S.Hrg.119-505

1[Senate Hearing 119-505]2[From the U.S. Government Publishing Office]34                                                        S. Hrg. 119-50556                 LESS HYPE, MORE HELP: AI THAT IMPROVES7                     SAFETY, PRODUCTIVITY, AND CARE89=======================================================================1011                                HEARING1213                               before the1415                SUBCOMMITTEE ON SCIENCE, MANUFACTURING,16                          AND COMPETITIVENESS1718                                 of the1920                         COMMITTEE ON COMMERCE,21                      SCIENCE, AND TRANSPORTATION22                          UNITED STATES SENATE2324                    ONE HUNDRED NINETEENTH CONGRESS2526                             SECOND SESSION27                               __________2829                             MARCH 3, 202630                               __________3132Printed for the use of the Committee on Commerce, Science, and Transportation3334                 [GRAPHIC NOT AVAILABLE IN TIFF FORMAT]3536                Available online: http://www.govinfo.gov3738                               ______3940                 U.S. GOVERNMENT PUBLISHING OFFICE414264-384 PDF                WASHINGTON : 20264344       SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION4546                    ONE HUNDRED NINETEENTH CONGRESS4748                             SECOND SESSION4950                       TED CRUZ, Texas, Chairman5152JOHN THUNE, South Dakota             MARIA CANTWELL, Washington,53ROGER WICKER, Mississippi                Ranking54DEB FISCHER, Nebraska                AMY KLOBUCHAR, Minnesota55JERRY MORAN, Kansas                  BRIAN SCHATZ, Hawaii56DAN SULLIVAN, Alaska                 EDWARD MARKEY, Massachusetts57MARSHA BLACKBURN, Tennessee          GARY PETERS, Michigan58TODD YOUNG, Indiana                  TAMMY BALDWIN, Wisconsin59TED BUDD, North Carolina             TAMMY DUCKWORTH, Illinois60ERIC SCHMITT, Missouri               JACKY ROSEN, Nevada61JOHN CURTIS, Utah                    BEN RAY LUJAN, New Mexico62BERNIE MORENO, Ohio                  JOHN HICKENLOOPER, Colorado63TIM SHEEHY, Montana                  JOHN FETTERMAN, Pennsylvania64SHELLEY MOORE CAPITO, West Virginia  ANDY KIM, New Jersey65CYNTHIA LUMMIS, Wyoming              LISA BLUNT ROCHESTER, Delaware6667                 Brad Grantz, Republican Staff Director68           Nicole Christus, Republican Deputy Staff Director69                   Lila Harper Helms, Staff Director70                 Melissa Porter, Deputy Staff Director7172                                 ------7374                SUBCOMMITTEE ON SCIENCE, MANUFACTURING,75                          AND COMPETITIVENESS7677TED BUDD, North Carolina, Chairman   TAMMY BALDWIN, Wisconsin, Ranking78MARSHA BLACKBURN, Tennessee          GARY PETERS, Michigan79TODD YOUNG, Indiana                  JACKY ROSEN, Nevada80ERIC SCHMITT, Missouri               JOHN HICKENLOOPER, Colorado81BERNIE MORENO, Ohio                  LISA BLUNT ROCHESTER, Delaware82CYNTHIA LUMMIS, Wyoming8384                            C O N T E N T S8586                              ----------8788                                                                   Page89Hearing held on March 3, 2026....................................     190Statement of Senator Budd........................................     191Statement of Senator Baldwin.....................................     292Statement of Senator Cruz........................................    2793Statement of Senator Blunt Rochester.............................    3094Statement of Senator Blackburn...................................    3295Statement of Senator Moreno......................................    3496Statement of Senator Hickenlooper................................    3697Statement of Senator Young.......................................    3898Statement of Senator Cantwell....................................    3999Statement of Senator Rosen.......................................    42100101                               Witnesses102103Demetri Giannikopoulos, Chief Innovation Officer, Rad AI.........     4104    Prepared statement...........................................     5105Brittany Ng, Vice President, Siemens Digital Industries Software.     9106    Prepared statement...........................................    10107Dr. Damion Shelton, Co-Founder and Chairman, Agility Robotics....    14108    Prepared statement...........................................    16109Mark Muro, Senior Fellow, Metropolitan Policy Program, Brookings110  Institution....................................................    17111    Prepared statement...........................................    19112113                                Appendix114115Derek Monson, Executive Director, Sutherland Institute, prepared116  statement......................................................    47117Letter dated March 3, 2026 to Hon. Ted Budd and Hon. Tammy118  Baldwin from Kristen Swearingen, Vice President, Government119  Affairs, Associated Builders and Contractors...................    49120Letter dated March 3, 2026 to Hon. Ted Budd and Hon. Tammy121  Baldwin from UVEye.............................................    50122Letter dated March 3, 2026 to Hon. Ted Budd and Hon. Tammy123  Baldwin from Graham Dufault, General Counsel, ACT | The App124  Association and Kedharnath Sankararaman, Policy Associate, ACT125  | The App Association..........................................    51126National Council on Disability, prepared statement...............    54127128                 LESS HYPE, MORE HELP: AI THAT IMPROVES129                     SAFETY, PRODUCTIVITY, AND CARE130131                              ----------132133                         TUESDAY, MARCH 3, 2026134135                               U.S. Senate,136       Subcommittee on Science, Manufacturing, and137                                   Competitiveness,138        Committee on Commerce, Science, and Transportation,139                                                    Washington, DC.140    The Subcommittee met, pursuant to notice, at 10:15 a.m., in141room SR-253, Russell Senate Office Building, Hon. Ted Budd,142Chairman of the Subcommittee, presiding.143    Present: Senators Budd [presiding], Cruz, Blackburn, Young,144Moreno, Baldwin, Cantwell, Hickenlooper, Blunt Rochester, and145Rosen.146147              OPENING STATEMENT OF HON. TED BUDD,148                U.S. SENATOR FROM NORTH CAROLINA149150    Senator Budd. This hearing will come to order. Good151morning, everyone. Thank you all for being here.152    I recognize myself for opening remarks, and thank you,153Ranking Member Baldwin and Chairman Cruz, Ranking Member154Cantwell, and our witnesses for working to put this important155hearing together.156    Artificial intelligence will undoubtedly usher in157significant improvements in quality of life for the American158people. It will make many workplaces safer and more productive,159helping to increase output, raise wages, and grow the economy.160    It will enhance manufacturing capabilities, especially161those of critical importance to our economic and national162security such as semiconductors and those in the defense163industrial base.164    It will make it easier to reshore manufacturing through165human-enhancing automation and digital twinning simulation.166    Smart systems and devices have the potential to167revolutionize health care, improving early detection of168diseases such as cancer, and helping people with disabilities169live better lives, not by replacing doctors but by augmenting170their diagnostic and treatment capabilities.171    AI-enabled research, aided by self-driving cloud labs,172could massively reduce barriers to the discovery of new drugs.173AI's potential in the health care industry presents unique174opportunities to save and improve lives.175    I have also noted that as I travel the state, some North176Carolinians share concerns about the growth of AI in automated177or autonomous technology.178    There is a natural hesitancy toward technology that may179disrupt incumbent industries or systems. It is normal to worry180about the impacts of advancements on your job, your children,181and your community.182    But if we have learned anything from our history, it is183that innovation is the lifeblood of the American economy. Our184nation's story has been shaped by technological advancements185that have time and time again expanded prosperity and improved186outcomes.187    I believe that AI, deployed in numerous ingenious ways, can188help people be better versions of themselves in their daily189lives.190    For those of us in this room, it is our job to listen to191those concerns and to work together in a bipartisan fashion to192address potential harms so that our Nation can reap the193tremendous benefits that AI has to offer.194    As I have said before, winning the AI race against China is195paramount for our national and economic security. The196administration is leaning in and providing important197leadership, ongoing work to first identify and then to aid in198the export of the American AI stack.199    From hardware to software, it will ensure that global AI200diffusion and standards are anchored in American values. The201Genesis Mission will build upon America's scientific dominance202by networking together world-class labs, computing power, and203datasets to turbo charge scientific discovery.204    The diffusion of AI into the economy will be a critical205dimension in this race. According to one report, AI is the most206rapidly adopted general purpose technology in history with207three in five U.S. adults surveyed having used AI less than 3208years after its release.209    However, other studies have found that U.S. businesses can210lag their Asian and even European peers in enterprise adoption211of AI tools.212    I am concerned that China, given its top-down command-and-213control structure, deep and sophisticated manufacturing214industry, and open-source heavy AI ecosystem is in a prime215position to diffuse AI quickly and broadly.216    Achieving a manufacturing renaissance in America is as217bipartisan and deeply held a goal as any in Congress. Given218demographic realities such as an aging, skilled workforce, and219stalled population growth, we will need to diffuse and scale220smart technologies and processes to make more critical goods221domestically.222    I am excited to discuss many of these technologies and223systems today. Our witnesses are at the front lines and cutting224edge of making our economy and our daily lives smarter.225    I look forward to hearing from them about what excites226them, what concerns them, and roadblocks we in Congress can227address.228    Ranking Member, do you have any comments?229    Senator Baldwin. Absolutely.230    Senator Budd. All right.231232               STATEMENT OF HON. TAMMY BALDWIN,233                  U.S. SENATOR FROM WISCONSIN234235    Senator Baldwin. Thank you so much, Mr. Chairman, and thank236you to the witnesses who are appearing before our subcommittee237today.238    Given the focus of today's hearing, I wanted to begin by239raising the troubling news from last week regarding the Trump240administration's actions to force artificial intelligence241companies to deploy systems without guardrails and safety242measures.243    Anthropic refused to have their technology used for mass244surveillance of Americans or for the deployment of autonomous245weapons without human oversight. The Trump administration then246labeled Anthropic as a security risk, and Secretary Hegseth has247taken the unprecedented step to blacklist the company.248    Our witnesses are here to testify about how artificial249intelligence can be used to improve worker safety and250efficiency, but we cannot ignore the stakes at play here.251Workers and consumers alike are rightfully worried about the252deployment of AI systems.253    A study done by Stanford last year found that 45 percent of254workers expressed concerns about the accuracy and reliability255of AI systems. Twenty-three percent feared job displacement and25616 percent worried about a lack of human oversight built into257the systems.258    And that does not even begin to touch on the concerns259regarding privacy, intellectual property, and energy use that260comes along with AI.261    But if designed with workers at the table during all262stages, from research and development to deployment to263negotiating guardrails with employers, artificial intelligence264has the potential to improve productivity, reduce safety risks,265and ease burnout in the workplace.266    The best innovations come from the factory floor. The267people actually doing the work know better than any senior268executive what the needs, inefficiency, and liabilities are day269to day.270    So I look forward to hearing our witnesses today and,271again, thank you for being here.272    Senator Budd. I thank the Ranking Member.273    I would like to introduce our witnesses for the day. Our274first witness is Demetri--and help me on this one, OK?275Pronounce your last name for me.276    Mr. Giannikopoulos. Giannikopoulos.277    Senator Budd. I even have a phonetic spelling here, but278thank you for that.279    Mr. Giannikopoulos is the Chief Innovation Officer at Rad280AI, a San Francisco-based company using AI to help radiologists281save time on reporting, allowing them to spend more time on282patient care. His experience applying new technologies to283health care spans over two decades.284    Our second witness is Brittany Ng. Ms. Ng is Vice President285at Siemens Digital Innovative Industry Software. Ms. Ng is an286expert in business matters and she has extensive expertise in287the deployment of AI across manufacturing context.288    Our third witness is Damion Shelton. Dr. Shelton is Co-289Founder and Chairman of the Board at Agility Robotics. He has290almost two decades of experience in academia, innovation, and291business, and Dr. Shelton founded two highly successful292robotics companies.293    Our final witness is Mark Muro. Mr. Muro is a Senior Fellow294at the Metropolitan Policy Program at the Brookings295Institution. At Brookings, his research focuses on how296technology development plays out differently across regions.297    All right. Mr. Giannikopoulos, you are recognized to298deliver your open statement if you are ready.299300             STATEMENT OF DEMETRI GIANNIKOPOULOS,301                CHIEF INNOVATION OFFICER, RAD AI302303    Mr. Giannikopoulos. Chairman, Ranking Member, and members304of the Subcommittee, thank you for the opportunity to testify.305    I come before you as a health care AI expert who has306deployed artificial intelligence nationwide, as a caregiver who307has supported my wife through cancer, as the spouse of a nurse308practitioner with more than a decade of frontline primary care309experience, and as a patient who has lived with multiple310sclerosis for more than two decades.311    These experiences have shaped who I am and are why I am312here today. They have fueled my more than 20-year career in313health care. They have given me a firsthand view of how care is314delivered, how decisions are made, and what happens when the315system works and when it does not.316    In each of these roles, I have seen the same reality.317Outcomes often hinge on whether a diagnosis is made quickly and318accurately, because in health care the most dangerous failure319is not a machine failure. It is a missed or delayed diagnosis.320    Research from Johns Hopkins Medicine estimates that nearly321800,000 Americans each year die or are permanently disabled322because of diagnostic error. Extensive research has shown that323errors often occur because clinicians are operating within an324increasingly complex and strained health care system, managing325rising volumes, time pressures, and expanding information.326    Imagine a patient arriving in a community emergency327department with sudden chest pain. A CT is performed. The cause328is something rare but deadly--an aortic dissection, a tear in329the body's main artery. Without rapid diagnosis and treatment,330a quarter of patients can die within 48 hours.331    Today, AI is helping at multiple points in that patient's332care. Within minutes, AI can analyze the images and help flag333the findings so it is reviewed immediately.334    When the radiologist opens the case, AI can help integrate335trusted clinical guidance directly into their workflow, drawing336on peer-reviewed evidence developed through organizations like337the Radiological Society of North America.338    It does not replace the physician. It strengthens339competence and trust in the diagnosis. Once the diagnosis is340made, the care team must act quickly. The patient may need to341be transferred for life-saving surgery.342    AI can help ensure the diagnosis is clearly communicated,343documented efficiently, and shared across care teams so344treatment can begin without delay.345    That is productivity. That is coordinated care. Research346published by the American Heart Association shows that347coordinated care pathways for aortic dissection reduce delays348and improve survival.349    We see the same reality in stroke care, where every minute350of delay risks permanent death of millions of brain cells,351controlling vital functions of the body. Faster diagnosis and352transfer can mean survival and less disability.353    But the story does not end there. That same scan may reveal354a quiet lung mass--not an emergency today, but something that355must be followed carefully. AI can help ensure follow-up356imaging is scheduled, care teams are alerted, and patients do357not fall through the cracks months or years later.358    Unfortunately, over half of patients never receive the359follow-up care their conditions require. For those Americans,360the risk is not theoretical. It becomes their reality.361    That is, this is not just about efficiency. It is about362trust. We are asking clinicians to manage rising imaging363volumes, expanding documentation requirements, and increasingly364complex patient needs, while the health care workforce is365shrinking and burnout remains high.366    AI, when implemented responsibly, is not replacing367clinicians. It is acting as a pressure release valve, helping368reduce cognitive burden and supporting clinicians in delivering369safe and timely care.370    I have deployed both FDA-cleared AI tools and other371clinical software that does not require FDA clearance. In both372cases, these systems undergo extensive clinical, privacy, and373governance reviews before deployment.374    Physicians remain responsible for patient care, and their375tools--these tools operate within existing health care laws and376professional accountability frameworks.377    Based on my experience deploying these systems nationwide,378the most important determinant of safety is not only how they379are evaluated before deployment, but how they are governed,380monitored, and supported once they are in clinical use.381    This is where thoughtful lifecycle governance and382consistent national standards are essential. This is especially383important for rural and underserved communities, where access384to subspecialty expertise may be limited.385    AI can help extend the reach of clinical expertise and386support clinicians caring for patients, regardless of geography387and access to local resources. AI is most powerful not when it388replaces human judgment, but when it strengthens it.389    AI will not replace physicians. It will help them do what390they train their entire lives to do, care for patients.391    Less hype, more help--that is not a future promise. It is392happening today, because behind every scan is a person, a393family, and a moment where getting the diagnosis right can394change everything.395    Thank you. I look forward to your questions.396    [The prepared statement of Mr. Giannikopoulos follows:]397398Prepared Statement of Demetri Giannikopoulos, Chief Innovation Officer,399                                 Rad AI400Introduction401    Chairman, Ranking Member, and Members of the Subcommittee:402403    Thank you for the opportunity to testify today on the role404artificial intelligence can play in strengthening healthcare for405patients and clinicians across the United States.406    My name is Demetri Giannikopoulos, and I serve as Chief Innovation407Officer at Rad AI, where I oversee the development and responsible408clinical integration of artificial intelligence technologies used by409radiologists and health systems nationwide. My work focuses on ensuring410these systems are safely integrated into clinical workflows, support411physicians in delivering accurate and timely diagnoses, and help412strengthen care coordination and patient outcomes.413    I previously served as Chief Transformation Officer at Aidoc, where414I led the clinical, operational, and governance protocol transformation415required to safely integrate multiple FDA-cleared artificial416intelligence medical devices into frontline physician workflows across417health systems nationwide. This work involved close collaboration with418clinicians, health system leadership, and regulatory stakeholders to419ensure these technologies functioned reliably in real-world clinical420environments and strengthened, rather than disrupted, the delivery of421patient care.422    In parallel, I have been actively involved in national governance423efforts focused on establishing standards for responsible artificial424intelligence in healthcare. I participate in the Applied Model Card425Workgroup of the Coalition for Health AI, a clinician-led coalition426advancing transparency, accountability, and safety in healthcare427artificial intelligence. I also serve on the Artificial Intelligence428Accreditation Advisory Committee for URAC, where we are developing429accreditation frameworks to help ensure artificial intelligence is430implemented safely, consistently, and responsibly across diverse431healthcare settings.432    I have also served on the Patient and Family Centered Care Clinical433Excellence Committee of the American College of Radiology, where I434contributed to the development of the Scanxiety Toolkit, which435addresses the anxiety patients experience while waiting for imaging436results. I bring this perspective not only through my professional437work, but also through personal experience. I supported my wife through438her cancer diagnosis and treatment, and I have also lived for more than439two decades as a patient with multiple sclerosis, undergoing regular440imaging and waiting for results that help determine the course of my441care. These experiences reinforced for me how deeply patients and442families depend on timely, accurate diagnosis, and clear communication443during some of the most vulnerable moments of their lives.444    Together, these experiences have given me a comprehensive445perspective on both the extraordinary potential and the profound446responsibility associated with integrating artificial intelligence into447the practice of medicine. Artificial intelligence in healthcare has448reached an inflection point. The central question is no longer whether449these technologies are promising, but how they perform in real clinical450environments and whether they meaningfully improve safety,451productivity, and patient care.452Imaging as a Critical Component of Modern Diagnosis453    Radiology plays a foundational role in modern medicine. Imaging is454often the step that confirms a diagnosis, rules out life-threatening455conditions, and determines the course of treatment. That matters456because diagnostic error remains a major source of preventable harm.457Research from Johns Hopkins Medicine, published in BMJ Quality &458Safety, estimates that each year nearly 800,000 Americans die or are459permanently disabled because of diagnostic error.\1\ Conditions such as460aortic dissection and stroke are among the exact types of time-461sensitive diagnoses that contribute to that harm.462---------------------------------------------------------------------------463    \1\ David E. Newman-Toker et al., Burden of Serious Harms from464Diagnostic Error in the USA, BMJ Quality & Safety (2023), https://465pubmed.ncbi.nlm.nih.gov/37460118/.466---------------------------------------------------------------------------467    Acute aortic dissection is a race against the clock. National468cardiology guidelines tell us that for dissections involving the469ascending aorta, the risk of death rises by about 1 to 2 percent every470hour without treatment.\2\ And we see the impact of that urgency very471clearly. Nearly one in four patients treated with medication alone died472within two days, compared to fewer than one in twenty patients who473received surgery.\3\474---------------------------------------------------------------------------475    \2\ 2022 ACC/AHA Guideline for the Diagnosis and Management of476Aortic Disease (James de Lemos et al., eds., American College of477Cardiology/American Heart Association 2022) (noting that acute aortic478dissection of the ascending aorta is ``highly lethal in symptomatic479patients left untreated, with an early mortality of 1 percent to 2480percent per hour after symptom onset''), https://www.ahajournals.org/481doi/10.1161/CIR.0000000000001106.482    \3\ Kevin M. Harris et al., Early Mortality in Type A Acute Aortic483Dissection: Insights From the International Registry of Acute Aortic484Dissection, JAMA Cardiology (2022), https://jamanetwork.com/journals/485jamacardiology/fullarticle/2795672.486---------------------------------------------------------------------------487    Stroke care follows the same pattern. Nearly 1.9 million neurons488are lost each minute that treatment is delayed.