Witnesses and representatives also discuss the issue of data privacy and the need for federal regulation. The subcommittee will come to order in the chair will recognize itself for an Opening Statement for five minutes. I will spend my first little bit to say we have news on our Ranking Member and we Exchange Messages over thanksgiving weekend. What a wonderful lady, what a wonderful person and someone who really puts this committee, the subcommittee and the institution of congress. We have another year to Work Together. You will be sorely missed. Thank you. [applause] oh my goodness. [applause] thank you. We still have a year to get things done and we will get things done. I would just like to thank our witnesses for being here today. Ive got to get my statement in and four minutes. That was important. I should not use all my time. Thank you to our witnesses for being here today. Issues regarding Artificial Intelligence. It is important that we shine a light on a roll ai can play in solving some of our most significant healthcare problems. They are already changing the way in which clinicians pay for their patients and how they conduct Clinical Trials. As ai continues to drive innovation in healthcare it is essential that they examine the meaningful benefits at any potential unintended consequences that these technologies could have. The potential benefits are seemingly without limit. Technology could help our Healthcare System save lives by better predicting potential diagnosis and help us reduce redundancies in our system. We have already seen this play out in real time and i watch unimaginable advances in healthcare. As a result of generative ai. Numerous Success Stories and using Critical Research and development to give research to the market sooner. This was a case by mit scientists to be used as an effective antibiotic. They use ai to help detect earlystage cancers. It is being used to augment existing. This is not to say we should let the use of these technologies go without guardrails. Over the next several months, some fundamental questions regarding the role they will play in our Healthcare System. Trained with supervised ai to drive outcomes. Are these technologies trained with unsupervised ai . Based off Human Behavior for healthcare consumers. Are these technologies trained with reinforced ai which they are really rewarding the systems for which that is generated. They are very complex and difficult things we need to explore as we move forward. These cases are important to remember that every decision comes with a cost. Both human and financial. There constantly monitoring someones heart rate caloric intake and outtake in sleep patterns can help. Extreme cardiac event. Using user data to predict better lifestyle habits. How are we ensuring this data is secure and ensuring they have full control over this information and it is not being used or sold without their consent. A major health event. Are there protocols to ensure individuals are taking unnecessary trips to the emergency room and incurring debt as a result. In closing i support the real possibilities they can bring to our Healthcare System and most importantly to patients. We should get the technology the license to coexist alongside clinicians, patients and innovators as well as regulators while also remaining vigilant of how this technology is being used. I look forward to the discussion today and i will yield back and continue my praise for the Ranking Member. We have a year to Work Together. You treated me with most respect in the areas where we did not agree, you challenge my thinking sometimes. Sometimes we had to move forward it was always with the utmost respect. With that, i will recognize the Ranking Member for five minutes for her Opening Statement. Thank you very much, mr. Chairman. Your words mean a great deal to me. I am deeply moved and touched by the expression of the members of the subcommittee. You know how much i love this committee. We have gotten so many important things done, gotten them over the finish line. Lets optimize our time so we can continue that tradition. Thank you to each one of you. You are all my friends. My fellow americans, my fellow colleagues. There is so much that i want to say. Really, there are not words to express how deeply deeply grateful i am. And the messages board and, i kept them all and i reread them before i go to sleep at night. They are really beautiful. Its like falling asleep on this magnificent cloud of goodwill. Thank you from the bottom of my heart. Here we are to discuss ai and healthcare. The nexus between the two is really very, very important one. It represents an Incredible Opportunity for our country. It has the potential to make our Healthcare System more efficient improve on physicians. New ways to use ai in the Healthcare System are consistently in the news. I think all of you see this in your national clip as you read them at the end of every day. New york times in the october of this year. New diagnosis, brain tumors on the operating table. Forbes in august of this year. Ai is a game changer for toughest areas of Drug Discovery wall street journal in november of last year. While straightbacked researchers use ai to probe for weaknesses. Despite this incredible permit, ai we know and some of the fathers and mothers of ai have instructed us on the potential that ai has, its the other part of the spectrum to worsen Patient Outcome and exacerbate inequities that we have in our Healthcare System if it is not deployed with adequate guardrails. Earlier this month reports found that a lawsuit now alleges that United Health group, one of the largest insurers in our country used in ai algorithm to wrongfully deny medicare to beneficiaries. Yay i algorithm made decisions about patient care that went against the recommendations of the patients own physicians. Another example is our nations children are being left behind as ai and medical imaging rapidly expand. There are no computer aided detections, computerassisted triage or computer aided diagnosis radiology progress authorized for pediatric use is. These pediatric radiologists are working with