\4\ In both cases, rapid489imaging is what makes timely treatment possible. It is what allows490physicians to see the problem, make the diagnosis, and act before491irreversible harm occurs. In these moments, imaging is not just492diagnostic. It is decisive.493---------------------------------------------------------------------------494    \4\ Jeffrey L. Saver, Time Is Brain--Quantified, 37 Stroke 263495(2006), https://pubmed.ncbi.496nlm.nih.gov/16339467/.497---------------------------------------------------------------------------498    Artificial intelligence can help strengthen this process. It can499flag life-threatening findings for rapid physician review, support500accurate interpretation, and help ensure that critical diagnoses are501communicated quickly so patients receive the care they need without502delay. This responsibility exists across all healthcare settings,503including rural and community hospitals where subspecialty expertise504may not always be immediately available. Artificial intelligence can505help reinforce this diagnostic infrastructure, supporting clinicians in506delivering timely and accurate care regardless of geography.507Artificial Intelligence Across the Diagnostic Continuum508    Artificial intelligence supports care across the full diagnostic509continuum, from triage to interpretation to communication and follow-510up.511    The first stage is triage and prioritization. Artificial512intelligence can analyze imaging studies shortly after acquisition and513help identify findings that require urgent attention. This helps ensure514that time-sensitive conditions are reviewed promptly, improving patient515safety.516    The second stage is interpretation. Artificial intelligence can517help integrate trusted clinical knowledge and evidence directly into518physician workflows, supporting radiologists as they make diagnostic519decisions. These tools support clinicians, not replace them, and help520improve consistency and efficiency in care delivery. Artificial521intelligence can also improve clinical documentation and communication522by helping ensure diagnoses are clearly recorded, reports are completed523efficiently, and information is shared accurately across care teams.524This supports faster clinical action and reduces delays in treatment.525    The third stage is follow-up and longitudinal care. Imaging526frequently identifies findings that require monitoring over time, yet527failure to complete recommended follow-up remains a significant patient528safety challenge. In one large academic health system study, only about529half of patients completed recommended follow-up imaging, with530completion rates of just 51.9 percent at one institution and 52.0531percent at another.\5\ Similarly, studies of pulmonary nodules have532shown that more than half of patients may not receive recommended533follow-up imaging.\6\534---------------------------------------------------------------------------535    \5\ Thusitha Mabotuwana et al., Automated Tracking of Follow-Up536Imaging Recommendations, Am. J. Roentgenology (2019), https://537ajronline.org/doi/full/10.2214/AJR.18.20586.538    \6\ Denitza P. Blagev et al., Follow-up of Incidental Pulmonary539Nodules and the Radiology Report, 11 J. Am. Coll. Radiology 378 (2014),540https://pubmed.ncbi.nlm.nih.gov/24316231/.541---------------------------------------------------------------------------542    Artificial intelligence systems can help address this gap by543identifying patients who need follow-up, tracking whether that care544occurs, and supporting care teams in closing the loop. This helps545ensure that important findings do not fall through the cracks and that546patients receive the care they need. Together, these capabilities547improve safety, strengthen productivity, and support better care for548patients.549Workforce Challenges and Cognitive Burden550    Healthcare systems today face growing workforce challenges. Imaging551volume has increased substantially over time, while workforce growth552has not kept pace with demand.\7\ This imbalance reflects broader553trends in healthcare, driven by population aging and expanded reliance554on diagnostic imaging. At the same time, clinician exhaustion,555cognitive overload, and burnout remains widespread, affecting a556substantial portion of the clinical workforce.\8\557---------------------------------------------------------------------------558    \7\ Eric C. Christensen et al., Projecting the Future Radiologist559Workforce Supply in the United States Through 2055, 21 J. Am. Coll.560Radiology (2024); Eric C. Christensen et al., Projecting Imaging561Utilization in the United States Through 2055, 21 J. Am. Coll.562Radiology (2024)., https://www.jacr.org/article/S1546-1440(24)00898-6/563fulltext, https://www.jacr.org/article/S1556446-1440(24)00909-8/fulltext.565    \8\ Tait D. Shanafelt et al., Changes in Burnout and Satisfaction566With Work-Life Integration in Physicians and the General U.S. Working567Population Between 2011 and 2023, 100 Mayo Clinic Proc. 1142 (2025),568https://www.mayoclinicproceedings.org/article/S0025-6196(24)00668-2/569fulltext.570---------------------------------------------------------------------------571    Artificial intelligence can help address this challenge by572improving productivity, reducing administrative burden, and supporting573clinicians as they manage increasing workload demands. A recent task-574based analysis by Dr. Curtis Langlotz of Stanford University, the575immediate past President of the Radiological Society of North America,576projects that artificial intelligence could significantly reduce the577time radiologists spend on certain tasks, particularly report drafting578and workflow coordination. However, the study also concludes that579because imaging demand continues to grow and the radiology workforce580has remained relatively stable, artificial intelligence is unlikely to581eliminate the need for radiologists.\9\ Instead, these technologies are582expected to help clinicians work more efficiently and focus more of583their time on direct patient care.584---------------------------------------------------------------------------585    \9\ Langlotz CP. The Effect of AI on the Radiologist Workforce: A586Task-Based Analysis. medRxiv [preprint]. Posted December 22, 2025.587doi:10.64898/2025.12.20.25342714, https://www.medr588xiv.org/content/10.64898/2025.12.20.25342714v1.589---------------------------------------------------------------------------590    In addition to improving efficiency, artificial intelligence can591help reduce cognitive burden by assisting with information management,592documentation, and follow-up tracking. This support helps clinicians593focus their attention on patient care and clinical decision making,594which is especially important as workforce shortages and burnout595continue to affect healthcare nationwide. Artificial intelligence596supports clinicians; it does not replace them.597Improving Access to Care598    Healthcare access challenges are particularly acute in rural and599underserved communities. More than 46 million Americans live in rural600areas, where access to specialists may be limited.\10\ Many rural601hospitals operate with constrained clinical staffing and may not have602immediate access to subspecialty expertise, including radiologists with603advanced training in specific conditions.604---------------------------------------------------------------------------605    \10\ U.S. Census Bureau, New Census Data Show Differences Between606Urban and Rural Populations (2016), https://www.census.gov/newsroom/607archives/2016-pr/cb16-210.html.608---------------------------------------------------------------------------609    Artificial intelligence can help address these gaps by supporting610clinicians in delivering timely and effective care. For example,611artificial intelligence can help identify urgent findings, assist with612interpretation, and ensure that critical results are communicated613promptly. These capabilities help clinicians work more efficiently and614reduce the risk that important findings are delayed or missed. Ensuring615that patients receive consistent, high-quality diagnostic care616regardless of geography is one of the most important opportunities for617artificial intelligence to improve healthcare nationally.618Governance, Oversight, and Accountability619    The safe and effective use of artificial intelligence requires620strong governance, clear accountability, and ongoing oversight.621    Based on my experience deploying these systems nationwide, the most622important determinant of safety is not only how artificial intelligence623is evaluated before deployment, but how it is governed, monitored, and624supported once it is in clinical use. Importantly, artificial625intelligence systems operate within existing legal and professional626accountability frameworks. Physicians remain responsible for clinical627decision making, and health systems are responsible for safe628implementation. Existing laws, including patient privacy protections,629medical malpractice standards, and civil rights protections, continue630to apply. Artificial intelligence does not replace these safeguards; it631operates within them.632    Governance initiatives such as those led by the Coalition for633Health AI and accreditation programs such as those being released by634URAC play an important role in supporting safe and responsible635implementation. Maintaining clinician leadership in patient care636decisions remains essential to ensuring patient safety and preserving637trust.638Regulatory Considerations and Responsible Innovation639    The United States has established strong regulatory frameworks to640evaluate medical devices and protect patient safety. Through my641experience implementing FDA-cleared artificial intelligence systems, I642have seen how regulatory clarity and predictability support responsible643innovation and safe deployment. Peer-reviewed research has consistently644shown that artificial intelligence performs best when used to support645clinicians, not replace them.\11\646---------------------------------------------------------------------------647    \11\ Eric J. Topol, High-Performance Medicine: The Convergence of648Human and Artificial Intelligence, Nature Medicine (2019), https://649www.nature.com/articles/s41591-018-0300-7.650---------------------------------------------------------------------------651    Artificial intelligence introduces new considerations because these652technologies operate within complex clinical environments and continue653to evolve over time. Ongoing collaboration among regulators, healthcare654providers, and developers can help ensure regulatory frameworks655continue to support patient safety while enabling beneficial656innovation. Thoughtful evolution of regulatory approaches can657strengthen both safety and public trust.658Conclusion659    Artificial intelligence is already helping strengthen healthcare660delivery when it is implemented responsibly, governed carefully, and661designed to support clinicians and patients. These technologies can662help clinicians diagnose disease earlier, improve productivity,663strengthen care coordination, and improve patient outcomes.664    Artificial intelligence supports clinicians; it does not replace665them. It helps clinicians do what they trained their entire lives to666do: care for patients. Behind every scan is a patient, a family, and a667moment where getting the diagnosis right can change the course of a668life. Thank you for the opportunity to testify. I look forward to your669questions.670671    Senator Budd. Thank you for your remarks.672    Ms. Ng, you are recognized for 5 minutes.673674           STATEMENT OF BRITTANY NG, VICE PRESIDENT,675              SIEMENS DIGITAL INDUSTRIES SOFTWARE676677    Ms. Ng. Good morning, Chairman, Ranking Member, and members678of the Subcommittee. Thank you for the opportunity to testify679today.680    The United States does not need AI as an abstract681capability. Our global competitiveness requires AI deployed on682factory floors, in shipyards, and across production systems683where it generates measurable productivity gains.684    I am proud to take on this challenge. I lead Siemens'685maritime business in the United States where we work hand in686hand with shipyards and manufacturers.687    At Siemens, we are a global leader in industrial AI. Last688year alone we invested $15 billion in the United States to689further our leadership in this transformational technology.690    We are continuing to bring industrial AI to customers and691seeing firsthand how AI is most powerful when it connects data692to operational decisions in the real world.693    Through our maritime work, shipbuilders are using our694physics-based, AI-enabled digital shipbuilding platform to695connect, design, simulation, and production planning so that696teams can identify bottlenecks before they occur.697    They can reduce rework, improve first-time quality, and698compress production timelines, and by training AI in virtual699environments, shipyards can improve planning and performance700even in complex production conditions.701    Industrial AI, alongside digital twins and software-defined702automation, is transforming manufacturing across industries. We703are seeing machine downtime reduced by 50 percent, energy704consumption cut by 20 percent, and quality control with 99.99705percent accuracy.706    Industrial AI is helping manufacturers to achieve up to 40707percent in productivity improvements and it is transforming708traditional factories into flexible digital enterprises.709    In modern shipbuilding, innovation looks like being able to710create a full digital twin of a vessel before steel is even711cut. This means that our customers can simulate how the ship712will be built, how systems will integrate, and how production713will flow through the yard, all in a virtual environment.714    On the deck plate, industrial AI can help sequence work715packages, flag quality issues earlier, predict equipment716downtime, and optimize the flow of material. The crews717deploying these technologies are spending less time waiting and718doing rework and more time building.719    When these tools are broadly adopted, they become economic720multipliers. However, technology does not just transform721industry unless the people understand it, trust it, and see how722it improves their lives.723    At Siemens, we have learned that digital transformation724works best when the worker is at the center. Manufacturing725today faces a shortage of 400,000 open jobs nationwide.726    The issue is not AI coming for industrial jobs. It is a727shortage of skilled workers amid increasing production728complexity. Industrial AI does not just solve the workforce729shortage; it changes the entire equation.730    These technologies reduce repetitive and hazardous tasks731while elevating the skill sets of American workers. As Congress732considers AI and manufacturing policy, I respectfully offer733three considerations.734    First, distinguish industrial AI from consumer AI.735Industrial AI operates in structured, safety-critical,736business-to-business environments with validated data and737rigorous testing.738    Second, focus on adoption. Federal policies and investments739should integrate digital modernization from the outset and740accelerate deployment through public-private collaboration.741    Programs like Manufacturing USA and the DOE's Genesis742Mission provide that bridge, moving from innovation into real-743world production.744    Last, the government should send a clear and consistent745signal that digital capability is core industrial746infrastructure. America's competitive edge will be determined747by who most effectively deploys AI in the physical world.748    Our nation is indisputably the world's greatest innovation749leader. Our challenge is scaling advanced technologies through750the industrial base.751    Thank you, and I look forward to your questions.752    [The prepared statement of Ms. Ng follows:]753754          Prepared Statement of Brittany Ng, Vice President,755                  Siemens Digital Industries Software756Introduction757    Chairman Cruz, Ranking Member Cantwell, Chairman Budd, Ranking758Member Baldwin, and distinguished Members of the Subcommittee, thank759you for the opportunity to testify today.760    I lead Siemens' maritime business in the United States, where we761work directly with shipyards, defense contractors, and manufacturers762modernizing some of the most complex production systems in the world.763    Siemens is a global technology company with deep roots in the764United States, including the global headquarters of our software765business in Plano, TX. Across our businesses, we help design, simulate,766build, and operate the infrastructure and industrial systems that767underpin the American economy. Last year alone, Siemens invested768approximately $15 billion in the U.S. to deepen our leadership in769Industrial AI, simulation, and digital engineering.770    Through that work, we have learned a simple lesson: AI is most771powerful not as a standalone capability, but when it connects data to772operational decisions in the physical world.773Siemens in the USA774    Siemens has been part of the American industrial landscape for over775160 years. Our presence in the United States spans manufacturing776floors, shipyards, software labs, and engineering classrooms. In many777ways, our technologies are embedded in nearly every chapter of the778American industrial story--from designing semiconductors and building779aircraft, to modernizing shipyards, powering data centers,780strengthening the electric grid, and advancing next-generation781manufacturing. Approximately 50,000 Siemens employees across the782country are designing, building, and deploying the technologies that783power American production.784    In Fort Worth, Texas, one of our newest advanced manufacturing785facilities illustrates what that commitment looks like in practice.786Before the first piece of equipment was installed, the factory existed787as a full digital twin--a physics-based virtual replica used to788simulate production flows, validate line configurations, and stress-789test material movement. By resolving bottlenecks in the virtual790environment first, the facility ramped up faster with greater precision791and flexibility. It is a clear example of how digital-first planning792accelerates real-world industrial capacity.793    That same approach extends beyond a single facility. Across the794Midwest--from Missouri to Ohio--thousands of Siemens engineers and795software developers build the digital engineering platforms that make796these deployments possible. The simulation and industrial software797developed by these teams form the backbone of AI-enabled production798systems used across the United States and globally. Industrial AI799adoption is not just happening on factory floors--it is being coded,800tested, and refined in American software hubs.801    Building on this momentum, Siemens is also advancing an AI build-802out in Washington state to scale AI and digital infrastructure where803industrial transformation is accelerating. The goal of our Data and AI804team, which will be globally headquartered in the United States, is to805unify data, software, and engineering platforms to deliver AI-native806solutions across the industrial lifecycle, from design through807operations. Coupled with Siemens' broader U.S. investment strategy--808totaling investments of over $100 billion in 20 years--this build-out809reinforces our belief that scaling Industrial AI is essential to810competitiveness.811    Our commitment to the United States is not limited to facilities812and software--it is equally a commitment to people. Siemens has pledged813to help train 200,000 electricians and manufacturing experts in the814United States by 2030 to close the skills gap that increasingly defines815industrial competitiveness. This initiative reflects a simple reality:816as production systems become more software-defined and AI-enabled, the817competitive advantage of the United States will depend not just on818innovation, but on whether America's workforce is prepared to lead that819innovation on the factory floor, in the software suite, and at the820intersection of design and execution.821    That commitment comes to life through deep partnerships across the822country with universities, apprenticeships, vocational schools, and823other workforce institutions preparing the next generation of824industrial leaders. At Purdue University, Siemens collaborates to825advance digital twin research and embed simulation tools directly into826engineering education--ensuring graduates are fluent in the digital827environments they will encounter in modern manufacturing. In Colorado,828Siemens drives microcredentialing initiatives that provide modular,829industry-recognized digital skills training. And in Kansas, Siemens830technologies power smart factory initiatives that demonstrate how AI-831enabled automation can scale across manufacturers of all sizes.832    The list can go on. But, all of our efforts reflect a consistent833theme: Siemens' role in the United States is operational, not abstract.834We design here. We manufacture here. We train the workforce here. And835we deploy Industrial AI here--in ways that strengthen American836productivity and industrial readiness.837Industrial AI in the Physical World838    In modern shipbuilding, we can now create a full digital twin of a839vessel before steel is ever cut. A digital twin is a high-fidelity,840physics-based virtual demonstration of a physical asset or production841process. In shipbuilding, that means we can simulate how the ship will842be built, how systems will integrate, how materials will flow through843the yard, how production constraints may emerge--all before physical844construction begins.845    For example, through our work with HD Hyundai, shipbuilders are846deploying an AI-enabled digital shipbuilding platform that connects847design, simulation, and production planning. Teams can test decisions848virtually, identify bottlenecks before they occur, reduce costly849rework, and compress production timelines.850    By training AI in virtual environments using digital twins and851synthetic data, shipyards can improve real-world planning and852performance in highly complex production environments. This is what AI853looks like in the physical world.854    On the deck plate, it means sequencing work packages more855efficiently, flagging quality issues earlier, predicting equipment856downtime, and optimizing material movement across congested yards.857Workers also get to spend less time waiting and correcting mistakes,858and more time building.859    AI also offers tangible improvements for worker safety in860manufacturing, as systems can identify hazardous conditions before they861escalate--whether that is equipment operating outside safe thresholds,862congestion in high-traffic work zones, or potential conflicts between863tasks. Digital twins allow crews to simulate complex lifts, confined-864space operations, and workflow changes virtually before executing them865in the physical environment. Workers spend less time navigating866uncertainty and fewer hours exposed to avoidable risks. Instead of867reacting to problems after they occur, teams can anticipate them--868improving both safety outcomes and operational confidence.869    What is true in maritime manufacturing is true across sectors.870Industrial AI and software-defined automation are delivering measurable871gains across aerospace, automotive, semiconductors, energy systems,872food & beverage, advanced manufacturing, and more.873    We're seeing machine downtime reduced by 50 percent, energy874consumption cut by 20 percent, and quality control with 99.99 percent875accuracy across industrial operations. Furthermore, Industrial AI876capabilities are helping manufacturers achieve up to 40 percent877productivity improvements through predictive maintenance, quality878control optimization, and automated decision-making--transforming879traditional factories into flexible, efficient digital enterprises.880    For example, through our partnership with JetZero--an American881aerospace innovator--we are helping bring to market a next-generation882blended-wing aircraft and a digitally native ``Factory of the Future''883in North Carolina. Using Siemens' AI-enabled digital twin and884engineering platforms, JetZero can explore thousands of design885configurations, model aerodynamic performance, and simulate its886production system before physical assembly begins. AI tools support887both engineering and shopfloor operations--from optimizing aircraft888performance to assisting technicians in diagnosing equipment issues.889The result is a more efficient aircraft and a smarter manufacturing890system built here in the United States.891    Across all these sectors--the result of this adoption is faster892time to market, reduced waste, lower cost structures, and more893resilient supply chains.894    AI is not an abstract capability. AI is deployed on factory floors,895in shipyards, and across production systems where it generates896measurable productivity gains.897Accelerating Adoption898    America's competitive edge will not be determined by who builds the899most AI models. It will be determined by who most effectively deploys900AI in the physical world.901    That requires adoption.902    The United States does not have an innovation problem--we have an903adoption challenge. We lead in developing breakthrough technologies.904The strategic challenge is scaling them across manufacturers of all905sizes and across every sector of our industrial base.906    This adoption challenge is particularly acute for small and medium-907sized manufacturers, which often lack the ability to experiment with908emerging technologies. Through our work with Haddy--an American909advanced additive manufacturing company--we are demonstrating how910automation-first production can lower those barriers. Haddy operates911localized micro-factories that combine large-scale robotic 3D printing912with digitally coordinated workflows. AI is embedded directly into913factory control and execution systems, enabling faster iteration and914scalable distributed production. This model shows how smaller915manufacturers can compete in advanced industrial supply chains without916the footprint of traditional larger-scale facilities.917    This is also true in national security contexts. In shipbuilding,918small improvements in production or quality can compound across919thousands of tasks and millions of labor hours. Adoption translates920directly into more predictable schedules, lower costs, and stronger921industrial capacity.922    At Siemens, we create technology to transform the everyday, for923everyone. But technology alone does not transform an industry.924Deployment at scale does. The companies and nations that integrate AI925into core production systems--not as isolated pilots, but as operating926infrastructure--will lead in productivity and competitiveness.927AI & The Manufacturing Workforce928    Technology cannot transform industry unless people trust it,929understand it, and see how