children. You cannot experiment on children. In many areas, the red lights are blinking and we need to Pay Attention to that. Children are not little adults. I am working on a proposal to address this gap for pediatric patients. In my view and the view both republican and democratic members of both the house and the senate have created legislation that would democratize a i. Today the resources, massive resources are really in the hands of very large hightechnology companies. So we can get that over the finish line in this congress. So then please the witnesses are here. Im happy to beat with each one of you and my colleagues. I think this commit it can lead the effort on ai as it applies to health care. I dont think this is you know easy stuff. We have two rise to this challenge and i think we have the capacity to do so so thank you mr. Chairman and i yield back. Thank you. The gentlelady is back in the chair will recognize the Charitable Committee chair rodgers for five minutes. Mr. Chairman. My heartfelt warm wishes to the ranking democrat eshoo. Shes been a trailblazer for so many including me and im grateful for her outstanding leadership and your friendship these years. This is not a Court Hearing that the energy and Commerce Committee is held across our subcommittees on the subject of Artificial Intelligence. Artificial intelligence has the potential to transform every aspect of our lives for better or for worse. Its critical that america, not china is the one addressing the ai challenges in meeting in the technologys development. The best way to start is by laying the groundwork to protect peoples information with a National Data privacy standards. This is a foundational First Step Towards a safe and prosperous ai future in health care and beyond. I look forward to continuing to discuss how we can improve protections for americans as we incorporate ai tools. Im proud of each of our subcommittee chairs for leading on this important issue. Ai has a unique role to play in the future of health care. Ai can help find the next break through here or improve their ability to catch deadly diseases earlier. Weirsdale Artificial Intelligence can aid in the assessment of medical Indemnity Medical and then in anonymity and ease burdens on Health Care Providers for the recall heard from providers in her district about the burdens of necessary. Cumbersome paperwork and how often this leads to burnout for doctors and nurses and how it eats up time we could be spending providing patient care. For my entire tenure in congress one of the top issues that we have struggled with has been finding ways to cut paperwork redundantly in our Health Care System so we can let doctors do what doctors do best, treat their patients are at 4 years we have nibbled around the edges of this issue but the future of ai could be transformative and hopefully let doctors be doctors instead of administrative staff. We will hear more from doctors schlosser on hca how its being tested out of hospital stay to be clear a i will not solve all the problems with americas Health Care System. One concern we have frequently heard is the potential of human biases to be implicitly based into Ai Technology. The first piece of Health Care Legislation this committee advanced is my bill to ban the usage of quality adjusted life years or quality which are discriminatory measures that are used by federal payers to deny Health Care Services to people with disabilities and chronic illness. If the ai is reliant on quality or other similar measures in clinical decisionmaking are most vulnerable will be left behind. No one here wants to advocate for discrimination and we need to be conscious about federal programs and Ai Technology incorporates these types of biases and what we should be thinking about in this area. I will close by saying that im optimistic about this technology. Think these technologies can make a difference in the lives of patients in this committee needs to lead the way in this innovation. For america to lead we must strike the right balance with ai. He gives businesses the flexibility remain agile as they developed these cuttingedge technologies while assuring responsible use. A National Standard for the selection and handling of data will help Health Care Providers and every american was with clear understandable protections wherever they are. Todays hearing hopefully sheds more light on the current landscape of ai in health care and hopefully provide us with further insight on the next steps that we should take. I yield back. The gentlelady yields back. The chair will now recognize the gentleman from new jersey Ranking Member for an Opening Statement. Thank you mr. Chairman i want to start out by thinking or two democratic members of the subcommittee for their retirement at the end of next year. I told them both they should change their minds that they dont want to. Let me start with Ranking Member eshoo who served as the top democrat on the subcommittee for the last five years including four years as chair and she let the subcommittee. The cobra 19 pandemic played a Critical Role in our success for efforts to user fees but thats just in the last session. And shes been on the Health Care Committee for a long time. A lot of you dont know that i have worked with and not even before energy and Commerce Committee which he played a Critical Role outside of the committee in many ways particularly with the armenian corridor. If it was not for anna lee would not have recognized the Armenian Genocide which is so important in our history when that happened and we did that. So thank you anna and tony to say tony is a long time leader on the subcommittee. Hes also served as the vice chair for Consumer Protection and commerce subcommittee for years. Hes led several other key efforts to put consumers first including a new law that protects babies from dangerous sleeping products. Again i want to emphasize tonys role outside of the Commerce Committee. With the caucus, he has played a tremendous role in promoting not only latino members but also the issues that are important to them. In both cases ended in this case as well as tonys case what they have accomplished here is important for committee that goes beyond the committee so thank you both. I had to say we still have another year left of their contributions and they are still more to be done. Let me go if i can to the issue that we are dealing with today. We are exploring how Artificial Intelligence is changing health care now and potentially in the future. This is an important hearing because the integration that ai present opportunities in patient care and streamline product to bring more efficiency to the Health Sector and at the same Time Congress must recognize the complex ethical and economic and social concerns raised by this sector of the greater deployment of ai in the Health Care System but as we will hear data is central to the use of ai in the delivery of health care as the patients medical data protecting individual information of privacy. I remain concerned the use of ai and health care is generated a significant risk and critical safeguards in place to protect the privacy and security of the patients data and ive said that each of our ai hearing this era strongly believe is the bedrock of any ai regulation we must have a strong protections for all consumers. Ai cannot function without large quantities of data and we must ensure that this increase data demand does not come at the expense of consumers right to privacy. Im going to continue to push for a comprehensive National Privacy standard. Under the chair is just as concerned but i believe its the only way we can limit the unscrupulous Data Collection until practice. Its only way we can ensure all of her personal medical information is protected on line and protected against algorithmic bias in security breaches. Ais role in rich education of medical claim statistically poses a grave concern to me. Its my potential to revolutionize the Health Care Landscape a i answered mrs. Could result in the denial of medical care potentially worsening Health Equity. Right now the classaction lawsuit against a major use of ai to deny medical claims made ai systems played a role in the survival of over 300,000 payment requests in a twomonth period. The average time supposedly revealed its claim is 1. 2 seconds. Ai tools can aid in Health Care Providers but the recommendation should not serve as a substitute for the nuanced judgment of her Health Care Professionals. Ai has the potential to supplement medical decisions however when Health Care Companies driven by efficiencies implement ai suggestions without critical scrutiny i worry that patients could be put at risk. So theres a lot of work to be done here and i want to thank the chairman of the subcommittee and the chair of the full committee for prioritizing this. Thank you and i yield back. At the gentleman would deal for a couple seconds. I didnt see the tony come into the room when i talked about annie. Several issues and one i remember when we had issues to help Small Businesses that were affected and we are focused on working together. It was a great experience so congratulations to you and we have another year to Work Together. The gentleman yields back. Thanks and we well have more time to move forward this year. We will see what happens, right . Now that concludes member statements and we will move to our witnesses Opening Statements but i will introduce each one of you and calling you. If i introduce you as a group and call a new to give your Opening Statement and those of you who have not testified some of you may have not you have a green light. The green light is for five minutes quick sale five minutes to testify and in four minutes of a green light then you get a yellow light and once you have a yellow light you have one minute and when the red light hits its time to wrap up and move forward. Today i will first introduce dr. Michael schlosser Senior Vice President of transformation and innovation of nca health care. Dr. Benjamin nguyen and i hope that with the proper pronunciation of your name. Mr. Peter shen dr. Christopher longhurst chief officer in Association Data u. S. San diego health and also dr. David newmantaker director of the division of neural visual vestibular Disorders Department of neurology professor of neurology at Johns Hopkins university. Appreciate you all for being here taken the time to be here today. This hearing is important and i will begin by recognizing dr. Schlosser for five minutes free Opening Statement. Make sure your microphone is on. And you have it up to your face fuel lifted up towards you. Thank you subcommittee and make member and chairman and Ranking Member toll on and members of this committee for inviting me to testify here today. Im dr. Michael schlosser Senior Vice President for transformation innovation in health care. The background in neurosurgery and operations and involvement in health care with Ai Technology. I want to share my perspective on ai in health care. Our commitment to integrating Ai Health Care is driven by a vision to enhance patient care and Operational Efficiency and effectiveness. Our cases are focused largely and removing administrative work from provisions and providers so we can align the focus on patients critical decisionmaking and other highrisk activity. A lot of my colleagues function at the top of their licensing. Expanded Health Care Workforce at the time of and the tools to deliver a superior standard of care. To achieve these goals are first up was to develop a responsible ai Program Involving a robust government structure to ensure ai applications are fair robust accountable and continuously evaluated for safety. The program is governed and enables the use of ai across organizations. Ensuring that technology is used responsibly but also ensuring to take full advantage of these innovations and the benefits they provide to our care team impatience but i comes to privacy and security we have several decades of experience in protecting patient data to meet the challenges of deploying ai in a secure and private manner prevailing on our experience under the hepa standards we ensure that all are ai applications appeared to be strict. Data and output of any ai