it improves their work. At Siemens, we view930AI as a people-centric transformation designed to help the workforce931make better decisions at a faster pace, rather than replace human932expertise.933    Manufacturing today faces more than 400,000 open jobs nationwide.934The issue is a shortage of workers combined with increasing production935complexity. Across the country, an aging workforce is retiring faster936than it can be replaced, and many employers struggle to attract and937retain the next generation of digitally native talent. Young workers938expect modern tools, intuitive systems, and career mobility--not paper-939driven processes and repetitive manual tasks.940    Industrial AI doesn't just solve the workforce shortage. It changes941the equation.942    Digital twins are a force-multiplier and allow workers to train in943simulated environments before stepping onto the production floor. AI-944powered copilots help technicians diagnose issues faster. Software-945defined systems make production more intuitive and adaptable. Across946environments, from the deck plate to the factory floor, this means947fewer repetitive and hazardous tasks, faster troubleshooting, and more948predictable execution.949    These technologies elevate the skill sets of American workers for950the higher paying jobs of tomorrow, while also increasing productivity951per worker. At Siemens, we've learned digital transformation works best952when the worker is at the center--when welders, machinists, planners,953and engineers are part of the process from day one.954    Industrial AI empowers workers to operate more advanced and955competitive ecosystems--thereby strengthening both productivity and956opportunity.957Policy Considerations958    As Congress considers AI and manufacturing policy, we offer three959considerations.960961    1. Distinguish Industrial AI from Consumer AI962963        Industrial AI operates in structured, safety-critical,964        business-to-business environments using validated operational965        data generated by machines, sensors, and engineering systems.966        These applications are deployed in regulated industries such as967        shipbuilding, aerospace, energy, and semiconductor968        manufacturing, where safety and quality are non-negotiable.969970        Industrial AI applications often occur upstream in design and971        simulation environments before physical products are built.972        Digital twins allow manufacturers to test and optimize973        processes virtually, reducing risk before deployment in the974        real world. In these settings, AI systems operate within975        defined parameters and controlled data environments--very976        different from open-ended consumer applications.977978        To foster innovation, policy frameworks should reflect these979        distinctions. Approaches designed for consumer-facing AI should980        not unintentionally slow adoption in manufacturing environments981        that already operate under rigorous technical and safety982        standards.983984    2. Focus on Adoption985986        The United States faces adoption and scaling challenges,987        especially as emerging technologies are being deployed988        worldwide at a rapid pace. To compete in the global economy,989        breakthrough technologies developed in labs and leading firms990        must be adopted broadly across the industrial base--including991        small and medium-sized manufacturers--to drive economy-wide992        productivity gains.993994        In tandem, Federal policies and investments--from shipbuilding995        to semiconductors--must integrate digital tools from the outset996        to ensure advanced manufacturing initiatives include the997        technology necessary for AI-enabled production.998999        Public-private collaboration plays a critical role in1000        accelerating this deployment. Programs that connect advanced1001        computing, simulation, and industrial software with real-world1002        production environments can shorten commercialization timelines1003        and reduce barriers to adoption. Through the Department of1004        Energy's Genesis Mission, Siemens is excited to help translate1005        Industrial AI, digital twins, and advanced computing into1006        scalable, operational industrial applications.10071008        Innovation creates meaningful potential. Adoption converts that1009        potential into measurable productivity, resilience, and1010        competitiveness.10111012    3. Send a Clear Signal for Digitally Native Production10131014        Both industry and government should send clear and consistent1015        signals that digital tools are core infrastructure for1016        competitiveness.10171018        For example, Manufacturing USA Institutes provide platforms1019        where manufacturers, suppliers, technology providers, and1020        workforce institutions collaborate on applied innovation.1021        Managing these partnerships as a true network to focus on1022        specific manufacturing challenges can help pursue the goal of1023        bridging the gap between basic research and commercialization.10241025        Tailored regulatory sandboxes complement this effort by1026        proposing structured environments where AI-enabled production1027        systems can be deployed and tested within defined guardrails.1028        This approach allows responsible experimentation to move1029        forward with clarity and confidence, putting technology and1030        innovation before regulation.10311032        Additionally, challenge-based partnerships such as the1033        Department of Energy's Genesis Mission help translate1034        breakthrough capability development into operational systems1035        that can create real-world impact. By connecting advanced1036        research infrastructure to real use cases, these programs1037        bridge the gap between innovation and deployment.10381039        Modernization scales fastest when expectations are clearly set1040        in Federal programs, such as procurement standards, financing1041        mechanisms, workforce programs, and public-private1042        partnerships. A coordinated signal from policymakers that1043        digitally native industrial strategy is a national priority1044        will accelerate technology adoption, strengthen domestic supply1045        chains, and reinforce long-term industrial readiness.1046Closing1047    In closing, consider the story about Albert Einstein riding on a1048train. When the conductor came by to collect tickets, Einstein began1049searching his pockets but couldn't find his ticket. The conductor1050recognized him and said, ``Professor Einstein, it's okay--I know who1051you are.''1052    Einstein kept searching. He looked in his briefcase, under his1053seat, even on the floor. The conductor said again, ``Dr. Einstein, I1054trust you.''1055    Einstein replied, ``Young man, I know who I am. What I don't know1056is where I'm going.''1057    America knows who it is: the world's great innovation leader. But1058if we don't focus on where we're going--on how we deploy these1059groundbreaking innovations across our industrial base--we risk losing1060direction.1061    If we lead in adoption, we lead in productivity.1062    If we lead in productivity, we lead in competitiveness.1063    Thank you, and I look forward to your questions.10641065    Senator Budd. Dr. Shelton, you are recognized for 51066minutes.10671068          STATEMENT OF DR. DAMION SHELTON, CO-FOUNDER1069                 AND CHAIRMAN, AGILITY ROBOTICS10701071    Mr. Shelton. Thank you, Chairman Budd and Ranking Member1072Baldwin, for inviting me to speak today. Also, thanks to the1073rest of the Committee.1074    I would like to open with a quote from ``Peter Pan''--all1075of this has happened before and it will all happen again.1076    The United States first started collecting census data1077around agricultural employment in 1820, almost exactly 2001078years before Agility Robotics launched its humanoid robot.1079    In 1820, the vast majority of Americans worked on farms1080using simple tools and assisted by animals. Over the next 2001081years, agricultural employment shrank to about 2 percent of the1082workforce.1083    This is not, however, a story of decline. Absolute1084employment, the total number of Americans working in1085agriculture, has actually increased by about 30 percent since1086the 1820s.1087    Why? In a single word, technology. In the 50 years after1088the patenting of the McCormick reaper in 1834, nearly 12,0001089additional farm implement patents were filed in the U.S. Far1090from destroying jobs, automation in agriculture expanded both1091direct agricultural employment and unlocked truly staggering1092growth in the United States, a 30X increase in population and a109340X increase in per capita GDP.1094    The modern AI and robotics boom promises to bring these1095sorts of transformative changes to the rest of our economy.1096Agility Robotics started in 2015 as a spinoff from Oregon State1097University.1098    From our DARPA-funded academic roots we have grown to more1099than 300 employees, primarily in Oregon, Pennsylvania, and1100California. Our Oregon facility co-locates both our R&D teams1101and our factory, RoboFab, where we assemble 100 percent of our1102robots in the U.S.1103    Our humanoid robot, Digit, was launched in 2020 as the1104first full-size humanoid robot available for purchase. Our1105customers include warehouse and logistics work, Amazon, GXO,1106and Mercado Libre, and manufacturing and automotive suppliers1107Schaeffler and Toyota with the common theme of providing a1108solution for repetitive material handling alongside human co-1109workers.1110    I would like to address the impact of automation on jobs1111and human workers. Many manual labor jobs in the U.S. are1112facing pressure from an aging workforce, high turnover, and a1113high injury rate, which is where Agility has focused its1114deployments.1115    The Bureau of Labor Statistics projects average annual1116vacancies of more than one million positions in warehouse and1117logistics work. Each robot deployed here not only does not take1118a job from a human, it enables the business to grow and expand1119overall hiring elsewhere in the organization.1120    To cite one example from an Agility partner, in 2012,1121Amazon employed about 88,000 people and acquired the robotics1122startup Kiva. Over the next 13 years, they deployed more than a1123million robots while also hiring an additional 1.4 million1124people.1125    Prior to the modern AI boom, it took a skilled engineer1126many months to develop a new application for a robot. Modern AI1127can dramatically shorten the time to deployment for small and1128medium-sized businesses that desperately need to solve labor1129challenges but lack the technical and financial resources to1130run their own IT department.1131    However, as we have seen from self-driving vehicles, safety1132often lags behind technical ability. It is imperative that1133robots not endanger their human colleagues, members of the1134public, or our homes and workplaces.1135    Focusing first on controlled environments has allowed1136Agility Robotics and our partners like Boston Dynamics to focus1137on developing responsible industry-led safety standards for1138robots in the workplace.1139    Gaining experience in these environments first with a focus1140on continuous safety improvement and industrial best practices1141is the fastest and most responsible path toward a successful1142long-term buildout of general purpose automation.1143    As an industry, we owe the general public a solid safety1144argument backed by data. I will conclude by briefly addressing1145the challenges posed by China's extremely rapid progress in1146humanoid robots.1147    To be as blunt as possible, they are doing a good job.1148Their technology is well-designed, highly capable, and backed1149by a formidable supply chain. Two recent papers by security1150researchers have identified a critical security vulnerability1151in a Chinese humanoid robot allowing for remote takeover as1152well as a phone home data logging mechanism that sends data to1153a remote server.1154    As of this week--and I did check this before the hearing--1155that robot is available for sale in the U.S. for less than1156$20,000. Combined with open source and open weight AI models,1157China is creating a compelling value proposition for early1158adopters of general purpose humanoid robots.1159    This is a siren's call that we would be well-served to take1160seriously or risk ceding the future of automation and, by1161extension, economic growth to others.1162    Thank you again for the opportunity to speak today and I1163thank for this committee's focus and leadership on these1164important issues, including the regulatory sandbox approach.1165    Agility Robotics is ready to help the U.S. solve its labor1166challenges, expand the economy, and ensure American1167competitiveness and economic security as we move into the next1168industrial revolution.1169    [The prepared statement of Mr. Shelton follows:]11701171 Prepared Statement of Dr. Damion Shelton, Co-Founder and Chairman of1172                      the Board, Agility Robotics1173    Thank you Chairman Budd and Ranking Member Baldwin for inviting me1174to speak to you today. And thank you as well to Full Committee Chairman1175Cruz and Ranking Member Cantwell, and all the other Members of this1176subcommittee.1177    I'd like to open with a quote from Peter Pan: ``All this happened1178before, and it will happen again.''1179    The United States first started collecting census data on1180agricultural employment in 1820, almost exactly 200 years before1181Agility Robotics launched its humanoid robot. In 1820, the vast1182majority of Americans worked on farms, using simple tools and assisted1183by animals. Over the next 200 years, agriculture employment shrank to1184about 2 percent of the overall workforce. This is not, however, a story1185of decline. Absolute employment--the number of total Americans working1186in agriculture--has actually increased by almost 30 percent since the11871820s. Why? In a single word: technology. In the 50 years after the1188patenting of the McCormick Reaper in 1834, nearly 12,000 additional1189farm implement patents were filed. Far from destroying jobs, automation1190in agriculture expanded both direct agricultural employment, and1191unlocked truly staggering growth in the United States: a 30x increase1192in population and a 40x increase in per-capita GDP. The modern AI and1193robotics boom promises to bring these sorts of transformative changes1194to the rest of our economy.1195    Agility Robotics was started in 2015, as a spin-off from Oregon1196State University. From our DARPA-funded academic roots, we've grown to1197more than 300 employees primarily in Oregon, Pennsylvania, and1198California. Our Oregon facility co-locates both our R&D teams and our1199factory--RoboFab, and 100 percent of our robots are assembled in the1200USA. Our humanoid robot, Digit, was launched in 2020 as the first full-1201size humanoid robot available for purchase. Our customers include1202warehouse and logistics work (Amazon, GXO, and Mercado Libre) and1203manufacturing and automotive (Schaeffler and Toyota), with the common1204theme of providing a solution for repetitive material handling tasks1205alongside human coworkers.1206    I'd like to address the impact of automation on jobs and human1207workers. Many manual labor jobs in the U.S. are facing pressure from an1208aging workforce, high turnover, and a high injury rate, which is where1209Agility has focused its deployments. The Bureau of Labor Statistics1210projects average annual vacancies of more than one million positions in1211warehouse manual labor roles.1212    Each robot deployed here not only doesn't take a job from a human,1213it enables the business to grow and expand overall hiring elsewhere in1214the organization. To cite one example from an Agility partner: in 2012,1215Amazon employed about 88,000 people and acquired the robotics startup1216Kiva. Over the next 13 years, they deployed more than a million robots,1217while hiring an additional 1.4 million people.1218    Prior to the modern AI boom, it took a skilled engineer many months1219to develop a new application for the robot. Modern AI can dramatically1220shorten the time to deployment for the small and medium-sized1221businesses that desperately need to solve labor challenges but lack the1222technical and financial resources to run their own IT department.1223However, as we've seen with self-driving vehicles, safety often lags1224behind technical ability. It's imperative that robots not endanger1225their human colleagues, members of the public, or our homes and1226workplaces.1227    Focusing first on controlled environments has allowed Agility1228Robotics and our partners like Boston Dynamics to focus on developing1229responsible, industry-led safety standards for robots in the workplace.1230Gaining experience in these environments first, with a focus on1231continuous safety improvement and industrial best practices, is the1232fastest and most responsible path towards a successful long-term1233buildout of general-purpose automation. As an industry, we owe the1234general public a solid safety argument backed by data.1235    I'll conclude by briefly addressing the challenges posed by China's1236extremely rapid progress in humanoid robotics. To be as blunt as1237possible: they're doing a good job. Their technology is well designed,1238highly capable, and backed by a formidable supply chain. Two recent1239papers by security researchers have identified a critical security1240vulnerability in a Chinese humanoid robot allowing for remote takeover,1241as well as a ``phone home'' data logging mechanism that sends1242continuous operational data to a remote server. As of this week, that1243robot is available for sale in the U.S. for less than $20,000. Combined1244with open source and open weight AI models, China is creating a1245compelling value proposition for early adopters of general-purpose1246humanoid robots. This is a siren's call that we would be well served to1247take seriously, or risk ceding the future of automation, and by1248extension continued economic growth, to others.1249    Thank you again for the opportunity to speak today, and I thank you1250for this committee's focus and leadership on these important issues,1251including the regulatory sandbox approach. Agility Robotics is ready to1252help the U.S. solve its labor challenges, expand the economy, and1253ensure American competitiveness and economic security as we move into1254the next industrial revolution.12551256    Senator Budd. Thank you, Dr. Shelton.1257    Mr. Muro, you are recognized for 5 minutes.12581259             STATEMENT OF MARK MURO, SENIOR FELLOW,12601261                  METROPOLITAN POLICY PROGRAM,12621263                     BROOKINGS INSTITUTION12641265    Mr. Muro. Mr. Budd, Ms. Baldwin, thanks so much. And to the1266distinguished members of the Committee, I really want to thank1267you for the opportunity to comment here.1268    I will just say my remarks here are my personal views, you1269know, separate from Brookings Institution, and I think I want1270to start by reaffirming the basic premise here that I think is1271very compelling, that America's innovators are day by day1272introducing incredible new tools and solutions that are1273allowing more and more people, firms, and entrepreneurs to1274expand and reach--reach and achievement of human expertise.1275    It is a compelling moment, and that is creating in the1276country significant excitement, and yet a degree of pessimism1277also complicates this moment. We should be frank about that and1278the Committee has been frank about that.1279    Some worry that the technology will not be smoothly1280adopted, that adoption will be challenging. Others fear what1281has been deemed the greatest automation technology in history,1282and still others worry about safety questions and which skills1283will serve them in the future to enable this work.1284    So I want to say just a few words about areas where Federal1285support can help maintain the sector's momentum, reinforce its1286value, and promote optimism about its possibilities.1287    In this direction, my fellow panelists have done an1288excellent job of inspiring confidence by detailing some of the1289possibilities. I want to touch on just a few ways my written1290testimony suggests we can sustain AI success.1291    First, broader AI innovation and adoption like we are1292seeing requires maintaining a dominant research base--abundant1293research flows generate talent--but also the ideas,1294intellectual property, innovation, and startups that we are1295hearing about.1296    I will provide more detail in the written remarks, but the1297Nation remains--does retain clear global leadership. With that1298said, it has fallen short of the doubling of AI research that1299have been suggested in 2019 by the National Security Commission1300on Artificial Intelligence, an important bipartisan watchword.1301    Beyond that, to address the scale issue, the Nation should1302prioritize a step change in total AI R&D outlays, at the same1303time seeking to rejuvenate the AI Research Institute's program1304for accelerating research on new topics, establish a new1305testbed programs involving Federal research units, private1306sector, and even data centers that can be brought into this,1307and then direct grant making toward investments in what some of1308us call pro-worker AI that supports such values as education1309systems, health, human learning, and human decisionmaking.1310    Accelerating AI adoption also requires promoting the growth1311of emerging AI clusters in geographic regions. Nvidia, OpenAI,1312Microsoft, and Anthropic have all advocated for this kind of1313development, sometimes using the term economic zones to1314envision regional investment areas that fuse permit and energy1315solutions with Federal and state creation, small business1316empowerment, and community level prosperity.1317    Such regional development, including through linkages with1318local testbeds, could foment a powerful surge of optimism in1319communities. We believe that bottom-up economic innovation in1320regions is an important part of this work ahead of us.1321    More boldly, Congress and the Trump administration could1322build new prize competitions or recent challenge grant1323programs, such as the tech hubs' and NSF's innovation engines1324to further develop these regions and hub engines that have so1325much to give.1326    Finally, AI adoption leadership will further hinge on1327ensuring that adequate pools of high-quality talent exist all1328across the country and in every sector and every region.1329    The challenge is both narrow and wide. The narrow part of1330the challenge is the Nation's diminishing homegrown share of1331the world's elite talent. Speaking of this, ITIF has shown that1332while the U.S. attracts and employs a high share of elite1333talent, its domestic production of such talents has slowed.1334That is a worry and needs to be addressed.1335    At the same time, concerns around the broad degree of AI1336readiness are important, and starting now virtually all workers1337will need to understand AI principles, be able to understand1338and direct AI effectively, and be able to evaluate its outputs.1339    AI literacy instruction, as the Department of Labor has1340been describing in recent months, will need to cultivate1341agility and install such human skills as judgment, teamwork,1342creativity, and problem-solving.1343    If that is widely achieved, workers will feel optimistic1344about the technology. If not, I worry they will not, and so a1345supportive AI talent strategy must foster AI readiness among1346both elite and general populations.1347    Fortunately, immigrants and U.S. higher ed are proven1348talent sources. Thoughtful visa reforms can help the Nation1349retain its edge on elite talent, and even more important is the1350need to prioritize broad AI literacy, not just at elite1351institutions but throughout the region and regional AI----1352    Senator Budd. Mr. Muro----1353    Mr. Muro.--the entire workforce system.1354    Senator Budd. Mr. Muro, if we could.1355    Mr. Muro. Yes.1356    [The prepared statement of Mr. Muro follows:]13571358 Prepared Statement of Mark Muro, Senior Fellow, Brookings Institution1359Introduction1360    Chairman Cruz, Ranking Member Cantwell, and distinguished members1361of the committee, I want to thank you for the opportunity to testify1362today on how we can maximize and sustain the value for human1363flourishing of America's extraordinary artificial intelligence (AI)1364sector.1365    My name is Mark Muro and I'm a senior fellow at the Brookings1366Institution. Acknowledging that affiliation, I should note here at the1367top that these remarks and anything I say today are my personal views1368and do not reflect the views of the institution or its other scholars,1369employees, officers, or trustees. These are my own thoughts about a1370timely topic.1371    Nearly every week, America's AI innovators are introducing1372incredible new tools and solutions that are allowing more and more1373people, firms, entrepreneurs, and communities to expand the reach and1374achievement of human skills and expertise. This progress has generated1375significant cause for optimism. In recent years, most notably, AI's1376capacity to drive productivity, advance science, promote health, and1377magnify what humans can do has been demonstrated time and again. These1378achievements reflect the power of AI's special ability to weave digital1379innovations and human skills into a transformative collaboration.1380    And yet, for all that, significant pessimism has begun to1381complicate the moment. Some fear what has been deemed the greatest1382automation technology in human history.i Others worry about1383how they will weather likely employment disruptions and about the1384uncertainty of what skills will serve them in the future. Still others1385worry about the impacts of data center developmentii on1386local communities, and the potential for highly uneven geographic1387build-out of the AI economy.iii1388---------------------------------------------------------------------------1389    \i\ Daron Acemoglu, David Autor, and Simon Johnson, ``Building pro-1390worker artificial intelligence.'' Working paper 34854. (Cambridge:1391National Bureau of Economic Research. 