model could include protecting Health Information about our patients as well as the models themselves protected in the same way. Private and secure is a key feature of our responsible ai program. We are deploying a new Data Architecture to support ai in our ai agenda which focuses on identifying datasets. The primary source of Training Data. This allows for the use of large data to develop models based analytics without another safeguard we have implemented is a human centric approach to ai. In all of our ai locations we provide a human in the loop approach for this interest we leverage a upper efficiency and accuracy we dont compromise on safety and responsibility. The human model allows for ongoing Model Development and direct feedback providing inside the workflow. The more the models are used therefore the better they become. When it comes to ai driven Decision Support tools these are the models where they are directly advising on the treatment and diagnosis of patients they believe its an exciting opportunity for ai but an area that require significant can be deployed safety safely. Youre my final time let me add three examples of how we can bring ai into health care for the first is enhancing clinical documentation. We have a system in a partnership where ai can listen to a provider and the patient of emergencyroom in the er and then transcribe the event and using natural language processing turned into structured clinical documentations. Moving documentation to ai assistants mode rather than the doctor as a data analyst. The second is streamlining nurse and as handoff occurs for in a thousand times every week across their 183 hospitals but its a risky time during a transition of care. We have a large language model to read data in there for up to be able to interpret that data and create a handoff tool to the nurses after they review it themselves. This will bring standardization and safety to a highly variable time during care delivery. Finally we are using ai for staffing and scheduling. We taught an ai algorithm to understand the data surrounding power care teams are deployed in our hospital to care teams are most valuable resource in ensuring what we have the team at the right place at the right time with the assistance of ai algorithm to create more balance, fair and equitable schedules. In conclusion we are at dedicated to exploring and leveraging ai to enhance and improve Operational Efficiency and unfold the highest standards of privacy and security. We are committed to ongoing dialogue with congress and the subcommittee to help ensure the pathway forward to provide all the opportunities. Thank you for your Opening Statement. The chair now recognizes dr. Nguyen for five minutes for your Opening Statement. How is that . Can you hear me now . Thank you. Chairman guthrie Ranking Member eshoo chairwoman Morris Rodgers and Ranking Member pallone and members of the committee is my pleasure to appear before you today to discuss the ai is changing health care. My name is dr. Benjamin nguyen. I worked at the intersection of technology and care delivery throughout my career with a special focus on Artificial Intelligence. They make it easy for people to access highquality Affordable Health care and controls for Health Care Consumers in employersponsored health plans. Its not a Standard Health plan with their service to and meet the Health Care Journey for 4 million members an easy one. We help make their existing medical plan easier to understand and use. A platform is personalized for each member guiding them to care. We offer access to ondemand care teams and connect an ecosystem with the virtual care solutions. Or affiliated Virtual Clinic provides chat of videobased telemedicine visits for wide spectrum of primary care needs. In significant dan demand we use ai controls to improve the experience for members for reducing the administered a burden on clinicians. When a patient comes the Virtual Clinic in ai system begins to gather information from them about the reason for their visit so by the time the clinician sees the patient they have a detailed summary of the patients history but they can spend their time discussing the treatment decisions and followup care with the patient. Ai is helping these clinicians reduce the administered a burden and it frees them up to spend more time on the patients who need the most without replacing clinical but it want to paint a picture verheim nextgen ai can transform the way patients experienced health care. Imagine a single mother who uses english as a second language. It may be a 10 to 15 minute doctor vista but she may need 60 minutes to answer all they questions about her sons care. It makes this engagement challenging in a modern medical practice. Their patients in the waiting room and there arent enough partition or is. The thoughtfully built a system using extend technology can build partnership with physicians and supply information to her level of comfort or transferred to any nation she prefers. She can spend as much time she wants and needs too. A welldesigned system like this can help us move to onesizefitsall approach to many sizes approach. Years ago would have been immensely difficult and expensive to tailor a chat box to this womans need. This Technology Generative ai is it different nature than you may be familiar with but it can be applied to any domain but the most common application is in Large Language Models of power in the chat box for the chat box powered by this technology can converse with users grasps nuance topics and engage with the user. To use an analogy the technological leaps in ai that happen in five years to enable this are so great that they go from locomotives to powered flight and powered flights brings with it many opportunities. Because generative ai is not perfect and it has shortcomings even among ai companies and experts theres disagreement of best practices for capability safety risks of these new generative technologies much less how to mitigate them. Health cares has unique challenges and opportunities. We need to develop her own expertise and generative ai for this comes from the outside