2026).1392    \ii\ Daniel Goetzel, Mark Muro, and Shriya Methkupally, ``Turning1393the data center boom into long-term local prosperity.'' (Washington:1394Brookings Institution, 2026).1395    \iii\ Mark Muro, Shriya Methkupally, and Molly Kinder, ``The1396geography of generative AI's workforce impacts will likely differ from1397those of previous technologies.'' (Washington: Brookings Institution,13982025).1399---------------------------------------------------------------------------1400    In view of that, I want to say a few words about areas where1401Federal support can help maintain the sector's momentum, reinforce its1402value, and promote optimism about its possibilities. Above all we1403should ensure that AI is pro-human and pro-community. In this1404direction, my fellow panelists have done an excellent job of inspiring1405optimism by detailing some of the possibilities. Now I'd like to follow1406them by noting a few areas of policy that have helped businesses like1407those we've just heard about innovate and that will now be needed to1408sustain more of such innovation.1409The nation should build a strong AI-support platform.1410    To speak about what it will take to sustain AI innovation to1411support human flourishing, I want to draw on my work at Brookings on1412``AI readiness'' to suggest that the Nation needs to build a strong AI-1413adoption platform.iv To that end, I would encourage the1414committee and Congress coalesce around a core set of AI-adoption1415readiness priorities. I'll touch on five areas of needed attention:1416---------------------------------------------------------------------------1417    \iv\ Mark Muro and Shriya Methkupally, ``Mapping the AI economy:1418Which regions are ready for the next technology leap.'' (Washington:1419Brookings Institution, 2025).1420---------------------------------------------------------------------------1421Research1422    First, broader AI innovation and adoption requires maintaining a1423dominant AI research base. Abundant research flows generate talent, but1424also ideas, intellectual property, innovation, and start-ups. Given1425that, there is work to be done.1426    To be sure, the United States maintains clear leadership in AI R&D1427and consistently outperforms other countries on innovation and1428investment. In 2024, for example, the U.S. developed 40 AI models while1429the second-best performer, China, produced 15. As to private1430investment, U.S. companies invested $109.1 billion in AI research in14312024, vastly outpacing China's $9.3 billion.xiii1432    Yet recent developments suggest these figures won't be enough for1433the U.S. to maintain dominance. According to the International Data1434Corporation's (IDC) spending report, China is expected to see an 861435percent compound annual growth rate in generative AI investments from14362022 to 2027, meaning the country's generative AI spending could grow1437to represent 33 percent of the world's AI investment--up from less than14385 percent in 2022.xiv1439    Against this backdrop, Federal AI research funding trends are1440concerning. To be sure, U.S. government investment in AI research did1441increase from $2.98 billion in 2024 to $3.32 billion in14422025.v However, this came at a moment when the authoritative1443National Security Commission on Artificial Intelligence (NSCAI) had1444advised doubling non-defense AI R&D investments annually to reach $321445billion by 2026.xvi1446---------------------------------------------------------------------------1447    \v\ NITRD, ``Artificial intelligence R&D investments--FY 2019 to FY14482025.''1449---------------------------------------------------------------------------1450    The nation's current mix of research activities also falls short on1451several areas of potential opportunity. First, U.S. private sector1452investments dwarf government outlays for basic (mostly academic)1453research. This may limit creative leaps forward and progress on novel1454use cases in underinvested sectors. Dwindling support for the NSF-led1455National AI Research Institutes--launched during the first Trump1456administration--is a case in point. Another missed opportunity is the1457thinness of AI research and computing flows into high-quality but1458farther-flung universities. As of now, the Nation's Bay Area Superstars1459and Star Hubs account for about 60 percent of the Nation's R&D flows,1460leaving institutions in many regions underserved.vi1461Likewise, computational support for basic research in academia too1462often remains spotty, likely limiting the Nation's innovation1463potential.1464---------------------------------------------------------------------------1465    \vi\ Muro and Methkupally, ``Mapping the AI economy.''1466---------------------------------------------------------------------------1467    And so, the Nation should build out its platform for broader AI1468adoption by expanding the scale of AI R&D and improving its character.1469To address the scale issue, the Nation should prioritize a step change1470in total AI R&D outlays in the next decade. Rather than cut research1471outlays, Congress needs to ``run faster'' if it wants to outpace China,1472because R&D is a critical accelerant.1473    At the same time, to improve the composition of overall AI1474research, the Nation should expand investments in basic R&D research1475and in mechanisms for broadening access to essential computational and1476data resources. On this front, increased basic research expenditure1477into universities appears critical. But so does rejuvenation of the AI1478Research Institutes program for accelerating research on new topics in1479new universities and locations. Additionally the build out of something1480like the National AI Research Resource (NAIRR) pilot program could1481provide a mechanism for increasing more and different researchers'1482gaining access to high-speed computing resources and datasets,1483including in more and different locations.vii Related to all1484of this, it would be valuable if some Federal research grant-making was1485directed towards investments in ``pro-worker AI'' that supports such1486values as education, human learning, and human decision1487making.viii1488---------------------------------------------------------------------------1489    \vii\ Mark Muro and Julian Jacobs, ``The case for promoting the1490geographic and social diffusion of AI development.'' Washington:1491Brookings Institution, 2024.1492    \viii\ Daron Acemoglu, David Autor, and Simon Johnson, ``Building1493pro-worker artificial intelligence.'' Working paper 34854. (Cambridge:1494National Bureau of Economic Research. 2026).1495---------------------------------------------------------------------------1496    In a word, an expanded and enhanced research agenda is critical for1497creating a national platform capable of bolstering AI innovation,1498entrepreneurship, and adoption in the country.1499Regional innovation clusters1500    Accelerating AI adoption also requires promoting the growth of1501emerging AI clusters in geographic regions. Dense, vibrant AI clusters1502are national assets that contribute to national progress. Neglecting1503such clusters is a missed opportunity that leaves innovation and1504adoption potential untapped.1505    With that in mind, the Nation's overall AI platform should promote1506broad AI adoption through region-focused industrial development.1507Nvidia, OpenAI, and Anthropic have all advocated for this kind of1508development, sometimes using the term ``economic zones'' to envision1509regional investment areas that fuse permit and energy solutions with1510Federal and state job-creation, small-business empowerment, and1511community-level prosperity. Such regional development work could foment1512a powerful surge of optimism in communities.1513    Given that, the Federal government should draw on its recent1514experience with ``place-based'' industrial investment to accelerate AI1515cluster scale-up in promising regions and sectors across the country.1516    What might this look like? Some of this region-catalyzing work1517could leverage the National AI Research Institutes from the previous1518Trump administration, as well as the NAIRR program, by orienting their1519research and computational supports toward the needs of promising local1520clusters. These steps would support early-stage activities in key1521clusters. More boldly, Congress and the Trump administration could1522revamp recent challenge grant efforts--such as the Commerce1523Department's Regional Technology and Innovation Hubs program and the1524NSF's Regional Innovation Engines--to focus new hubs and engines1525specifically for AI.ix A number of the current hubs and1526engines are leveraging AI technologies already. Why not develop new1527centers fully focused on compelling AI verticals grounded in dynamic1528regional ecosystems?1529---------------------------------------------------------------------------1530    \ix\ Muro and Methkupally, ``Mapping the AI economy.''1531---------------------------------------------------------------------------1532    Taken together, approaches like these will help ensure the national1533AI adoption platform promotes new expansion across the Nation's1534regions. They will also foster optimism and counter concerns that AI is1535somehow an industry working exclusively for others somewhere else.1536Talent1537    AI adoption leadership will further hinge on ensuring that adequate1538pools of high-quality AI talent exist all across the economy, available1539to every sector and every region. The challenge is both narrow and1540wide.1541    The narrow part of the challenge is the Nation's diminishing home-1542grown share of the world's elite AI talent. Speaking to this, the1543Information Technology and Innovation Foundation (ITIF) has shown that1544while the U.S. attracts and employs a high share of elite talent its1545domestic production of such talent has slowed.x That's a1546worry.1547---------------------------------------------------------------------------1548    \x\ Trelysa Long, ``AI is powering the U.S. economy, but who's1549powering AI?'' (Washington: ITIF, 2025).1550---------------------------------------------------------------------------1551    At the same time, there are also concerns about the broad degree of1552AI-readiness needed across the Nation's workforce. Starting now,1553virtually all workers will need to understand AI principles; be able to1554understand and direct AI effectively; and be able to evaluate AI's1555outputs, as notes the Department of Labor's new Artificial Intelligence1556Literacy Framework.xi In addition, such ``AI-literacy''1557instruction--as continues the DOL--will need to cultivate agility at1558scale and instill such ``human'' skills as judgement, teamwork,1559creativity, and problem-solving. If that is widely achieved, workers1560will feel optimistic and engaged about AI. If not, they won't.1561---------------------------------------------------------------------------1562    \xi\ Employment and Training Administration, ``Training and1563Employment Notice No. 07-25.'' Washington, 2026.1564---------------------------------------------------------------------------1565    And so, a supportive Federal policy platform for AI needs to foster1566AI readiness among both top scholar echelons and everyone. Fortunately,1567U.S. immigrants and higher education stand out as proven sources of1568talent. Given that, thoughtful visa reforms will help the Nation retain1569its talent lead. But even more important is the need to prioritize1570broad AI education and workforce literacy at higher education1571institutions and all across the workforce system, not only in the usual1572elite locations. Likewise, Congress should support the creation of1573regional AI learning networks, with employer-led, cross-sector1574partnerships that serve as training and innovation centers for the AI1575economy. Aligned to industry demand, all of this training will go a1576long way toward ensuring AI unleashes creativity and optimism among1577American workers. Connecting all of this to the emergence of AI1578industry clusters near universities and community colleges will help1579educate, engage, employ, and retain critical technical talent.1580Infrastructure1581    Boosting regional AI adoption will further depend on the delivery1582of key infrastructure that is not now solidly in place, but on which1583national leadership depends.1584    On this front, the AI era is elevating the need for large-scale1585chip production, vast data and computational resources, new energy1586sources, and the build out of huge data centers in accordance with1587important electricity, water, and other permitting1588issues.xii To be sure, work has begun on some of these1589issues, such as through the CHIPS and Science Act's subsidies for1590semiconductor plant construction and the launch of the NAIRR pilot for1591giving more scientists, innovators, and educators access to the1592computing and data resources necessary for game-changing research.1593---------------------------------------------------------------------------1594    \xii\ Muro and Methkupally, ``Mapping the AI economy.''1595---------------------------------------------------------------------------1596    With that said, AI-related infrastructure gaps stand as major1597impediments to regional and national scale-up. The demand for computing1598resources and energy is projected to challenge available supplies.1599Permitting and grid hurdles exacerbate the delivery problem. And to1600many communities, data center siting decisions seem secretive and1601disruptive--divorced from regional economic planning.xiii1602---------------------------------------------------------------------------1603    \xiii\ Goetzel, Muro, and Methkupally, ``Turning the data center1604boom into long-term local prosperity.''1605---------------------------------------------------------------------------1606    In light of these challenges, the Federal government should work to1607facilitate timely, carefully planned, and environmentally sound data1608center and power development. A portion of that work must clearly1609involve policy and regulatory efforts to bring new clean energy1610generation sources and grid links online. Some of this will involve1611speeding up the complicated federal, state, and local siting and1612permitting process for conventional or nuclear power plants, including1613by leveraging suitable public lands or replacing coal plants. But it1614will also be important to streamline the permitting processes for clean1615energy generation and related transmission capacity.1616    Otherwise, the Federal government should do what it can to1617facilitate strategic data center development. With data center1618development at times disruptive and localities increasingly wary of it,1619Federal and other stakeholders should work with industry to optimize1620the process so that it supports AI build-out that maximizes local AI1621gains.1622    Development could be streamlined and rationalized through the1623establishment of AI economic zones within states or through the release1624of suitable public lands. Likewise, the government could encourage data1625center developers to negotiate beneficial partnerships with local1626stakeholders, which would complement construction with community AI1627development. In this vein, Brookings has suggested how regions might1628trade expedited data center regulatory approvals for shared computing1629resources, research collaborations, and talent1630initiatives.xiv The Federal government could aid in that.1631Federal agencies could create and fund an AI tech hubs program, where1632data center developers and the Federal government are co-investors in1633region-level tech ecosystems along with universities and regional1634firms. Alternatively, Congress could approve funding for AI test-bed1635collaborations involving the co-location of national labs, data1636centers, universities, and startups on Federal land. The Department of1637Energy has already moved in this direction with its plan to leverage 161638Federal land parcels for rapid data center construction on sites that1639have in-place energy infrastructure and fast-track-1640permitting.xv Some of these sites could be managed to create1641community economic development as part of the build out.1642---------------------------------------------------------------------------1643    \xiv\ Ibid.1644    \xv\ Office of Policy--Department of Energy, ``Request for1645information on artificial intelligence infrastructure of DOE lands.''1646Request for information.1647---------------------------------------------------------------------------1648Worker security1649    Finally, any national platform for regional AI scale-up needs to1650include strategies to provide basic worker security. Such provisions1651are necessary because successful AI adoption will involve both gains1652for many workers and dislocation for others. Minimizing worker1653dislocation will smooth adoption, keep talent engaged, and maintain1654morale.1655    Recent work from Brookings shows that higher-tech, higher-value,1656information-based industries--especially in cities--are likely to see1657elevated levels of AI impact.xvi Specifically, Brookings1658analysis suggests 30 percent of all workers could see at least 501659percent of their occupation's tasks disrupted by generative AI in the1660coming years, with higher ``exposure'' levels for higher-skill computer1661and office activities.xvii While some of those impacts will1662enhance worker well-being and create new jobs, others could bring about1663sudden task shifts, depressed work demand, or even chronic under-or1664unemployment. Others may shred long-reliable pathways for worker1665mobility.xviii This matters because such disruption could1666produce ``adjustment'' challenges for local labor markets, weaken1667confidence in the AI revolution, and undermine support for regional AI1668scale-up.1669---------------------------------------------------------------------------1670    \xvi\ Mark Muro, Shriya Methkupally, and Molly Kinder, ``The1671geography of generative AI's workforce impacts will likely differ from1672those of previous technologies.''1673    \xvii\ Mark Muro, Shriya Methkupally, and Molly Kinder, ``The1674geography of generative AI's workforce impacts will likely differ from1675those of previous technologies.''1676    \xviii\ Forthcoming research from the Brookings Institution and1677Opportunity@Work.1678---------------------------------------------------------------------------1679    Given that, the Nation's Federal AI platform needs to provide1680elements of a worker-adjustment strategy that helps regions deliver on1681Vice President JD Vance's promise that AI adoption will bring workers1682``higher wages, better benefits, and safer and more prosperous1683communities.'' xix1684---------------------------------------------------------------------------1685    \xix\ Reuters., ``Quotes from U.S. Vice President JD Vance's AI1686speech in Paris.'' February 11, 2025.1687---------------------------------------------------------------------------1688    Much remains to be worked out on how to deliver this. But for sure,1689the Nation will want to invest more in ``active labor market policies''1690that help people shift into new jobs. These policies may involve rapid1691retraining programs for individuals impacted by AI-related job1692displacement, such as the pilot efforts the Trump administration has1693advanced.xx Relatedly, these policies may entail flexible1694benefits, including for wage insurance, and health care that is not1695tied to one employer. Other supports may include policies that give a1696measure of economic security to workers who want to be retrained and1697learn new careers. For example, Brookings has described the idea of a1698``Universal Basic Adjustment Benefit'' that would help displaced1699workers transition to new work with the help of temporary income1700support that allows for intensified training access.xxi Such1701provisions can provide a measure of stability as the nature of work1702evolves while also kindling optimism among workers who may currently be1703discouraged due to their fears of displacement costs.1704---------------------------------------------------------------------------1705    \xx\ U.S. Departments of Labor, Commerce, and Education.1706``America's talent strategy: Building the workforce for the golden1707age.'' (Washington, 2025).1708    \xxi\ Mark Muro and Joseph Parilla, ``Maladjusted: It's time to1709reimagine economic `adjustment'' programs.'' (Washington, Brookings1710Institution, 2017).1711---------------------------------------------------------------------------1712    In sum, innovative firms of all kinds--whether AI developers1713themselves, the increasing millions of AI-adopting firms, or the1714thousands of AI start ups entering the space--are providing abundant1715grounds for excitement about AI's future potential. Their1716entrepreneurship gives much cause for optimism. And yet, sustaining1717that optimism requires sustaining those firms' growth and Americans'1718confidence in the future--and that means reinvesting in the1719fundamentals of American AI strength and vibrancy.1720    Which is why Congress should build a strong AI-support platform,1721one that begins with robust investments in R&D and regional innovation1722clusters, and that leans in on talent, infrastructure, and worker1723security. Implemented well, such a platform will foster both continued1724growth and a broader, more widely shared confidence in the AI future.17251726    Senator Budd. Thank you so much. If we could take a few1727questions for the rest of the panel.1728    Mr. Muro. Absolutely.1729    Senator Budd. If we have additional thoughts, we will come1730back to that and--when you are recognized, if that is OK.1731    Mr. Muro. Great, thank you.1732    Senator Budd. Yes, thanks again for being here. Thanks for1733your opening remarks.1734    You know, it has been said that the four main inputs of AI1735are talent, compute, energy, and data. The Federal Government1736houses a tremendous amount of scientific and important datasets1737that could be leveraged as strategic assets in the U.S. AI1738leadership.1739    The Open Government Data Act of 2018 requires data assets1740owned by the Federal Government whose sharing would not1741otherwise be prohibited by law to be published in machine-1742readable format.1743    If each of you first three with particular companies--1744operating companies--what would each of your companies' efforts1745to deploy AI--how would that be affected by more access to1746data? If it is AI-ready and if it is machine-readable and1747compatible, how would that affect each of your companies?1748    We will start with you.1749    Mr. Giannikopoulos. Senator Budd, the access to data is a1750critical aspect of development for artificial intelligence.1751    However, in health care the ability to have validation data1752by which you can measure the quality of your solutions that are1753developed by which deployers at the institutional level can1754assess the fit within their personal institution is one of the1755greater challenges.1756    So having more robust access to this machine-readable,1757translatable, and ideally in health care outcomes-linked data1758will provide opportunities to actually assess these solutions,1759not in a vacuum but against real American data as part of that.1760    So, you know, more robust access to that would be a1761significant enabler of adoption and innovation of this new1762technology.1763    Senator Budd. So in the world of radiology, what would be a1764specific dataset that you would look to make more accessible?1765    Mr. Giannikopoulos. Radiology, there are two key areas. It1766is the images and the reports, and the reports in radiology are1767one piece of siloed information. Yet, what happens with that?1768    If a--that quiet lung mass that I referenced in my opening1769statement, if that is identified what percentage of that turns1770into cancer? How do you understand the progression of that?1771Being able to access that longitudinal information at scale1772would enable significant development.1773    Senator Budd. And for something like that in particular1774with radiology, that would not violate patient privacy?1775    Mr. Giannikopoulos. If appropriately de-identified, which1776would be a very important aspect of this, absolutely not.1777    Senator Budd. Great. Thank you for that.1778    Ms. Ng, what are your thoughts for Siemens Industries?1779    Ms. Ng. Chairman, thank you for the question.1780    More available, highly trusted data is gold in industrial1781AI environments and something that is very important to1782recognize is that for industrial AI, the underlying data comes1783from machines. It comes from manufacturing systems and1784engineering datasets within the four walls of that factory or1785the shipyard or production environment.1786    So the more readily available that data is and in usable1787formats, the better the generative insights and recommendations1788that the industrial AI products can make.1789    Senator Budd. That is very helpful, thank you.1790    Dr. Shelton?1791    Mr. Shelton. So I think from an operating company1792perspective, we have to remember that there are really only1793three ways that we can actually have access to data.1794    The most expensive is we can go out and generate it1795ourselves, and you see robotics companies trying to do that1796today by a tele-operation.1797    The second is when the data is available on the public1798Internet for free and this is one of the ways that the modern1799AI frontier labs have made such rapid progress training their1800LLMs specifically.1801    But that data does not exist for robots generally. There is1802no description of human movement or industrial tasks that you1803can really find on the open internet.1804    And then the third way is simulation, and although there1805has been a lot of rapid progress--I am sure you have seen news1806reports with Nvidia and others who have done that kind of1807work--we are still in the early days of that really being1808viable for large scale training of robots.1809    So the government release of data is actually helpful a1810couple of ways. First is it gives us a very broad look at the1811economy as a whole.1812    Although Agility has been focused to date on these internal1813warehouse jobs, there are broader classes of job that would1814pull on things like large geospatial datasets and other large1815public data sources that would be very exciting to have access1816to.1817    And second is I think it gives us a broad representative1818look at what people are doing. That is part of the reason why I1819cited the census data is that the government has a very large1820role to play here identifying areas where there are1821opportunities for robots to get out and do useful work. So I1822will excited to be a part of that.1823    Senator Budd. Dr. Shelton, do you have an example of those1824ideas inside the census data?1825    Mr. Shelton. Yes. I mean, I think what is fascinating with1826this is how large of an opportunity there really was with1827underemployment in sectors like the warehouse and logistics1828world.1829    Feeding things like accident data, human performance data,1830and stuff allows us to focus on the very high-value tasks that1831actually improve worker safety, improve worker health, and1832allow those workers to redeploy elsewhere within the org.1833    So we are excited about being able to tailor what our1834product offerings are to the needs of the broader workforce.1835    Senator Budd. Dr. Shelton, do you think that would help1836with adoption? I mean, there is a general fear out there about1837job replacement as sort of a generalized concern.1838    But if you show massive areas of underemployment in a1839sector such as warehouse and logistics where there is need for1840much more workers, do you think that would help with adoption1841in those areas?1842    Mr. Shelton. Absolutely. In fact, we see a lot of pressure1843from the small and medium business community as well that would1844love to adopt this kind of technology and, you know, it is--1845without name checking a specific retailer, we were approached1846by a small but nationwide retailer a number of years ago,1847looking for a back warehouse automation solution because they1848were having staffing challenges.1849    Unfortunately, it was not possible at the time with the1850technology that was available to us circa 2020 to really engage1851with them. But I think the broader look at the labor pressures1852that are faced by those businesses in particular would be super1853helpful.1854    Senator Budd. To give us a quick snapshot of the timeline,1855if you had that conversation in 2020.1856    Mr. Shelton. Correct.1857    Senator Budd. Now in 2026, if you had the same1858conversation, do you have more capabilities?1859    Mr. Shelton. We do. I think there is a rollout question1860that we have, which is do you tackle the low hanging fruit1861first or do you go for the harder end of the spectrum.1862    But, certainly, technologically the advances that AI has1863shown just in the last I would call it 3 years significantly1864lowers the barrier to entry to address those businesses.1865    Senator Budd. Thank you.1866    Ranking Member.1867    Senator Baldwin. So the administration suggests through its1868recently released Maritime Action Plan that AI can be used to1869increase domestic shipbuilding capacity. Currently, the United1870States produces less than one percent of ships globally.1871    We used to be dominant in the industry. The effort to1872revitalize domestic shipbuilding requires cooperation and input1873from government, the private sector, labor, and more.1874    So, Ms. Ng, where do you see the most promising1875opportunities to use AI to increase domestic shipbuilding1876capacity and how is Siemens working with organized labor and1877workers in the shipbuilding sector to identify opportunities to1878use AI to expand and enable our workforce to increase output,1879including repair and maintenance?1880    I am going to give you a cluster of questions here. Do you1881think we can attract workers to expand our capabilities if they1882continue to hear that AI might displace their jobs?1883    So go at it.1884    Ms. Ng. Ranking Member, thank you very much for the1885question.1886    I will start by sharing a little bit about what industrial1887AI can do to recapture the American dominance of shipbuilding1888here in the U.S.1889    What we are seeing is that by the usage of industrial AI in1890all aspects of the life cycle for shipbuilding is really1891critical, and I will give three quick examples.1892    First of all, in design. Industrial AI allows us to be able1893to test and model thousands of configurations of a ship in1894hours as opposed to months. This allows us to do quicker design1895churn and to be able to design ships very quickly.1896    Second, we are seeing with production we are using1897industrial AI to help basically simulate material flow,1898workforce allocation, and sequencing of work packages so that1899work done in a back shop, on a shipyard, is very efficiently1900sequenced and done quickly.1901    This allows us to execute on that work and on those great1902designs even faster.1903    The third example that I can give is in sustainment. We are1904using industrial AI applications with predictive maintenance to1905be able to preserve uptime so that all of our machines and all1906of our capabilities in the shipyards and in the back shops are1907able to operate efficiently.1908    We are seeing that bringing that entire life cycle view1909together, we are very able to be able to reestablish that1910dominance.1911    The other question that you had around how do we continue1912to support the workforce and make sure that we are being able1913to retain that workforce is, I would say, pretty1914straightforward.1915    We have to have a strategy to attract workforce and retain1916them, meaning the workforce is becoming increasingly digitally1917native. They have expectations. They want to work with top-of-1918the-line software and AI-enabled tools.1919    So we are really honored to be able to provide that as part1920of talent acquisition strategies for manufacturers and1921shipbuilders specifically.1922    Senator Baldwin. And I certainly want to just encourage you1923to have them at the table from the beginning, not introduce1924them at the end.1925    Ms. Ng. Absolutely.1926    Senator Baldwin. Mr. Muro, I have a question for you.1927    Last month, you co-authored a piece titled ``Turning the1928Data Center Boom into Long-Term Local Prosperity.'' In this1929article, you note that local officials have the leverage to1930engage data center developers on becoming true local partners1931in their community.1932    The article highlights some efforts in Wisconsin,1933specifically Microsoft's partnership with the University of1934Wisconsin-Madison, in the development of the AI Co-Innovation1935Lab and their partnership with Gateway Technical College to1936train workers.1937    What should local officials be trying to get out of these1938negotiations? We have got a lot of them going on in Wisconsin.1939    Mr. Muro. Wisconsin actually is a demonstration, almost1940like an aerial view of data center development that is1941benefiting a region that is tied to broader economic1942achievement for the region, too.1943    That is because the--they are technology themselves. They1944are a source of computer processing, high-speed technology of--1945that could be tied in and is being tied into regional1946university activity, research, and so on. And then there are1947users of energy and places for energy innovation.1948    So data centers can be thought of in a very broad way along1949with their important national demonstration or their national1950push toward powering the whole technology.1951    But places have the possibility to work--enter into1952interactions with the hyperscalers early in the process. We1953think there is an opportunity for them to trade essentially1954very quick initial ability to save--excuse me, to establish,1955you know, quick permitting in response and in a trade with the1956companies to build these kind of partnerships in the region.1957    So I think the more a region has a sense of its technology1958goals and where those intersect with the hyperscalers, and I1959think there are lots of areas for particular research, work on1960energy issues, testbeds of all sorts. So I think there is just1961a wider array of possibilities in regions.1962    Senator Budd. Thank you, Mr. Muro.1963    Senator Cruz, you are recognized.19641965                  STATEMENT OF HON. TED CRUZ,1966                    U.S. SENATOR FROM TEXAS19671968    The Chairman. Thank you, Mr. Chairman. Welcome to all the1969witnesses.1970    Let us start with a broad question. AI is a fundamentally1971transformational technology, and like past waves of innovation,1972we do not know exactly how it is going to impact our economy,1973including what it is going to do to employment.1974    One of the greatest fears that I hear from people is1975concern that AI is going to take their jobs and it is going to1976lead to fewer jobs.1977    Mr. Giannikopoulos, what do you say to those concerns? What1978are you experiencing in your current job markets and how do you1979anticipate AI shifting the job market in respective industries?1980    Mr. Giannikopoulos. Chairman Cruz, thank you for the1981question.1982    The crisis in health care is real and it is happening right1983now. There are around 900,000 physicians in the United States.1984The projected shortage of physicians by 2030 is 187,000--excuse1985me, by 2037.1986    The projected shortage of nurses by 2030 is 194,000. So the1987gap is growing. The need is there, specifically within1988radiology. If you look at the attrition rate over the past 51989years, it is up by 50 percent compared to historical norms.1990    Meanwhile, the projection is that imaging volume will rise1991by 26 percent in the next 30 years. So a continued growth and a1992continued need.1993    Dr. Curtis Langlotz, the immediate past president of the1994Radiological Society of North America, recently did a task-1995based analysis of the work of a radiologist, and he estimated1996that within the next 5 years the amount of work a radiologist1997will need to do could possibly go down by 33 percent.1998    Yet, the need for them to perform that understanding and1999ultimately, you know, make those diagnoses so the patient can2000get the care will only increase.2001    So AI is not replacing the physician at any point along2002this. It is enabling them so that we can leverage them to get2003where we need to be as a health care society.2004    The Chairman. Ms. Ng, how would you answer the same2005question?2006    Ms. Ng. Chairman Cruz, in our experience, advanced2007technologies support workforce expansion, not contraction. This2008is especially true in my sector of shipbuilding, where we are2009seeing these advanced technologies drive an increased demand2010for highly skilled trades.2011    AI is not diminishing the need for skilled workers. It is2012amplifying it. AI's expansion reinforces the need for strong2013technical skills and for people who are passionate about2014innovating it.2015    The best example that I have is actually in Fort Worth,2016Texas, where we have the latest and greatest Siemens facility2017that was just stood up. In Fort Worth, we used our digital twin2018and AI-enabled tools to be able to model the entire production2019floor and simulate it before even breaking ground.2020    This facility created hundreds of new jobs in that area,2021and was--and the employees that are working there are operating2022in reduced complexity and much better training.2023    So in summary, the greater risk to jobs, in our opinion, is2024that--is losing industrial competitiveness. It is not2025responsible modernization. Thank you.2026    The Chairman. Dr. Shelton.2027    Mr. Shelton. Yes, thanks for the question.2028    I can give two examples with human labor. The first in the2029warehouse and logistics world is there are sort of two2030problems.2031    First right now is an extreme growth in that sector over2032the last, say, 10 or 15 years--all of us love our next-day2033Prime shipping--has created a completely unsustainable growth2034trajectory in that industry.2035    So there are only two ways you can do that. One is you2036could pull slack out of the rest of the economy from a human2037labor standpoint, and second is you can deploy automation.2038    Companies right now have tried to do both, like the Amazon2039example that I cited. However, as a sector, it is still coming2040up short from a total employment count standpoint. So that is2041the backfill side of this.2042    The second and the more exciting piece of this is I think2043it changes fundamentally the nature of what a job is. So as a2044private pilot, there was a time in the U.S. history where2045delivering air mail meant you put on your goggles, you got into2046your biplane, and you flew across the U.S. personally carrying2047the mail.2048    There is now an enormously complex logistics system for2049shipping things by air, and you could say that you work in that2050industry while someone from the 1930s would simply not2051recognize what the job has evolved into.2052    That is why I got into robotics. That kind of stuff is2053super exciting to me and the evolution of things like2054agriculture. As a notoriously terrible grower of corn, I think2055people do not--are not aware of how hard modern agricultural2056jobs are and just how far we have come on that side of the2057economy.2058    So I am an optimist about this. I do think we should be2059cognizant that there are evolutions of jobs over time and we2060should be sensitive to that, but it is also the path forward.2061    The Chairman. Two final questions. What is something that2062you hope AI could do in your industry that we cannot currently2063do yet, number one? And number two, what is a surprising way2064you have seen AI used that you did not expect?2065    Mr. Giannikopoulos. I will answer the same way, but from2066two different perspectives. If you look at precision medicine2067pathways, AI has already opened those up in ways that we have2068never seen.2069    Take my personal diagnosis journey. It was 10 years to2070that, ultimately--10 years of varying symptoms that were2071ultimately dismissed because I was a white male with Greek2072heritage in Florida, which is not exactly common for multiple2073sclerosis.2074    Being able to identify that this, you know, varying2075numbness, all these different parts and pieces personalized to2076me as an individual and understanding in our broader health2077care system, could have shortened that cycle of diagnosis.2078    That is what it is already starting to do, but with better2079integration, better understanding, better tailoring of the2080medicine and the understanding to the individual, it will be2081able to take that to the next level.2082    This is where we are getting into, you know, predictive2083cancer scores, different things like that, so that you can2084identify a path that a person is on, not when they are already2085on it but before they start it, and really make a big2086difference.2087    The Chairman. OK. Briefly, Ms. Ng, and then Dr. Shelton.2088    Ms. Ng. To answer your first question, we want to see AI2089adopted throughout the entire life cycle as opposed to in silos2090so, say, maybe just design or just production or just2091sustainment.2092    For the second question, what has surprised me the most is2093seeing the application of industrial AI with the U.S. Navy,2094which is my customer.2095    We have been able to work with the Navy to digitally model2096the four shipyards and help them to be able to plan for future2097work.2098    So when you build a submarine, you have a whole maintenance2099plan that comes associated with that, and what is surprising to2100me is the ability to use industrial AI to actually model out2101new dry docks, new back shops, all of this new physical2102infrastructure that does not disrupt the current availabilities2103and work happening there.2104    Thank you.2105    Mr. Shelton. So the very first time we deployed one of our2106robots, Digit, the humanoid, doing a task, it took an engineer,2107as I recall, somewhere about five and a half weeks to really2108prototype that.2109    By 2023--and this video is actually on our YouTube channel2110for anybody who is curious, we decided to hook an early version2111of ChatGPT up to it to see if it could write code, if it could2112run the robot.2113    And shockingly, and I was completely gobsmacked by this, it2114worked on the first try. Nothing in technology ever works on2115the first try so that was a really interesting outcome.2116    Now, at the time, it was not hardened. It was not2117deployable. But as sort of a shot across the bow of things that2118were to come, it was super interesting.2119    I would love--and this gets back to Chairman Budd's2120question about the dataset availability--that that worked is a2121sign that we can take descriptions of tasks that we want robots2122to do and use automated tool sets to get them deployed as2123rapidly as possible.2124    It is an exciting and completely novel side of the industry2125that continues to surprise most of us who have been in the2126field for a while.2127    The Chairman. Thank you.2128    Senator Budd. Thank you, Chairman.2129    Senator Blunt Rochester.21302131            STATEMENT OF HON. LISA BLUNT ROCHESTER,2132                   U.S. SENATOR FROM DELAWARE21332134    Senator Blunt Rochester. Thank you, Chairman Budd and2135Ranking Member Baldwin, and thank you so much to the witnesses2136for this hearing.2137    I get very excited about this topic. I was former secretary2138of labor in Delaware as well as head of state personnel.2139    But I also had the opportunity in the House when I served2140in the House to start a bipartisan Future of Work Caucus,2141because for me, if anybody says they know what the answers are2142or whether the economy is going to--you know, we are going to2143have more jobs or less jobs, you really do not know.2144    This is all so new and so--and so impactful in so many2145parts of our lives. I am on the HELP Committee, so health care.2146I have a nursing workforce shortage bill, so health care is2147important.2148    I am from an ag state so we think about precision2149agriculture. I saw a woman use an iPad in her kitchen to2150control her crops.2151    So to me, this is very important. Our state has a statewide2152AI commission and we are setting up clear guardrails on ethics,2153safety, transparency across models, on training, creating a2154regulatory sandbox, and also new legislation.2155    We also have two land grant institutions, the University of2156Delaware and Delaware State University, that are actually2157turning AI into real-world innovation and workforce pathways,2158and we have a lot of local companies like Qnity who are2159deploying new hardware to improve AI systems.2160    And we also have to pair AI adoption with real safety2161guardrails and policies that ensure that we strengthen our2162workforce instead of sidelining it, and I think, you know, that2163is one of the--for me, there is the--I am a pragmatic optimist.2164    Let us put it that way. So there are the good things, but I2165also understand that we do not want to leave people behind.2166Whether it is a podcast, whether it is magazines like the2167Atlantic, the AI and the Future of Work, it is truly2168everywhere. We are reshaping how we work.2169    I want to really kind of tailor my questions to something,2170kind of picking up on Senator Baldwin about the workforce and2171how the workforce is included as these changes are happening,2172and I will start with Ms. Ng.2173    What does engagement with your workforce look like when it2174comes to designing and implementing AI-related changes? Can I2175start with you?2176    Ms. Ng. Absolutely. Thank you for the question.2177    As I mentioned in the opening statement, what we have found2178at Siemens is that digital transformation truly does work best2179when the worker is at the center, and that is over a multitude2180of ways.2181    The best way that I can provide a story, though, is2182actually a story from the Puget Sound Naval Shipyard. There is2183a superintendent of Backshop 31 at Puget Sound and she truly2184demonstrates what effective partnership looks like when we are2185trying to solve a problem using a digital technology.2186    And what I mean by that is this superintendent, she leads2187from a place of empathy. She knows the struggles and the pain2188points that her workforce is managing. So working with leaders2189like that.2190    Senator Blunt Rochester. Do you have like a--is there a2191specific--and for all of you, I will just submit questions for2192the record because I have a lot of questions and I do not have2193a lot of time.2194    But I will follow up with you on, like, is there a specific2195process that you use of a way to get that information and then2196turn it into your policies?2197    I know that, Mr. Muro, you said in your testimony Brookings2198analysis suggests 30 percent of all workers could see at least219950 percent of their occupation tasks disrupted by generative AI2200in the coming years.2201    So even with the updating of tasks, can someone share with2202me how you include workers in making sure that their jobs--they2203are re-skilled, that they are trained, that they have what they2204need to keep up?2205    And I will start with--maybe I will start with you, Mr.2206Muro. And I have 41 seconds.2207    Mr. Muro. Yes. No, it is--first, those statistics are2208directional. Clearly, there is tremendous uncertainty around2209the--but we are talking about a pathway. We know that there2210will be significant disruption.2211    Disruption, though, is--can be positive or negative and we2212think significant aspects of AI's impact on work will be2213disruptive, but also beneficial.2214    Senator Blunt Rochester. I mean, I understand the2215difference between a horse and a buggy, and now we have cars. I2216get that part.2217    I am looking for--and we will follow up with each of you--2218what specific things are we doing to prepare the workforce and2219also include them in the help and decision-making of how to2220make sure that it is not so disruptive?2221    I love the issue of coding. We were pushing people to go2222into coding, and now the machines can code.2223    And so as I am thinking about the future and how I think2224Dr. Shelton said the changing nature of the workforce, I would2225love to have a conversation with each of you about what can we2226do as Congress to ensure that we are not leaving anybody2227behind, but at the same time we are benefiting from the2228technology that is before us.2229    And I am out of time. I yield back.2230    Thank you, Mr. Chairman.2231    Senator Budd. Thank you, Senator.2232    Senator Blackburn.22332234              STATEMENT OF HON. MARSHA BLACKBURN,2235                  U.S. SENATOR FROM TENNESSEE22362237    Senator Blackburn. Thank you, Mr. Chairman, and to each of2238you, thank you for being here.2239    Establishing a framework for AI is something that is going2240to be very important for Congress to do. President Trump asked2241me to take the first stab at drafting the Trump America AI Act,2242and it is based off of his executive order.2243    I was pleased that the executive order included protections2244for what I call the four Cs: children, creators, communities2245from job loss and high electric rates, and then censorship.2246    We know that AI and the LLMs are biased against a lot of2247conservatives, and in Tennessee I like to say we have a good,2248bad, and ugly relationship with AI. Our manufacturing,2249logistics, health care, really like it.2250    A lot of our innovators that hold patents and trademarks2251and work with our auto industry and then, of course, our2252entertainment industry with our musicians, our songwriters, our2253screenwriters, our script writers are really quite concerned2254about what it is going to do and their ability to protect their2255name, image, likeness.2256    So having the proper guardrails in place are important, and2257we know that every industrial sector has had those guardrails2258except the virtual space, and once we define what the rules of2259the road are, we free up innovators to take off and do great2260work because they know what the playing field is so thereby2261they can develop a strategy to win.2262    So that is the purpose of our having the Trump America AI2263Act and putting these protections in place, and I want to--I2264think it is Mr. Giannikopoulos. Am I saying that anywhere near2265right?2266    Mr. Giannikopoulos. Very close, Giannikopoulos.2267    Senator Blackburn. Kopoulos. I did get very close.2268    Mr. Giannikopoulos. You did.2269    Senator Blackburn. All right.2270    As you know, Nashville area really is the health care2271informatics and health care interactive technology hub, and we2272are seeing so much innovation that is taking place there and we2273are really--when it comes to predictive diagnosis, disease2274analysis, remote surgeries, telehealth even, we see the benefit2275of these applications.2276    The but comes into place when we talk about creating a2277digital health ecosystem and privacy concerns that are there.2278So in your work, I want you to talk about that importance as an2279innovator and a patient.2280    I am going to give you a total of a minute, and then if you2281want to submit something longer form to me, I would welcome2282that.2283    Mr. Giannikopoulos. Excellent. As a developer within health2284care governed by business associate agreements, we are, you2285know, clearly within the scope of HIPAA. HIPAA is an elastic2286rule. It started out as an insurance----2287    Senator Blackburn. It needs to be modernized.2288    Mr. Giannikopoulos. It does need to be modernized, but it2289started out as an insurance portability.2290    Senator Blackburn. In order to include all of this.2291    Yes, as I say, it came about--it covers the fax and the fax2292machine, the paper in the fax machine. So we can modernize it.2293    Mr. Giannikopoulos. It covers the access to data and with2294electronics we have the ability to audit it.2295    Senator Blackburn. That is right.2296    Mr. Giannikopoulos. That is really key with the2297transformation we have gone through.2298    Nashville, as you have mentioned, has led the pack in that,2299and what I see out of the large institutions is they are2300approaching this as workforce transformation. It is not simply2301leveraging