thes brings me to my last point which there is a significant growing gap in the health care. We need more doctors and nurses who are comfortable in speaking with ai and i know we are rightfully focused on teaching bedside medicine. Ensuring ai products are all americans equitably demands active participation from all levels of the Health Care System. We need the incentive frameworks and collective effort to create the opportunities to ensure that ai achieves the potential and changing the Health Care System for the better. Thank you for the opportunity to testify and id be happy to answer any questions you have. Thank you for your testimony. And mr. Shen you are recognized for five minutes free Opening Statement. Members of the subcommittee on the health of a nearly 17,000 employees in the United States and approximately 71,000 employees totally thank you for the opportunity to testify today, topic of Artificial Intelligence health care. My name is peter shen. My career focuses on the introduction of technologies in the Health Care Market including Artificial Intelligence. Its a leading Technology Company with more than 120s of history and break your innovation to market that enables Health Care Professionals to deliver the best care for their patient. Our corporate role includes imaging diagnostics and therapies augmented by Digital Technologies and Artificial Intelligence. We partner with more than 90 of the leading providers of health care address issues around population growth and disease management Health Care Workforce shortages and lack of access to care in underserved areas. But we have the distinction of being the only medical Technology Company capable of end and can secure from diagnosis to screening and survivorship or this is a responsibility we take very seriously as we keep patients at the center of everything we do. Human health has been working on applying technology for more than 20 years. Enter at all this in new jersey we new jersey we built one of most powerful supercomputing infrastructure dedicated to developing ai and health care but does allow the two collector. Organize correct and secure medical data and deliver accurate ai. We create and maintain a transparent Quality Assurance process which involves clinical validation guaranteeing the data being used between the ai is accurate for diagnosis and treating disease. This Training Data is based on a balance cohort of people of different ages and genders and ethnicities. The century we develop reliable accurate and unbiased a algorithms reflective of the Patient Population they will be applied towards. The patient journey is at the heart of human health ai work in ai helps improve outcomes for the patient at a ai helps patients undergoing a ct scan. Optimizing generating images while minimizing the time the patient is in the scanner. Radiologists can utilize their ai software as a companion to identify small nodules and other at rallies that they previously werent able to visualize without the assistance of ai. Suspicious lung nodules diagnosed to be ken chesebro declension could potentially be treated by Radiation Therapy which includes the task of manually drawing a unique contour to target radiation while observing healthy tissue. Our ai enabled software can automatically detect the contours of the cancerous area speeding up the patients time to treatment and eliminate unwarranted radiation. As siemens help commercial ai goes through approval process. They follow all ai enabled medical device regulatory requirements for freemarket review and close market surveillance to ensure safety and efficacy of our devices. We believe the rapid acceleration and innovation of ai medical devices the need for Regulatory Environment should be balance in innovation and adoption. While we believe the current rate of troy framework is sufficient to support innovation in ai we support the continuation of flexibility in the approval process as was the metric to facilitate global harmonization in the development of appropriate International Consensus standards. While cms is recognizing the value of the complex nature of ai the agencys reimbursement decisions have not uniformly and consistently ensured appropriate levels of payment for these ai products. To view and approach and limits access to patients benefiting from Ai Technology across their Health Care System especially in in underserved areas. We support a solution that ensures a predictable consistent approach to cms and approach that recognizes the cost of ai and reimbursement, ai analysis with separate Payment Systems so more data can be evaluated. See nisquallys ai has a Great Potential to include access to care diagnosed disease faster and make more precise treatment decisions. Rick said to see what the future holds for ai and health care. Again thank you for their pretended to testify before you today and i look forward to your questions. They think you for your test the money. Dr. Longhurst you are recognized for five minutes. Thank you members of the subcommittee for their pretended to speak with you today better experience in using ai models to improve health care delivery. Im a practicing pediatrician at the privilege of serving as the chief medical officer chief associate dean at uc health. We have been carefully evaluating and implementing models to enhance quality and safety for over five years and we believe our experience can be helpful in health care ai. As a leader at the intersection of care and Delivery Technology its disappointing to see so little progress of the past two decades. A recent study from boston suggest one of her patients admitted to the hospital continues to experience an adverse event many of which Health Care Organizations are complex systems using new ai tools will bend the Patient Safety curve in a better direction. One instructive example of her use of this technology is early in the pandemic. Serving as one of the first two sites for evacuation of wuhan expatriates we monopolize some of the first in the country in february 2020. This early experience led us to deploy an imaging which helped identify covid and chest xrays. This