technology to do what you have always done.2302    It is how can we leverage the technology to do something2303new with it, and as a patient that is where the benefit sits2304for us, if we can receive better care pathways as a result of2305that.2306    Senator Blackburn. Ms. Ng, I want to come to you and your2307work.2308    Let us talk about the difference between industrial2309applications and commercial applications, because I think this2310gets lost many times as we talk about how consumers approach2311the utilization of a technology and their expectation of that.2312    And as we have developed tech policy through the years, I2313have told people, look at the computer and look at the backside2314and the front side, and there is truly a difference that is2315there.2316    And with AI and the fact that AI, coupled with the other2317computational sciences like quantum, is going to yield faster,2318more accurate results. But talk about the difference in the2319industrial and the commercial.2320    Ms. Ng. Thank you, Senator.2321    You touched on it perfectly, which is that in a2322manufacturing and industrial environment, oftentimes the work2323is safety critical and operating in a high tolerance2324environment, meaning that precision matters extremely2325importantly.2326    So consumer AI is using, you know, all of the Internet to2327analyze data and do various things. Industrial AI is using much2328more controlled datasets from machines on the floor, from2329manufacturing systems, and from digital engineering datasets,2330and it is using that control data to generate insights and make2331recommendations.2332    Ultimately, the human is still the decision authority to2333confirm what to do next.2334    Thank you.2335    Senator Blackburn. And, you know, as we look at training2336our children and we need to have STEM labs in all of our2337schools. Even though much coding will be done by AI from the2338consumer side, kids need to code just like they need to learn a2339language. They need to have that understanding.2340    So we look forward to helping build this policy out as we2341move forward. Thank you.2342    Thank you, Mr. Chairman.2343    Senator Budd. Thank you, Senator Blackburn.2344    Senator Moreno.23452346               STATEMENT OF HON. BERNIE MORENO,2347                     U.S. SENATOR FROM OHIO23482349    Senator Moreno. Thank you. I did not see where--I did not2350see that pattern coming. That is great.2351    Thank you for this hearing. It is, obviously, very, very2352important and very topical.2353    I will start with you, Demetri, because I am not going to--2354I am not going to take a swing at your last name so I will just2355call you Demetri. We are going to make it informal, part of I2356will call you by your first name, so you can call me Bernie.2357    In the 1920s, would it surprise you that there was people2358saying machines will devour man?2359    Mr. Giannikopoulos. I believe they were called cars.2360    Senator Moreno. Exactly. And in the 1950s, an MIT professor2361did a study that said that basically we are stumbling into the2362automation era and all workers will be replaced. Does that2363sound familiar as well?2364    Mr. Giannikopoulos. Absolutely.2365    Senator Moreno. And in the 1980s, there was a book written,2366``The Robot Is After Your Job.'' Does that sound familiar as2367well?2368    Mr. Giannikopoulos. The Excel spreadsheet was also going to2369replace accountants.2370    Senator Moreno. Right, exactly. So the doom and gloom has2371been around forever, and I think it is important for the record2372is just to talk about employment numbers in America from the23731920s to the 1950s to the 1980s.2374    So in 1920s, when machines were going to devour man, there2375was 42 million Americans employed. In the 1950s, that number2376went to 63 million. So, obviously, decently wrong.2377    In the 1980s, that number is 108 million, and in 1990s with2378the Internet that was going to replace all retail businesses.2379There was no point in being in retail--I remember that being in2380the car business--because the Internet was just going to2381completely disintermediate everybody, 131 million people2382employed.2383    In the 2000s, at the beginning of smart phones, 160 million2384people employed, and thanks to the policies that we put in2385place in the--this Congress, we now have a record number of2386Americans employed, 172 million.2387    So the Mark Twain expression, ``The news of our death is2388sorely exaggerated,'' meaning employment continues to grow.2389    Now, with that said, I think what we can all agree on--and2390I want to get your thought, and I will start with you,2391Brittany, sorry, because there is no way I am taking a shot at2392your name either--on what is different about this period of2393time, from your point of view, with the introduction of AI2394versus what I have just laid out?2395    Ms. Ng. Thank you for the question, Senator.2396    There is an incredible story about Albert Einstein riding2397on a train, you know, back in the day, and the conductor is2398coming through, stamping tickets, and he sees Einstein2399scrambling, looking for his ticket.2400    And he looks up and says, ``Mr. Einstein, it is OK, I trust2401that you bought a ticket. I know who you are.'' Continues on2402his way. Sees Einstein still scrambling on the floor.2403    Goes back and says, ``Mr. Einstein, truly, it is fine. I2404know who you are. I know that you bought a ticket.'' And Albert2405Einstein looks at him and says, ``Thank you, sir, I also know2406who I am. What I do not know is where I am going.''2407    And I think that that story really gets at the heart of2408what you are mentioning, which is ultimately what we need to be2409doing now is partnering with industry to define what that2410compelling vision and that roadmap is, and that is what I do2411every single day on the ground at Siemens with my customers is2412partnering with them to define what are the actual business2413outcomes that you are trying to achieve, and how can we fully2414channel these industrial AI solutions and applications toward2415achieving that outcome.2416    So to answer your question, I truly believe that what is2417interesting and new about this technology is that we have the2418ability to explicitly help specific business outcomes.2419    Thank you.2420    Senator Moreno. Yes, and obviously, productivity2421improvements is key, and I think the expression has been used,2422you are not going to lose your job to AI but you may lose your2423job to somebody who uses AI.2424    And, Dr. Shelton, I will kind of shift to you on this part2425of the questioning. Part of what I see as a worry is that this2426is moving much faster than these other rollouts, right?2427    So this is definitely at a speed which we are not used to,2428and what I worry about is are our institutions that train2429people, are they prepared for the speed?2430    What are your thoughts on that?2431    Mr. Shelton. It is a qualified yes, I think, to your2432question, although I would agree that the speed here is very2433different.2434    I think something that everybody in this room has had to2435grapple with with AI is we did not perceive automation coming2436for our jobs or evolving the nature of work for us as, largely,2437knowledge workers.2438    And I think, you know, ironically, people more on the2439manual labor and the blue collar side of the universe has had,2440you know, several hundred years to get used to the idea of2441using tools to accomplish more than they could have2442accomplished on their own.2443    That is why I used the farming example in my opening2444statement. Those of us who deal primarily on the knowledge side2445of the universe have not had options to deploy autonomous2446agents until the last, let us call it 6 months, to go off and2447do portions of the tasks that we are using them for, not to do2448the job, per se, but to do a sub-component of it.2449    So I think all of us who are on the more knowledge end of2450the economy are going to have to evolve very rapidly to start2451understanding our job as managing fleets of autonomous workers2452who are working on our behalf as opposed to doing that work2453ourselves, and that is quite different.2454    From a university standpoint, I think there is going to be2455a rapid pivot back to the liberal arts. This is something that2456my wife, who is an English professor, likes to remind me with2457some rapidity is she has known all along that it was all about2458talking, and I have to concede that that is probably looking2459like it is going to be the case over the next 20 years.2460    Senator Moreno. All right. Thank you.2461    Senator Budd. Thank you, Senator Moreno.2462    Senator Hickenlooper, you are recognized.24632464             STATEMENT OF HON. JOHN HICKENLOOPER,2465                   U.S. SENATOR FROM COLORADO24662467    Senator Hickenlooper. Thank you, Mr. Chair. Thank you all2468for being here. Certainly, a very timely hearing, and I commend2469the Chair for that.2470    Let me start just to talk a little bit about product2471testing.2472    Mr. Giannikopoulos--I think that is close--thank you for2473sharing how Rad AI is integrating the software to support2474quality--the delivery of quality health care to patients.2475    As you know, software, hardware, are rigorously tested2476before they are used with patients. There is endless testing2477and backing up.2478    We are working on bipartisan legislation called the Vet AI2479Act, which promotes evidence-based best practices to help2480companies have their AI tools independently evaluated by2481independent third parties. Let us leave it at that.2482    This helps increase transparency, promote accountability2483for responsible system design. I think, to a large extent, it2484creates trust in customers.2485    So, Mr. Giannikopoulos, would you describe how--what Rad AI2486is doing now to test--you know, to looking at what are its--2487whether its products are sufficiently tested before they go2488into action?2489    Mr. Giannikopoulos. We are a company founded by a2490radiologist for radiologists. Dr. Jeff Chang, the youngest2491radiologist at the time of his graduation, 17 years of age----2492    Senator Hickenlooper. Jesus.2493    Mr. Giannikopoulos. Yes, a very, very smart man. Has been2494central to the design ethos of our solutions. He was a2495practicing radiologist when he created the idea of Rad AI to2496begin with.2497    We have continued to engage with the radiologists, validate2498the usage, and understand exactly how it integrates into the2499workflow.2500    The other key piece that really needs to be built with the2501adoption of AI in health care in particular is trust, and a way2502to gain trust is--there is the technical nature, there is the2503transparency, the information about how you have built your2504models.2505    But there are also leveraging relationships. For example,2506we have one with the RS&A Ventures Group to integrate the2507century of knowledge that RS&A has available directly into the2508application at the point of care for the radiologist.2509    So it is not AI assisting them with the diagnosis, it is AI2510plus trusted and validated information right at the point of2511care.2512    Senator Hickenlooper. Cool. It is worthy of a longer2513discussion.2514    Ms. Ng--is that right? Somehow close? See, I am not scared2515of these, well, you know, difficult names. Unusual names, we2516will call them. I have a difficult name, Hickenlooper.2517    Thanks for your testimony and sharing how all the work that2518Siemens is doing around, you know, how AI is going to enter new2519sectors of the economy and especially the industrial economy.2520    It is clear that we are in the middle of a revolution. I2521think of it as a great transition and I think we are doing2522several things at once. We are going into AI. We are moving2523toward clean energy. This will look back--50 years or 100 years2524from now, they will look back into the beginning of this great2525transition.2526    I think it is going to--in the terms of AI, it is going to2527define how AI supports this growth of small businesses and2528entrepreneurs, but also attracts students to STEM fields and2529ultimately will transform our workforce.2530    Now, when two companies enter into a contract, it is2531essential that the service agreements are transparent but also2532that they are enforceable for--well, for how an AI system can2533be used or not used or misused.2534    So, Ms. Ng, how does Siemens ensure it is transparent with2535its customers about the design limitations of the AI products2536it sells? Do you understand what I am asking?2537    Ms. Ng. Yes. Thank you, Senator, for the question.2538    So transparency is critical to trust, which we were just2539speaking about, and there are several ways that we as Siemens2540ensure that we are providing full visibility into not only how2541industrial AI is being used in environments like a beer2542manufacturing customer that we have in Colorado, but throughout2543all of the sectors in the United States.2544    And it really comes down to two things. One is allowing the2545industrial AI application to be able to share the source data,2546so pulling from which machines, which engineering datasets, and2547which manufacturing systems that it is producing from.2548    The second is providing explanation or evaluation of why2549that recommendation is being made, which comes down to the2550ability that we were speaking about earlier of training folks2551to be able to work in systems oversight types of roles so that2552they can engage more effectively with that industrial AI2553capability.2554    Thank you.2555    Senator Hickenlooper. Great. I am going to hold off. I have2556a couple other questions but I will submit them in writing. I2557appreciate--they are slave drivers here in terms of keeping us2558on our time.2559    Senator Budd. Only because more showed up. So thank you for2560the question.2561    Senator Young, you are recognized.25622563                 STATEMENT OF HON. TODD YOUNG,2564                   U.S. SENATOR FROM INDIANA25652566    Senator Young. Well, thank you to our witnesses for being2567here today.2568    And Ms. Ng, I would like to build maybe on the previous2569questioning and your response. We need to unlock more2570industrial data, spatial data, as I understand it, a lot of it,2571so that embodied AI or AI used to control machines in the2572physical space can continue to move forward.2573    China, a country we care about because we compete with them2574on a number of different levels, they lead the world in terms2575of deploying industrial robots, collecting this sort of data.2576We are going to have to come up, of course, with a different2577set of rules and practices, ones consistent with our values and2578laws.2579    But I think for starters, we need to distinguish between2580industrial AI and consumer AI. I have some legislation with2581several colleagues called the AI Public Awareness and Education2582Campaign Act.2583    This seeks to provide transparency into AI, its promises,2584limitations, and what consumers can expect from it. Too often,2585we think about AI and our minds immediately--I think2586naturally--move toward generative AI rather than the AI we have2587been living with and using for the past several decades.2588    You are correct in your testimony that what we in Congress2589do on consumer-facing AI should not impede the innovation and2590deployment of technology in manufacturing like with the work2591you are doing in the maritime space, for example.2592    As you may know, I introduced some legislation called the2593Ships for America Act. This is legislation to reinvigorate2594American shipbuilding, an item that, fortunately, Republicans2595and Democrats alike have gotten behind.2596    It is something that President Trump has prioritized for2597his administration, and I understand that Siemens recently2598created a new maritime business to address the opportunities2599the company perceived in the shipbuilding and ship repair2600industries.2601    Can you walk me through, Ms. Ng, the benefits these types2602of industrial AI capabilities can have in the effort for2603America to reclaim our shipbuilding dominance?2604    Ms. Ng. Absolutely. Thank you for the question.2605    So what I would start with is by saying that if America2606truly does want to reclaim shipbuilding dominance globally, we2607have to outpace our competitors, not just in labor but in2608building out these digital capabilities, and industrial AI is2609indisputably a force multiplier here.2610    We think about tackling industrial AI and shipbuilding2611across three different areas of a ship's lifecycle. So starting2612in design, through production, and then into sustainment, we2613apply applications of industrial AI across each area of that2614lifecycle, everything from doing faster, more precise design2615using AI-enabled simulation, through predictive maintenance,2616making sure that the machines and the back shop are available2617and ready to roll for every single task that is needed.2618    But if I were to summarize, the most important thing is2619that we are using industrial AI to help sequence work so that2620it is done more efficiently and more quickly, and what this2621does is it reduces rework and it makes work better on the deck2622plate in the shipyards because people are able to execute first2623time--you know, the first time, and operate in a more2624productive environment.2625    Thank you.2626    Senator Young. And as some of my constituents watch this2627hearing they may think, wait a second, more productivity--that2628means higher wages too, right? So that is--that is a byproduct2629oftentimes of these types of investments, whether it is2630yesterday's capital investments or today's AI investments.2631    So I do want to underscore that. What is relatedly Siemens2632doing, Ms. Ng, to upskill the American workforce more broadly2633to ensure workers can succeed in these high-tech, high-demand2634AI-empowered roles?2635    Ms. Ng. Thank you for the question, Senator.2636    We are doing a lot, but if I were to summarize I would pick2637two different things. One is we are working with an ecosystem2638of partners to develop what we call micro credentials for2639shipbuilding specifically but, more broadly, for manufacturing.2640    What this is doing is it is creating pragmatic, kind of2641ready-to-use credentials that might be a day, a week, a month2642long, and they prepare the workforce to go into modern2643shipbuilding and start operating on day one.2644    The second piece is that I am incredibly proud that Siemens2645has committed to training 200,000 electricians and2646manufacturing experts by 2030.2647    We are not doing it alone. We are working with public-2648private partners, universities, community colleges, and trade2649programs to be able to do that, and I am extremely proud to2650share that today.2651    Thank you.2652    Senator Young. Great. Thank you for your answers, Ms. Ng.2653    Mr. Chairman.2654    Senator Budd. Thank you, Senator Young.2655    Senator Cantwell, you are recognized.26562657               STATEMENT OF HON. MARIA CANTWELL,2658                  U.S. SENATOR FROM WASHINGTON26592660    Senator Cantwell. Thank you, Chairman, and I thank you and2661Ranking Member Baldwin for this important hearing, and the2662witnesses for being here.2663    We are here today to talk about some of the most innovative2664things that could help our economy, going forward, but I want2665to mention we do have an NSF AI Education Act that Senator2666Moran and I have introduced, and it talks about some of the2667workforce issues you guys have been discussing and a Small2668Business AI Training Act, that also with Senator Moran, so that2669we are getting this across all aspects of our economy, and a2670Future of AI Act that my colleagues Senator Young and2671Hickenlooper and Blackburn and I. This is about voluntary2672standards for infrastructure and innovation, mostly on safety2673and security.2674    But the thing I wanted to focus on is the science side of2675the equation. I represent a big national lab and--but the2676laboratories across our country represent a unique opportunity2677to take science research that basically might take you years to2678do and now drive it down to months.2679    And so one of the things I am very worried about that we2680cut 10,000 STEM Ph.D.s from the Federal agencies in 2025, so2681that is not a good idea.2682    I represent also one of the most scientific regions of our2683country and so we like scientists because they are in there2684creating the next generation of economic competitiveness and2685solutions and, certainly, when you look at the massive amount2686of AI investment that China's making, you want to keep your2687scientific workforce because they are going to help you.2688    But one of the things that I am most interested in hearing2689from maybe you, Mr. Muro, or Dr. Shelton is how transformative2690biology and chemistry lab work could be compressing it into a2691few months.2692    As I mentioned, the Pacific Northwest Lab is doing that2693already on deploying autonomous experimental platforms that you2694are basically, you know, creating everything from discovery in2695bio energy to advanced materials and fusion, but you are2696getting it done in months instead of years.2697    And so I would assume that people think that this kind of2698AI--I would call it AI-accelerated discovery--is, you know,2699worth the commitment and worth getting this done and probably2700one of the most important things that we could be doing.2701    Dr. Shelton or Mr. Muro, either one.2702    Mr. Muro. I would just say one general thing here is that2703we can make alignments of our--or our computing, our AI2704systems, our data, and our talent, but we have to have the2705ongoing basic scientific activity functioning at a high level2706as well because, really, all of that is essentially trading2707data for the next iteration of all of that innovation.2708    So I think that is one of the things that a comprehensive2709attempt to leverage AI for national good would want to look at2710is making sure that you have all dimensions of the AI machinery2711working, and we do think that actual science--scientific2712activity and the talent it collects is absolutely central.2713    Senator Cantwell. Dr. Shelton.2714    Mr. Shelton. So logistics and warehouse work has already2715seen the problem that in that industry we call islands of2716automation, so you may have a conveyor belt and you may have a2717ground vehicle that is carrying something and you have to get a2718piece of material transferred from the autonomous ground2719vehicle over to the conveyor belt.2720    Academic research has similar problems, although typically2721with higher tech devices where you have existing automated2722processes, say, a PCR machine that does DNA analysis and some2723other piece of lab equipment that you have to connect together,2724and robotics and specifically general purpose platforms are a2725way to address that.2726    There is also--and this is not my field, although I do have2727friends who work in this--one of the most exciting things I2728think that AGI or narrow AGI specifically is able to offer up2729is vastly outsized performance within a particular problem2730area.2731    So if you are familiar with the AlphaFold project out of2732DeepMind looking at protein folding, you can get AIs that are2733superhumanly good at a very narrow task and that, feeding into2734a lab structure that is highly automated, allows you to go from2735the conceptual computation side of it down to the wet lab work2736in a very compressed process.2737    Senator Cantwell. OK. So somebody at home, how do they2738understand what you just said?2739    Mr. Shelton. Sure.2740    Senator Cantwell. I mean, I am just--I am just trying to2741say, we spend a lot of money on our national labs. We have2742already decided they are critical to our competitiveness as a2743nation, and we are also proud of our universities.2744    But when I look at UW versus PNNL, we are talking about a2745size--a huge size difference in the amount of research that is2746done.2747    So now you are saying you are going to apply AI to that and2748basically translate that science into faster application, as I2749am saying, not years, but months, then I think this is a huge2750initiative that we should be undertaking is to take all of that2751research.2752    You just gave one example. You are basically taking one of2753our big research arms and you are basically saying, let us make2754sure we apply AI to it because it really is one of our most2755competitive R&D efforts, right?2756    Mr. Shelton. One of the things, and we have seen this in2757robotics research itself, is you have, you know, a human who2758comes up with an idea and wants to see that translated into2759something practical.2760    So for your question of how would I explain this to someone2761who is not sitting here in the perspective of robotics land,2762you have an idea. You have to get that translated into some2763sort of physical reality and then you have to test that2764physical reality.2765    I think where AI allows us to inject the best short-term2766safety-controlled piece of this is to work with the human2767researcher to get the math part of the process turned into2768something physical that can then be tested, whether it is in2769the medical space or elsewhere in industry, and allow us to2770actually reduce something to practice as rapidly as possible.2771    Senator Cantwell. Yes, I--just the last point. I know I am2772over my time.2773    It is that we are spending, like, about $10 billion on lab2774efforts that are all about all these issues. So I am just2775saying one area to double your investment is to basically say,2776we are spending $10 billion here. So we have already decided we2777think this is really, really important.2778    We are saying apply AI to that effort to make it more2779efficient, because we have already decided that is where we are2780spending our money, and you just gave two really good examples2781of what that acceleration can deliver.2782    So thank you, Mr. Chairman.2783    Senator Budd. Thank you, Senator.2784    Senator Rosen, you are recognized.27852786                STATEMENT OF HON. JACKY ROSEN,2787                    U.S. SENATOR FROM NEVADA27882789    Senator Rosen. Well, thank you, Chair Budd, Ranking Member2790Baldwin. It