was months before wide strip widespread testing became available. Our evaluation showed clinical decisionmaking for one of five patients in Emergency Department over the course of the summer of 2020. After processing 60,000 returned it off at the end of 2020 because it was no longer useful to our clinicians when testing which demonstrates the importance of ongoing monitoring to ensure that ai tools continue to be safe and effective. The study was recently sided with her review of research about cove cold in ai which found their publication was one of just four of over 9000 which demonstrated impact on Clinical Health outcomes for this demonstrates another key point which is a huge gap between the christian algorithm and implementation at measuring patients. What we refer to as the ai cycle. The second example comes from her use of ai to support earlier identification and treatment of sepsis. Uc san diego has chosen to develop a local model using local data. We trained users as i dont know when confidence was low. The results have been associated with a decreased risk of death among patients with sepsis in our Emergency Department. The case study highlights importance of not only creating algorithms. Ensuring their transparent to generate trust and doing the hard work of integrating the workload where they can impact air quality. A final sample his recent use of generative ai to help the mission answer patient messages which is breached unprecedented levels in virtual care busy san diego authored a study showing generative ai is highquality quality. On these results it became one of of the person at the nation to implement generative ai to help our clinicians respond to patient messages. Importantly with full transparency with their patients by ensuring every message has a message automatically generated by the doctor for these messages cannot be without a clinician review and the results of cms has been well received by clinicians and patients and may save a cognitive burden. The study illustrates the importance of transparency in keeping humans in the loop at all times. The privacy and security perspective i want to know no information as leapt r. Hipaa protect the environment. We found cybersecurity with a focus on improving her digital protections and resiliency. This summarizes the Health System engages in the use of largescale Machine Learning models to perform a while at wide for a test. These benefits mitigate brummel five years or helped ai committee has been evaluating all Machine Learning models from an ethical and Health Equity framework to ensure safety and security and trust will find that the model proposed by the National Quarter for health i. T. To ensure safe use of ai. Awesome advocate for Testing Process our experiences local audits to be more effective for ensuring alignment with these principals in these models must be evaluated within the context of care. Finally with the generous support of the center for Health Innovations we see an opportunity to move this industry together with you and the frustration of responsible ai use metaphor to answer any questions you have. Thank you. We appreciate your Opening Statement in the chair recognizes dr. Newmantaker. Thank you could epshteyn to address congress on this critically important subject. Im a physician scientist with training Public Health and the Research Focus including the development and deployment of diagnostic technologies such as ai. Im currently a professor of virology and director of the center for diagnostic and a past president of the society for diagnostic medicine. My testimony will focus on challenges with a health care from a Public Health perspective with an emphasis on ai to medical diagnosis. Id like to say for the record the pin is express my written testimony are my own and do not necessarily the university. A is a branch of Computer Science concerned with and on computers with a billet to simulate Human Behavior. Most complex Cognitive Task and medicine would be diagnostic to cause a patients symptoms. Diagnosis accounts for an 800,000 deaths or permanent disability disability each and the west more than 80 with her clinical reasoning failures. This creates an opportunity for ai patients to save american lives. Potential benefits include better Health Outcomes and lower costs greater access to and efficiency of care delivery es pecially those underserved and disadvantage and decrease Health Care Costs workforce for now. Under these benefits would be realized without tackling data challenges facing ai. Implementing ai systems is no longer the technology. The sources of data which technologies are trained. They are three Critical Data problems. First using data that are wrong the garbage in out problem relying on the wrong kinds of data and third not having the right kinds of data at all especially Health Outcomes or inaccurate diagnoses unexpected adverse events for data Quality Health care is not uniform even for diagnosis. Most reliable complete dataset in radiology and Laboratory System diagnostic systems are being built. The least reliable or from routine clinical encounters. Patients at thems or terminations finding the Electronic Health record are often missing or inaccurate. There must be a radical shift in the way we capture diagnostic information. Faulty data will make the same mistakes that simply its available Electronic Health care data to train ai systems in the best we can help force a systems replicated with out vices. If systems are to play without e will drop. The biggest Public Health gains for welldesigned ai can present expect parts of the Health Care System for their large quality apps that will be closed for many individuals. Diagnostic errors lack of access to care in underserved areas and health disparities. For any and health care to that effect and help all americans the following are essential. First ai systems must be trained on Gold Standard datasets that are unbiased including complete information on clinical and carry outlets. It must integrate clinical workloads leveraging the strength of computers and humans together to reach better result being achieved. Wherever ai is used systems to monitor maintain and enhance Skills Including diagnostic once should continue to fact check each other. Ai systems must be held to a high rated her standard and demonstrate scientifically to improve hair kit care quality. New payment incense will be made to ensure ai systems are unbiased and health come outcomes or monitor. Targeted Research Funding to address these areas essential for special consideration should be given to funding programs to support the development of large Gold Standard datasets for its highquality ai systems are trained but in summary and ai has Great Potential to transform health care for the better. Absent carefully crafted legislation and target resources risk will dominate. The guiding principle for policy changes should be Public Health impacts including emphasis on the aqua distribution of and risks across the population. Thank you for this opportunity and id be pleased to answer any questions you may have. Thank you protest to me. That includes Opening Statements were witnesses. I think dr. Nguyen you guys have a lot of information we have a lot of curiosity so each of you will get five minutes. I know what would be hard to answer some of the questions briefly. If one of the say im sorry im moving to the next questioner we are being. We just want to make the best of our five minutes. And some of my colleagues on more than others will ask a detail questioned for five seconds left in their time. Ill let you answer as much as we can but im not being its getting things done so i can appreciate your time as well. I will say that and i will begin to the five minutes by recognizing myself for five minutes. You mentioned the various types of ai and how it genitive ai is the focus. Can you walk us through how you might deployed to various forms of ai supervised, unsupervised reinforced into these are the main drivers of how you deploy ai then please give us your approach. Absolutely. I think its helpful to think about ai is a highlevel into different categories. One is a will call it narrow ai narrow specialized ai tools and they typ. Event. That is a narrow type. You want to apply those types of narrow ai systems when you have highrisk, tasks that you must get right. Things such as supporting diagnoses and the patients deterioration. Those are what use the systems four. The other half is the systems are genitive ai systems. These systems are much more flexible. They are built for specific specialized tasks. We see them most often in technologies like chat box and chat cpt. They are built to be flexible and to do things like take language and output of language response. They are good at flexible task like assisting with an Administrative Burden of generating educational context. You need flexibility rather than a specialized nature. You may thank you. Mr. Shen for health care especially ensuring that patients their guardrails in place ai. We want to make sure we protect safety while promoting Better Outcomes for the taxpayer dollar. How can we strike a proper rated tory balance on the front into ensure their safeguards in place without stifling innovation and growth . Thank you for that question. Our algorithms got a rated tory process with the fda and we follow the Machine Learning enabled device regulatory requirements in premarket and review to make sure the solutions are safe for patients and effective for them as well. We also have regular dialogue with the fda regarding ai Machine Learning and provide feedback on the ways we can ensure safe and effective also we work closely with the ftm and temptation of the predetermined change control plans for ai and i think this committee for their support in this effort. Those helpless and sure we continue to innovate in this area will having the right regulatory components in place. At the same time one of the challenges right now is around the adoption of Artificial Intelligence and for providers and physicians to take advantage of the great benefits that we havent talked about here. Right now seem needs to create a consistent and predictable approach for the payment of ai solutions that are fda approved rather than the current ad hock approach we have seen to date for certain technologies based on cost. Other ones dont receive that payment. The uncertainty for providers is to make investment to Artificial Intelligence and enforcement because of that uncertainty patients get lost in terms of their ability to take advantage of the technology. Thank you and i want to move to dr. Newmantaker. Some of my concerns is the data that goes into ai and what we need to be aware for what could, but the rated tory policy challenge that we need to consider. Example would be to make sure we talked about quality adjusted life years and we want to make sure a data system factors that in to see if people are qualified for care. What are they on pristine challenges that you think we need to be aware of as an example . And their significant data challenges for ai systems particularly when we look at Clinical Data. For example laboratory and imaging datasets whether they are digitized and Clinical Data in electronic notes. They have lots of errors and problems and i do think to some extent thats a key focal point where we should be making sure that they earned over relying on faulty data sources as the try to move forward. I didnt leave you much time to answer so thank you for that and hopefully we will toward that more through this hearing. So i back i recognize the Ranking Member for five minutes for her questions. Thank you mr. Chairman of thank you do each one of you for being here today. As i was listening, i was listening hard to absorb what you were saying and i had to admit that there were different parts of your testimony where were really didnt understand what you are talking about. Now thats not to be interpreted as you being less than perfect. Thats a condition for the entirety of humanity. I think that we are very hungry to hear and pedestrian terms if youll excuse that terminology exactly how this is going to work and how you think its working now. I recall a book written by john torit