is really an important hearing and I want to thank2791all the witnesses for being here today.2792    And so I am going to start with you, Ms. Ng. I am going to2793talk a little bit about AI standards and trust.2794    Earlier this year, I led a congressional delegation to CES2795in Las Vegas with a few senators. I have been leading one2796almost every year--senators from this committee. We were able2797to stop by the Siemens booth, hear how Siemens is innovating on2798AI, other new tools, and I am excited you are here to testify2799today.2800    In your testimony, you outlined how using AI in advanced2801manufacturing has a potential for enormous benefits. However,2802the financial risk of an inaccurate outcome from an unreliable2803AI tool is still very high.2804    Therefore, your customers have to have a clear incentive to2805ensure AI tools that they integrate with their products are2806trustworthy and reliable.2807    And so what standards of trust and reliability are your2808customers demanding? Are there best practices or standards that2809this Center for AI Standards and Innovation should consider2810from the industry?2811    Ms. Ng. Absolutely, Senator, and I am so glad that you got2812to visit CES. It is always a wonderful event.2813    To answer your question, the most important thing in2814defining trust is having really strong underlying data that2815informs the industrial AI application.2816    So what that means is being able to tap into machine-2817generated operational data on a shop floor, manufacturing2818systems data, and digital engineering datasets that all come2819together to be able to support analysis and recommendations and2820generate insights.2821    So what is interesting is that that is a huge piece of my2822business today at Siemens is supporting our customers and2823delivering what we call the sort of digital authoritative2824backbone for engineering.2825    That serves as the foundation of which all operational2826capabilities are built on top. Thank you.2827    Senator Rosen. Thank you. And, you know, I really do2828believe, and I am sure you do too, without clear Federal2829standards and guardrails, AI tools that are sold to consumers,2830small businesses, large businesses, but small businesses in2831particular, they have less, if any, leverage. They may not be2832safe and reliable.2833    They could have potential to cause significant harm. So we2834do have to be sure that we are paying attention. And so I want2835to talk about small business and AI adoption because many2836companies have prioritized AI adoption, trying to find where2837this tool can improve efficiency, add value to their business,2838right? How do you grow?2839    However, we need to ensure that the potential benefits, as2840we are seeing from AI, do not continue to disproportionately2841benefit only the largest corporations that have the capital and2842the influence, really, to adopt safe AI and large language2843models and all the other things that go along with it.2844    So I want to stay with you, Ms. Ng. You have a range of2845customer sizes, large legacy brands like PepsiCo to very small2846businesses, startups.2847    What is the biggest challenge you would say that your2848smaller customers face adopting AI? And what can Congress do as2849we think about--I am also on the Small Business and2850Entrepreneurship Committee so I want to think about the large2851businesses and the small.2852    What can Congress do to ensure that the AI adoption gap it2853just does not widen, leaving our smaller innovators potentially2854behind?2855    Ms. Ng. Absolutely, Senator, and this is an incredibly2856important question.2857    We think about adoption for small and mid-sized businesses2858in three ways. It is awareness, ability, and willingness.2859    So awareness is being able to communicate and understand2860the art of the possible, of truly what are the business2861outcomes that industrial AI can unlock for that business.2862    Ability is, yes, to your point, sometimes capital, but it2863is also the aligned incentives with both leadership and, as2864Ranking Member Baldwin mentioned, the ability to also have the2865factory floor workers involved in the process.2866    The third piece is willingness, and this is my biggest2867passion in life is organizational change management. So making2868sure that there is an intentional rollout plan to all segments2869of the workforce to ensure that that adoption happens.2870    Senator Rosen. You set me up for--I only have a few seconds2871for my last question because your passion is going to lead me2872to Mr. Muro to ask this question.2873    AI literacy, right? Because you are enthusiastic about2874getting it out to everyone, but can we talk about how AI2875literacy closes that AI adoption gap that we want to happen?2876    So thank you for setting me up for that question.2877    Mr. Muro. I would just say that this is where regional2878ecosystems can be extremely helpful. Regional learning2879ecosystems, regional technology ecosystems, and as part of that2880work to make sure that adoption does include these features.2881    More and more regions are beginning to find their own level2882on these technology solutions, and I think the Federal2883Government has historically, you know, had a role in supporting2884regional economic development.2885    I think we can be more pointed about that and maybe making2886this a central area, because the talent piece--and the talent2887piece are intimately connected to the--to the kind of broad2888adoption that we are talking about technologically.2889    Senator Rosen. Well, thank you. Thank you again for all of2890your work for being here.2891    Great hearing. Thank you, Mr. Chair.2892    Senator Budd. Thank you, Senator Rosen.2893    Ranking Member Baldwin, you are recognized for additional2894questions.2895    Senator Baldwin. Thank you. Senator Rosen has cued me up2896for my next question and my last question as we do a quick2897second round before closing.2898    Mr. Muro, I appreciate your highlighting worker security in2899your testimony. In Wisconsin, we are very proud of being the2900first in the country to pass a law relating to apprenticeships2901and registered apprenticeships. We did that in 1911.2902    The U.S. did not follow suit until 1937, but with the first2903apprenticeship law in the country we continue to break2904enrollment records in part because our apprenticeships focus on2905in-demand fields.2906    I would like it if you could talk a little bit more about2907the opportunities to integrate artificial intelligence into2908registered apprenticeships and how these programs can help with2909retraining workers in an AI world?2910    Mr. Muro. I mean, first, I think our view of2911apprenticeships is highly shaped by its industrial past, and2912that is an incredibly important theme and a place that it can2913be utilized and merged with AI.2914    AI can also add to the learning and training dimensions of2915apprenticeship through tutorials, AI--AI online supports, and2916all of those.2917    But I think that we should think of apprenticeship as2918especially a way to get hands-on work experience, and we are2919going to need that more and more because it is going to become2920clearer that strictly, you know, higher education degrees are2921not going to be maybe more--maybe more vulnerable to change2922than very specific subject matter, hands-on experience with2923people learning a technology.2924    So I think that this is an extremely important dimension2925that you have put your finger on here because in some ways,2926apprenticeship points exactly at a--at a gap that I think we2927are going to face in especially launching careers and setting2928up pathways for development.2929    Senator Baldwin. Thank you.2930    Senator Budd. Thank you, Ranking Member Baldwin.2931    Mr. Giannikopoulos, there is--since you are in the2932radiology world, there was a famous prediction made by the2933father of AI. I am sure you are familiar with it.2934    I think it is Geoffrey Hinton. He said that with the rise2935of AI there would be no more need for radiologists, but someone2936as smart as he was is completely wrong, and your company, your2937work, just proves that wrong every day.2938    How could someone as smart on AI as he was originally get2939that sort of prediction wrong? And how could we use that as2940sort of wisdom for the future as we think about the application2941of AI in--personally and commercially?2942    Mr. Giannikopoulos. As we look at all of this, all of this2943transformation, it is not unique to radiology. That was a field2944that was early identified as potential for, you know,2945automation and replacement. That did not work out.2946    It is part of the way the AI is trained. You know, it is2947trained on this medical information that is generated by a2948radiologist through an interpretation process. That judgment2949cannot be replaced. There are going to be edge cases. There are2950going to be zebras.2951    Again, my personal diagnostic journey, I do not fit any2952algorithmic assessment that would normally get to that2953diagnosis, which is why we need the human directly involved.2954Like, medicine is both a science and an art, being able to put2955that together and synthesize it.2956    Now, what AI offers as an opportunity is the ability for2957the radiologists, the clinicians, to synthesize more,2958understand more, access it without having to dig through2959records and, you know, all these complicated systems, but2960instead see it presented to them in a really easy way so that2961they can make that determination.2962    And if you look at rural health care in particular, that is2963an area where we need to augment those institutions, and in2964North Carolina, ARA Health Specialists in Asheville where they2965were originally based out of, they cover most of western North2966Carolina right now and serve as the safety net to make sure2967that patients do not have to be flown over mountains literally,2968you know, to get to other health care systems.2969    They do that by early adoption of technology. I met them 202970years ago on the documentation side, and in my last company2971they were the first--some of the first in the states to adopt2972image analysis for improvement of understanding.2973    Now they are using new documentation tools like ours to be2974able to augment that and integrating the Radiology Society of2975North America evidence directly into that to really speed it2976up.2977    So I cannot comment on how Jeff got that quite so wrong,2978but I can share he did get it wrong and today, they also said2979self-driving cars would take over by this time, too.2980Radiologists are still driving to work in their cars.2981    Senator Budd. We are still waiting on that one.2982    I just want to thank each of you and thank the--those that2983came and asked great questions. I thank the Ranking Member and,2984again, each of your companies and allowing time for you all to2985be here today.2986    Senators will have until the close of business on March 102987to submit questions for the record. The witnesses will have2988until the close of business on March 24 to respond to those2989questions.2990    This concludes today's hearing. The Committee stands2991adjourned. Thank you.2992    [Whereupon, at 11:56 a.m., the hearing was adjourned.]29932994                            A P P E N D I X29952996        Prepared Statement of Derek Monson, Executive Director,2997                          Sutherland Institute2998                 25 Ways AI is Helping People Flourish2999I. Introduction: A Pro-Human Framework for Artificial Intelligence3000    Chairman Budd, Ranking Member Baldwin, and members of the3001Subcommittee, thank you for the opportunity to submit this statement3002regarding the intersection of Artificial Intelligence (AI) and human3003flourishing. While much of the public debate surrounding AI focuses on3004controversy and risks, AI is already creating concrete improvements to3005human lives behind the scenes.3006    Drawing from the Sutherland Institute's recent report, 25 Ways AI3007is Helping People Flourish (February 2026), authored by Ford Copple and3008myself, my statement emphasizes a pro-human approach to AI policy. As3009articulated by Utah Governor Spencer Cox during the recent Utah AI3010Summit, such an approach empowers workers with better tools,3011strengthens communities through accessible innovation, and enables3012problem-solving at an unprecedented scale. Below, I outline 25 specific3013examples where AI is currently improving life in education, government,3014health care, and family life. It is vital that AI regulation does not3015jeopardize these proven benefits, as it is contradictory to call3016safeguards pro-human if they eliminate the very tools helping Americans3017today.3018II. Education: Empowering Teachers and Students3019    AI innovations are helping teachers prioritize their time more3020efficiently and effectively while aiding students in far-reaching ways.30213022   1.  Reducing Burdens on Teachers: Tools like SchoolAI in Utah help3023        teachers spend less time on administrative duties and more time3024        focusing on instruction. This shift allows educators to give3025        more personalized attention to their students' unique needs.30263027   2.  Protecting Student Safety: New monitoring systems powered by AI3028        automatically detect firearms on campuses through security3029        camera feeds. By enhancing situational awareness, these tools3030        provide an additional layer of protection against school3031        shootings and enable faster emergency response.30323033   3.  Personalizing Student Learning: AI tools are adapting3034        instruction to each student's unique strengths, weaknesses, and3035        pace. This technology creates personalized learning plans at3036        scale, which increases student engagement and improves overall3037        outcomes.30383039   4.  Helping with Learning Disabilities: Software such as Dysolve AI3040        is specifically designed to help students with dyslexia improve3041        their reading skills. Local parents have noted that this kind3042        of support levels the playing field for children with learning3043        disabilities.30443045   5.  Increasing Student Engagement: Tools like MagicSchool AI allow3046        students to personally engage with historical figures, bringing3047        history to life. These unique lessons allow for a level of3048        personal interaction with curriculum that was previously3049        impossible.30503051   6.  Improving Reading and Language Skills: AI tools like Amira are3052        improving the fluency of bilingual students. Students who do3053        not speak English as their first language can interact with3054        these tools to practice and refine their reading and speaking3055        skills.3056III. Government: Efficiency and Integrity3057    AI offers innovations that create efficiencies that save taxpayer3058dollars and make essential government services more effective.30593060   7.  Saving Taxpayer Dollars on Infrastructure: AI-generated3061        ``digital twins'' allow public infrastructure projects to be3062        tested under simulated scenarios before construction begins.3063        This predictive capability reduces costs and improves3064        efficiency by an estimated 20 percent to 30 percent.30653066   8.  Reducing Wasted Time: AI tools reduce routine administrative3067        tasks, allowing government employees to save up to two weeks of3068        time annually. This frees civil servants to focus on high-value3069        work rather than paperwork.30703071   9.  Improving Election Integrity: AI helps election officials flag3072        errors in processes and simulate various security scenarios.3073        These applications strengthen election security while3074        simultaneously reducing unnecessary waste.30753076  10.  Better Emergency Response: In Utah, AI helps screen non-3077        emergency calls to speed up 911 response times. This ensures3078        that emergency operators can stay focused on life-threatening3079        situations while other needs are addressed.30803081  11.  Faster Approvals while Protecting Public Safety: The FDA uses3082        the AI tool Elsa to streamline clinical protocol reviews and3083        reduce the duration of scientific assessments. This allows3084        lifesaving and life-improving treatments to reach the market3085        quicker without sacrificing patient safety.30863087  12.  Cutting Government Red Tape: Cities like Portland use AI to ease3088        the burden of complex permitting requirements for citizens and3089        businesses. This saves significant time for both the public and3090        government employees by simplifying bureaucratic hurdles.3091IV. Health Care: Better Lives and Better Outcomes3092    AI is helping improve and save American lives, putting citizens in3093the driver's seat for their own health, providing better health3094information, and helping practitioners make more informed decisions3095about care.30963097  13.  Treating Chronic Diseases: A new Utah pilot program uses AI to3098        speed up prescription renewals for chronic disease patients.3099        Simultaneously, the system detects dangerous medication3100        interactions to ensure patient safety.31013102  14.  Better Lives for Prosthetic Patients: Researchers at the3103        University of Utah have integrated AI with prosthetic limbs to3104        create ``bionic'' limbs. These advancements help patients3105        perform everyday tasks, such as reaching and gripping objects,3106        in more natural ways.31073108  15.  Saving Lives from Breast Cancer: AI detects breast cancer in3109        mammograms with increased speed and accuracy. Because early3110        detection is a critical variable in survival rates, this3111        technology is a literal lifesaver.31123113  16.  Quicker Treatment for Debilitating Disease: The Cleveland Clinic3114        Genome Center uses AI to detect Parkinson's Disease using3115        genetic, proteomic, and pharmaceutical datasets. This allows3116        doctors to spot early warning signs far earlier than was3117        previously possible.31183119  17.  Upgrading Heart Disease Diagnosis: University of Utah Health and3120        Intermountain Health successfully used AI to predict heart3121        disease onset and outcomes. This is especially helpful for3122        analyzing large health data records that are traditionally3123        difficult for humans to process.31243125  18.  Faster Skin Cancer Treatment: Doctors use AI algorithms that can3126        scan for skin cancer in five minutes with 99.9 percent3127        accuracy. This dramatically increases the speed of diagnosis3128        and leads to much faster treatment for patients.31293130  19.  Improving Recovery from Injuries: Apps like Wound Assistant3131        allow patients to monitor their own healing process from home.3132        Additionally, AI wearable devices using miniature cameras are3133        being developed to rapidly improve the healing process for3134        physical injuries.3135V. Family Life: Support and Connection3136    While family policy is often fiercely debated, we must not lose3137sight of the benefits AI is already offering to parents and families.31383139  20.  Connecting Loved Ones with Language Barriers: AI-generated sign3140        language avatars and interpreters help the deaf and hearing-3141        impaired connect with family members. Generative AI translators3142        are assisting over 70 million sign language users globally in3143        communicating with loved ones.31443145  21.  More Time to Spend with Children: Virtual assistants like3146        Ohai.ai help busy parents manage daily tasks like school3147        schedules. This technology frees up parents to spend more3148        quality time with their children.31493150  22.  Better Customer Service: Companies like Remi use AI to simplify3151        the process of replacing a home's roof. Customers can scan3152        their roof with software and receive a quote for repairs almost3153        immediately.31543155  23.  Saving Money, Boosting Home Values: Tools like Neighborbrite3156        generate custom landscaping designs tailored to a homeowner's3157        specific environment and taste. This reduces both the cost and3158        the hassle of home improvement projects.31593160  24.  Healthier, Quicker Family Meals: AI tools craft meals catered to3161        a family's specific dietary requests and restrictions. These3162        tools simplify the often-complex daily question of ``what's for3163        dinner'' for busy working families.31643165  25.  Improving Child Safety: Smart baby monitors like Monai use AI to3166        ensure infants are safe and healthy in their cribs. The monitor3167        sends alerts to a parent's phone if it detects concerns like a3168        covered face.3169VI. Conclusion: A Call for Balanced Innovation3170    As a society, we are in the initial stages of a civilization-3171altering technological change. While controversial uses draw attention,3172AI is quietly and consistently improving our lives behind the scenes,3173and promising even bigger life improvements in the future. The3174regulatory decisions made today will determine whether AI remains a3175source of human flourishing or becomes a generational lost opportunity.3176We urge the Subcommittee to follow a pro-human path that protects3177against harm without stifling the many essential ways AI is helping3178Americans thrive, both today and tomorrow. Thank you for the3179opportunity to submit this written statement.3180                                 ______31813182                        Associated Builders and Contractors3183                                      Washington, DC, March 3, 202631843185Hon. Ted Budd,3186Chair,3187Senate Committee on Commerce, Science, and Transportation,3188Subcommittee on Science, Manufacturing, and Competitiveness,3189U.S. Senate,3190Washington, DC.3191Hon. Tammy Baldwin,3192Ranking Member,3193Senate Committee on Commerce, Science, and Transportation,3194Subcommittee on Science, Manufacturing, and Competitiveness,3195U.S. Senate,3196Washington, DC.31973198Dear Chairman Budd, Ranking Member Baldwin and Members of the U.S.3199            Senate Commerce, Science, and Transportation Subcommittee3200            on Science, Manufacturing, and Competitiveness:32013202    On behalf of Associated Builders and Contractors, a national3203construction industry trade association representing 67 chapters, more3204than 23,000 member companies and millions of construction workers, I3205thank you for holding this important hearing to examine how artificial3206intelligence can enhance safety, increase productivity and improve care3207across industries. As the voice of America's merit shop construction3208industry, ABC and our contractor members are committed to building the3209Nation's infrastructure safely, efficiently and responsibly. AI and3210emerging construction technologies are transforming the way we work,3211empowering contractors to deliver projects faster, safer and more3212efficiently than ever before.3213    ABC launched its Technology and Innovation Initiative in 2020 to3214integrate construction technology into every facet of our strategic3215priorities, particularly industry-leading safety, total human health3216and workforce development. Through this initiative, ABC is helping3217contractors thoughtfully assess, adopt and leverage technology3218solutions that strengthen safety performance, address labor shortages3219and manage compressed schedules.3220    At ABC, safety is the foundation of everything we do. AI and3221advanced technologies are reshaping jobsite safety management in3222profound ways. Contractors are digitizing safety inspections, audits,3223checklists and incident reports, replacing paper-based systems with3224real-time, data-driven tools. By tracking both leading and lagging3225safety indicators, contractors can identify risks earlier and intervene3226before incidents occur. According to ABC's 2025 Health and Safety3227Performance Report, companies that track these indicators experience a322862 percent reduction in total recordable incident rates and a 653229percent reduction in Days Away, Restricted, or Transferred rates3230compared to those that do not. These measurable improvements show that,3231when technology is integrated into a strong safety culture, it leads to3232safer jobsites and ensures more workers return home safely each day.3233    Beyond safety, AI and related technologies are transforming3234productivity and cost control. Labor and productivity tracking systems3235provide real-time visibility into workforce deployment and materials3236usage, enabling contractors to manage resources more effectively and3237protect thin margins. Jobsite monitoring tools, including 360-degree3238imaging and drone-based documentation, allow owners, general3239contractors and subcontractors to collaborate with unprecedented3240transparency.3241    Paul Hedgepath, director of virtual construction for ABC member MJ3242Harris, serves as chair of ABC's Construction Technology and Innovation3243Committee. He explained that one of the industry's biggest challenges3244is simply accessing the information contractors already have. ``The3245data is there--but it is fragmented and time-consuming to retrieve,''3246he noted, pointing to the thousands of requests for information,3247submittals, specifications, schedules and safety documents stored3248across multiple systems on a typical project. Secure, construction-3249specific AI platforms allow field leaders to ask direct questions3250within their own project data and receive source-linked answers in3251seconds. ``This is not automation replacing people. It is decision3252support that reduces search time and increases clarity,'' Hedgepath3253added. With faster access to verified information, superintendents can3254prevent mistakes before work begins, and project managers can shift3255time away from document hunting toward risk management and execution,3256improving both safety and productivity when deployed with proper3257governance and clear use cases.3258    Through ABC's Tech Alliance--a curated group of leading3259construction technology companies--and our Tech Marketplace, we provide3260contractors, most of which are small businesses, access to cost-3261effective digital solutions. These partnerships deliver tools that3262support bidding, project management, safety analytics, workforce3263management and field collaboration. Our annual Tech Reports, including3264insight papers from Hensel Phelps and Dodge Construction Network,3265provide case studies and forward-looking insights to help contractors3266evaluate and implement AI responsibly.3267    With innovation comes responsibility. AI evolves rapidly, and3268without clear guidance, its use can introduce risks, including data3269privacy concerns, misinformation and bias. ABC encourages contractors3270to establish clear AI usage policies that ensure technologies are3271deployed safely, ethically and in compliance with applicable laws.3272    Congress plays an important role in fostering an environment where3273innovation can thrive while protecting workers, taxpayers and national3274security interests. Policymakers can support responsible AI adoption in3275construction by promoting regulatory clarity and ensuring that small3276businesses have access to the tools and training necessary to compete3277in a digital economy.3278    AI is not replacing the skilled men and women of the construction3279industry; it is empowering them. By augmenting human expertise with3280real-time data, predictive insights and intelligent automation, AI3281helps contractors complete projects on time, reduce costs and, most3282importantly, protect the health and safety of their workforce. The3283result is stronger infrastructure, greater productivity and safer3284jobsites across America.3285    ABC and our members stand ready to work with Congress to advance3286policies that support innovation and strengthen the construction3287industry's ability to build safely, ethically and efficiently.3288            Sincerely,3289                                        Kristen Swearingen,3290                                Vice President, Government Affairs.3291                                 ______32923293                                                      UVEye3294                                         Teaneck, NJ, March 3, 202632953296Hon. Tedd Budd,3297Chairman,3298Subcommittee on Science, Manufacturing, and Competitiveness,3299Committee on Commerce, Science, and Transportation,3300United States Senate,3301Washington, DC.3302Hon. Tammy Baldwin,3303Ranking Member,3304Subcommittee on Science, Manufacturing, and Competitiveness,3305Committee on Commerce, Science, and Transportation,3306United States Senate,3307Washington, DC.33083309Dear Chairman Budd and Ranking Member Baldwin:33103311    On behalf of UVeye, a New Jersey-based developer and manufacturer3312of advanced vehicle inspection systems, I am writing to commend the3313U.S. Senate Committee on Commerce, Science, & Transportation3314Subcommittee on Science, Manufacturing, and Competitiveness for its3315upcoming hearing entitled, ``Less Hype, More Help: AI That Improves3316Safety, Productivity, and Care.'' As a company focused on vehicle3317safety, we believe AI-driven technologies can bring tangible benefits3318in safety and productivity to the American people and help preserve and3319improve the safety of vehicles nationwide.3320    UVeye is dedicated to creating safer vehicles and safer roadways3321for drivers, passengers, and pedestrians, by providing objective,3322consistent, and instant evaluations of the condition of a vehicle that3323help identify issues before they become safety hazards. Our patented3324AI-driven system uses 360+ imaging to scan each vehicle in seconds,3325detecting under-body damage, tire wear, exterior dents or scratches,3326alignment issues, and windshield damage. The system operates in all3327weather conditions--rain, snow, or mud--and delivers standardized,3328easy-to-read inspection reports. UVeye systems are installed in3329Original Equipment Manufacturing (OEM) facilities, auctions,3330dealerships, and heavy-duty/commercial fleets, scanning more than 2.53331million vehicles each month. This includes customers testing and3332deploying autonomous vehicles (``AVs''), another AI-enabled technology.3333Ensuring Vehicle Safety3334    UVeye's technology is helping make vehicles safer and more3335efficient by identifying issues that may be invisible during manual3336inspections--such as leaks, under-body damage, and worn or outdated3337tires--that can lead to breakdowns or accidents. UVeye provides an3338``always on'' inspection process to substantially increase the3339likelihood of detecting safety issues. Faster inspections also keep3340vehicles on the road and mission ready.3341Protecting Consumers3342    For new-and used-vehicle buyers, UVeye provides transparent,3343documented condition reports reducing ``hidden defect'' risk and3344liability disputes. At vehicle auctions, the system reduces disputes3345over damage by providing visual proof and standardized inspection data.3346By improving the accuracy and consistency of vehicle inspections, UVeye3347supports fairer pricing, improved confidence in used-vehicle markets,3348and fewer ``surprise'' repair costs for consumers.3349Boosting Efficiency3350    UVeye's automation reduces the burden of manual inspection,3351resulting in faster, more cost-efficient inspections. Earlier detection3352of damage or malfunctions means fewer breakdowns, less downtime, and3353lower warranty and repair costs, generating savings for businesses and,3354ultimately, consumers. When scaled across commercial fleets, AV3355operators, and OEM supply chains, UVeye's technology yields multiplier3356benefits in safety, transparency, and cost-efficiency.3357    These are just some of the ways UVeye's AI-enabled technologies are3358benefiting Americans across the economy. As the Subcommittee considers3359how it can support the further development and deployment of AI-enabled3360systems, we urge you to consider more real-world success stories like3361ours, that demonstrate the true potential of AI technologies.3362    Thank you again for holding this critical hearing. Please do not3363hesitate to contact our counsel, Ariel Wolf (ASWolf@Venable.com) if you3364have any questions on this letter, or if there is any further3365information we can provide to the Subcommittee. We look forward to3366engaging with members of the Subcommittee on this and other issues as3367it continues its work to support innovative and tangible uses of AI3368technologies.3369            Sincerely,3370                                                                 UVeye.3371                                 ______33723373                                                      March 3, 20263374Hon. Ted Budd, Chairman,3375Subcommittee on Science, Manufacturing, and Competitiveness,3376Committee on Commerce, Science, and Transportation,3377United States Senate,3378Washington, District of Columbia.33793380Hon. Tammy Baldwin, Ranking Member,3381Subcommittee on Science, Manufacturing, and Competitiveness,3382Committee on Commerce, Science, and Transportation,3383United States Senate,3384Washington, District of Columbia.33853386RE: Subcommittee hearing, ``Less Hype, More Help: AI That Improves3387            Safety, Productivity, and Care''33883389Dear Chairman Budd and Ranking Member Baldwin,33903391    ACT | The App Association (ACT) appreciates the opportunity to3392submit this Statement for the Record for the Senate Committee on3393Commerce, Science, and Transportation's Subcommittee on Science,3394Manufacturing, and Competitiveness hearing titled ``Less Hype, More3395Help: AI That Improves Safety, Productivity, and Care.'' This hearing's3396focus on concrete uses of AI that improve safety, productivity, and3397care is timely. Americans most often encounter AI incrementally through3398practical improvements to existing tools, such as streamlining3399workflows, enabling accessibility, improving diagnostics, optimizing3400logistics, and strengthening cybersecurity.3401    Small businesses are leading the way on AI. As some of the leading3402consumers, developers, and adapters of AI tools, ACT members have a3403major stake in how policymakers view AI markets. ACT represents an3404ecosystem valued at approximately $1.8 trillion domestically,3405supporting 6.1 million American jobs.\1\ ACT members are innovators3406that create the software bringing your smart devices to life. They also3407make connected devices that are revolutionizing healthcare,3408agriculture, public safety, financial services, and virtually all other3409industries. We are concerned that state-level efforts to regulate AI3410technologies before the risks and the benefits of their use are fully3411understood could unnecessarily preempt ACT members' ability to compete3412in AI markets and leverage the technologies.3413---------------------------------------------------------------------------3414    \1\ https://actonline.org/wp-content/uploads/APP-Economy-Report-3415FINAL-1.pdf.3416---------------------------------------------------------------------------3417    AI is an evolving constellation of technologies that enables3418computers to simulate elements of human thinking, including learning3419and reasoning. In practice, Americans most often encounter AI3420incrementally through improvements to existing digital services, such3421as streamlined workflows, image analysis, voice recognition, and3422predictive analytics. We urge policymakers to recognize these3423applications as ``narrow'' AI, meaning systems that already deliver3424significant societal benefits and are widely deployed by small and3425medium-sized innovators. Examples of these ``narrow'' AI applications3426include detecting financial fraud, strengthening cybersecurity,3427enabling accessibility tools for people with disabilities, improving3428healthcare diagnostics, and supporting more efficient infrastructure3429and resource management. These ``narrow'' and applied AI systems3430already deliver significant benefits in the real world. The policy3431challenge is to ensure that governance approaches remain risk-based,3432interoperable, and grounded in existing legal guardrails, while3433supporting the infrastructure, standards, and workforce needed for3434responsible deployment at scale.3435    ACT urges the Subcommittee to keep the following principles in3436mind.3437Existing Law Already Provides Strong Guardrails Against Harmful Conduct3438    A wide range of Federal and state laws already prohibit harmful3439conduct regardless of whether AI is involved,\2\ and there is no AI3440exemption. Section 5 of the Federal Trade Commission Act\3\ prohibits3441unfair or deceptive acts or practices (UDAP), and state consumer3442protection statutes apply similar standards to digital services. These3443authorities reach a broad set of practices relevant to AI-enabled3444products and services, including misleading claims about safety or3445efficacy, failing to mitigate reasonably foreseeable risks, or3446designing systems in ways that create foreseeable harms. Together,3447these existing frameworks already provide meaningful guardrails for AI3448developers and deployers.3449---------------------------------------------------------------------------3450    \2\ How Existing Laws Apply to AI Chatbots for Kids and Teens,3451Georgetown Law Institute for Technology Law & Policy (Nov. 10, 2025),3452https://www.law.georgetown.edu/tech-institute/insights/how-existing-3453laws-apply-to-ai-chatbots-for-kids-and-teens/.3454    \3\ Federal Trade Commission Act Sec. 5(a), 15 U.S.C. Sec. 45(a)3455(prohibiting ``unfair or deceptive acts or practices in or affecting3456commerce'').3457---------------------------------------------------------------------------3458    Congress could take meaningful action by passing a comprehensive3459Federal data privacy law, laying out a single set of rules for how3460companies should handle consumer data. In the absence of a Federal3461privacy law, we urge Congress and Federal agencies to continue3462evaluating the application of these longstanding authorities before3463considering new legal structures. To support a more grounded approach,3464ACT launched a research initiative examining how existing federal,3465state, and local laws already apply across common AI use cases.\4\ This3466project aims to map notable current legal obligations across areas such3467as civil rights, consumer protection, privacy, safety, labor,3468competition, and intellectual property, demonstrating that AI systems3469are already subject to extensive legal accountability. By establishing3470a clear baseline of existing law, this effort aims to inform Federal3471coordination, reduce regulatory duplication, and help policymakers3472distinguish between genuine gaps and areas where improved guidance or3473standards may be more effective than new legislation.3474---------------------------------------------------------------------------3475    \4\ ACT | The App Association. Mapping Existing Laws to AI, https:/3476/actonline.org/wpcontent/uploads/3477ACT_Mapping_Existing_Laws_to_AI_Outline.pdf.3478---------------------------------------------------------------------------3479Use Risk-Based Governance and Voluntary Standards, Not Fragmented,3480        Precautionary Regulation3481    ACT strongly supports risk-based approaches to AI governance3482aligned with recognized standards of safety, efficacy, and reliability.3483Frameworks such as the National Institute of Standards and Technology3484(NIST) AI Risk Management Framework (RMF) provide a flexible and3485interoperable foundation for managing risks across diverse sectors and3486use cases.\5\ In addition, industry-led, voluntary, consensus-based3487standards are a cornerstone of U.S. technological leadership. The U.S.3488approach reflected in OMB Circular A-119 has historically supported3489innovation, interoperability, and global adoption while allowing3490standards to evolve with rapidly changing technologies.\6\ Congress3491should continue to support agency participation in voluntary standards3492development and remove barriers that limit small business engagement in3493these processes. A standards-forward, risk-based approach is especially3494important for the types of applied AI at issue in this hearing,3495including industrial systems, robotics, and health-care workflow tools.3496These domains often depend on multi-layer supply chains and3497interoperable technology stacks where clarity, testability, and3498practical risk controls matter more than formalistic compliance.3499---------------------------------------------------------------------------3500    \5\ National Institute of Standards and Technology. AI Risk3501Management Framework. U.S. Department of Commerce, https://3502www.nist.gov/itl/ai-risk-management-framework.3503    \6\ https://www.whitehouse.gov/wp-content/uploads/2017/11/Circular-3504119-1.pdf.3505---------------------------------------------------------------------------3506Clarify Roles and Shared Responsibility Across the AI Value Chain3507    Effective governance depends on clearly understanding the roles and3508responsibilities of different actors across the AI value chain, from3509developers to deployers to downstream integrators and end users.3510Assigning obligations based on demonstrated harms and each actor's3511ability to mitigate risk promotes accountability while avoiding3512misplaced burdens. To support this approach, ACT developed an AI Roles3513& Interdependencies Framework aligned with the NIST AI RMF, describing3514key stakeholders across development, distribution, deployment, and use,3515and identifying practical responsibilities that support safety and3516reliability.\7\ This shared-responsibility model reflects the reality3517that risk mitigation is most effective when implemented by the actors3518best positioned to identify and reduce risks at each stage of the3519lifecycle.3520---------------------------------------------------------------------------3521    \7\ ACT | The App Association. AI Roles & Interdependencies3522Framework (May 2024), https://actonline.org/wpcontent/uploads/ACT-AI-3523Roles-Interdependencies-Framework-final-text-May-2024-UK-English.pdf.3524---------------------------------------------------------------------------3525The Need for a National AI Framework to Avoid a Patchwork of State Laws3526    Congress should avoid a fragmented AI regulatory environment.3527States are already advancing a growing number of AI-specific laws, many3528of which would impose overlapping or inconsistent requirements for3529disclosures, data practices, and risk assessments.\8\ This patchwork is3530creating significant uncertainty for developers and deployers,3531especially small businesses that lack the resources of their larger3532competitors to navigate 50 different compliance regimes. Early evidence3533shows that even small deviations in state requirements can trigger3534substantial compliance costs, diverting limited resources from product3535development to legal review, recordkeeping, and bespoke technical3536implementations. Recent analyses show that state AI laws modeled on3537European-style precautionary approaches can impose unknown and3538potentially significant costs to companies,\9\ while delays in state-3539level implementation underscore the difficulty of managing complex AI3540obligations at the state level.3541---------------------------------------------------------------------------3542    \8\ Stevens, Morgan. ``State of Confusion: How a Patchwork of AI3543Laws Hurts Small Businesses and U.S. Competitiveness.'' ACT | The App3544Association, Oct. 2025, https://actonline.org/2025/10/10/state-of-3545confusion-how-a-patchwork-of-ai-laws-hurts-small-businesses-and-u-s-3546competitiveness/.3547    \9\ ``The Hidden Cost of AI Regulations: A Survey of EU, UK, and3548U.S. Companies.'' ACT | The App Association, Oct. 2025, https://3549actonline.org/the-hidden-cost-of-ai-regulations-a-survey-of-eu-uk-and-3550u-s-companies/.3551---------------------------------------------------------------------------3552    A single national baseline is preferable to a system where the most3553restrictive state rules dictate outcomes nationwide. Federal preemption3554of conflicting state AI requirements, paired with state attorney3555general enforcement of that Federal baseline, would give consumers3556clarity, reduce compliance burdens, and ensure that small business3557innovators can safely build and deploy the beneficial AI tools that3558enhance safety, productivity, and care for all Americans.3559Conclusion3560    ACT appreciates the Subcommittee's leadership in focusing on3561practical AI tools that improve safety, productivity, and care.3562Congress can reinforce the benefits of applied AI by emphasizing: (1)3563risk-based governance grounded in existing law; (2) voluntary standards3564and interoperable frameworks; (3) clarity on shared responsibility3565across the AI value chain; and (4) a national approach that avoids3566fragmented, inconsistent state-by-state mandates. We stand ready to3567work with the Subcommittee, Federal agencies, and standards bodies to3568support responsible AI deployment that delivers measurable benefits for3569workers, consumers, and patients, while ensuring that small business3570innovators can continue to compete and contribute to American3571technological leadership.3572    Thank you for your consideration.3573            Respectfully submitted,3574                                            Graham Dufault,3575                                                   General Counsel,3576                                             ACT | The App Association.3577                                   Kedharnath Sankararaman,3578                                                  Policy Associate,3579                                             ACT | The App Association.3580                                 ______35813582        Prepared Statement of the National Council on Disability3583    Dear Chairman Budd, Ranking Member Baldwin and Members of the3584Subcommittee:35853586    I am writing as Acting Chairman of the National Council on3587Disability (NCD), an independent, bipartisan Federal agency that3588advises Congress, the President, and other Federal agencies on matters3589affecting the lives of people with disabilities, to provide this3590statement for inclusion in the written record of this Subcommittee's3591hearing, ``Less Hype, More Help: AI That Improves Safety, Productivity,3592and Care.'' NCD is providing this statement to make policymakers aware3593of the susceptibility of artificial intelligence (AI) to develop3594explicit and implicit biases about people with disabilities and to3595advise policymakers on effective ways to ensure that these technologies3596are developed with data sets that include people with disabilities.3597    While NCD's research recognizes the potential benefits of utilizing3598AI to improve healthcare outcomes for people with disabilities, there3599exist some vulnerabilities in these technologies that could negatively3600impact the diagnosis and treatment of people with disabilities and3601provide policymakers with erroneous information rather than accurate3602solutions.3603    The ultimate goal of AI is to create machines that can make the3604same decisions as humans.\1\ NCD's 2024 report, titled The Implicit and3605Explicit Exclusion of People with Disabilities in Clinical Trials,3606analyzed the use of AI in clinical trials.\2\ One study NCD examined3607described how technologies such as AI, machine learning, and natural3608language processing can be incorporated into several aspects of3609clinical trial research.\3\ Some examples include data mining,3610prescreening for possible participants, and automating invitations to3611possible participants who have been prescreened through automation.3612Academic researchers and the pharmaceutical industry are using AI to3613mine and utilize data from electronic sources such as health records3614and devices.\4\3615---------------------------------------------------------------------------3616    \1\ Harrer S, Shah P, Antony B, et al., ``Artificial Intelligence3617for Clinical Trial Design,'' Trends in Pharmacological Science,36182019;40(8):577-591. doi:10.1016/j.tips.2019.05.005.3619    \2\ National Council on Disability, ``The Implicit and Explicit3620Exclusion of People with Disabilities in Clinical Trials,'' (2024)3621available at: National Council on Disability | Federal report3622illuminates need for disability inclusion in clinical trials.3623    \3\ Von Itzstein MS, Hullings M, Mayo H, et al., ``Application of3624Information Technology to Clinical Trial Evaluation and Enrollment: A3625Review,'' JAMA Oncology, 2021;7(10):1559-1566. doi:10.1001/3626jamaoncol.2021.1165.3627    \4\ Woo M, ``An AI Boost for Clinical Trials,'' Nature,36282019;573(7775): S100-S102. doi:10.1038/D41586-019-02871-3.3629---------------------------------------------------------------------------3630    Our 2024 report found that a contributing factor to health care3631disparity outcomes for people with disabilities was physicians'3632erroneous assumptions about the values and expectations of their3633patients, assumptions that mirror widespread, stigmatized societal3634views about the disabled.\5\ Due to these concerns, NCD found that3635disability cultural competence should be a core strategy for the3636healthcare system in order to reduce healthcare disparities for people3637with disabilities.\6\ Strong evidence exists that cultural training for3638health care professionals improves providers' knowledge, understanding,3639and skills for treating patients from culturally, linguistically, and3640socioeconomically diverse backgrounds.\7\3641---------------------------------------------------------------------------3642    \5\ Shakespeare T, Iezzoni LI, Groce NE, ``Disability and the3643Training of Health Professionals,'' Lancet, 2009;374(9704):1815-1816.3644    \6\ Association of American Medical Colleges, Cultural Competence3645Education for Medical Students, 2005. https://www.aamc.org/download/364654338/data/culturalcomped.pdf.3647    \7\ Govere L, Govere EM, ``How Effective Is Cultural Competence3648Training of Healthcare Providers on Improving Patient Satisfaction of3649Minority Groups? A Systematic Review of literature,'' Worldviews on3650Evidence-Based Nursing, 2016;13(6):402-410. Accessed January 16, 2024.3651https://sigmapubs.onlinelibrary.wiley.com/doi/full/10.1111/wvn.12176.3652---------------------------------------------------------------------------3653    While NCD's 2024 research was limited to AI and clinical trials, we3654reasonably believe our findings and recommendations can be generalized3655to other forms of AI technology in healthcare. Our findings were based3656on published studies, legislation, and clinical trial protocols as well3657as subject matter expert interviews, trial participant interviews,3658healthcare provider and participant surveys as well as feedback from3659stakeholders at the National Institutes of Health (NIH) and Food and3660the Drug Administration (FDA).3661    Because AI is intended to develop the same decision-making3662capabilities as humans, NCD is similarly concerned that AI technologies3663may inadvertently develop the same assumptions and biases about people3664with disabilities. For this reason, NCD advises policymakers to review3665the use of AI in healthcare in general and establish regulations as3666needed to ensure that these technologies are built on data sets that3667include people with disabilities, so that implicit and explicit biases3668are not accidentally developed or ``learned.''3669    Thank you for the opportunity to provide a brief summary of NCD's3670relevant research, analysis, and recommendations on ways to improve the3671use of AI technology in the healthcare system for people with3672disabilities. We welcome the opportunity to brief the Committee and its3673staff in depth on any of these or related topics at your direction and3674request.3675    To that end, please do not hesitate to contact our Executive3676Director, Ana Torres-Davis, atorresdavis@ncd.gov, and Director of3677Legislative Affairs and Outreach, Anne Sommers McIntosh,3678amcintosh@ncd.gov, who will be glad to address any request for follow-3679up you may have or provide a more in-depth briefing on any of our3680reports and advisement.3681            Respectfully,3682                                               Neil Romano,3683                                                   Acting Chairman,3684                                        National Council on Disability.36853686                                  [all]