and i think you all know who he is. Google him because hes one of the greats of our country but the title of the book was measure what matters and thats what im trying to extract from your testimony. I dont know if some of this is forward ministers and when you talk about Administrative Burdens, what does that mean . What is ai going to do about that and what does it mean in terms of patients at their bedside . I mean in real life not only is Important Information that medical specialists can have access to, to enlarge their understanding and as you said doctors dr. Longhurst is so wonderful you took all of your experience to the Childrens Hospital tuc. They are first cousins, stanford and university of california. So my questions are really more about the practical, the real practical advantages of ai. I would ask the entire panel if you get into this. Without any congressional statutes yet, how do you guarantee all the positives that you have presented to us today . Answer briefly. You are all saying this and i believe in the potential of this. I think we need to think long and hard about it. You are doing it right now with your systems. Let me see if i can briefly give you an impact winter. We said minister. We are anywhere between 25 and 50 of time during the day that a clinician doctor underspend some activities that dont directly relate to patient care. They are entering data into the system and searching for data in multiple different systems. They are putting the information together and riding it down and organizing it. They are communicating with other physicians they are bringing all of the information together. They are waiting it down, they are organizing it, they are communicating with pharmacy and other departments in the hospital, all of that so that they can have the right information and be able to make good information for their patients. That is a model that ai is almost custombuilt for. Are you losing using it today and measuring what matters . What are the outcomes . Are you reversing outcomes . What are they saying . For example, for the e. R. Doctors, they are seeing, upwards of 20 or 30 of their time returned to them so that they can spend more time with the patient and communicate with the patient, not having to do the documentation themselves. Anybody else . There are direct benefits and sort of indirect benefits. What we have been discussing are the indirect benefits of having additional time with, you know, less time spent on unnecessary tasks and i think, the future of ai, that we want to look towards is one where iai is helping improve health directly. For an example, improving the accuracy of the diagnoses and improving the application of treatments, avoiding adverse events from mistakes made in the delivery of healthcare. And those are the kinds of things that we are getting at. We want to improve patient health. Is it happening now . I would say that it is not happening at that level yet. But it is a place where we need to focus our attention. You all are going to get my specific other questions that i plan to ask. We are off script. So, thank you for your testimony today and the expertise that you are bringing to this. We need it and i hope that you would all weigh in in some way, shape, or form about the act. I would like to know where you all are. I think it is important for us to pass it. The chair recognizes chair rogers for five minutes for her questions. Dr. Michael schlosser, as you know, i believe that it is the first step congress should take, as we work through the guardrails that are needed in regards to Artificial Intelligence. You state the importance of using and developing ai in healthcare. Would you share any comments on this issue and the importance of privacy in Artificial Intelligence we privacy is critical to everything that we do with patient data. Even with Artificial Intelligence. We have been operating under the hipaa standard for decades and i think that has actually given us a great roadmap to understand how to do a really great job in protecting patient data. Ai strategy is a data strategy. The two are intrinsically linked. We need good, quality data, diverse data sources, large data sets to train and fine tune these models. We have to think about both sides of this. How do we keep the data private and secure . But also, we need to do it in a way that enables us to use the data to train these models to get smarter and better. If we want to achieve the outcomes that my colleagues and i have mentioned, the data is the fuel for that, so we completely agree that data has to be kept private and secure. We obviously would be happy to work with the committee and yourself on the approach to data privacy, you mentioned in the act. I would add, as a provider, we have a lot of deeper experience and how to do this and a lot of insight into how to do this well. Ai is just now another Software Application we have to put in the umbrella of hipaa so that we can continue to protect patient data. Recognizing that it is critical that the fda keep pace with how the technologies are being utilized in the benefit and risks involved. The fda must ensure that patience patients must have time. They need to provide industry with predictable, regulatory pathway and rules. Can you discuss how fdas current regulatory process works for aienabled medical technologies and are there any improvements in them . As it relates to the fda, there are several different pathways for ai solutions to get the regulatory approvals. These pathways include different regards that are available to approve, for organizations to approve, that they are ethical, safe, and secure, in terms of how they are treating patient data. And also how that application can be applied to the Patient Population going forward. In fact, the weight of the construct that the fda has today, it actually provides good ways for how software can be updated and how ai algorithms can be updated going forward. Where we see some of the challenges, as it relates to regulation is not in terms of the approval of fda solutions, but as mentioned earlier, the adoption of these fda solutions. Leveraging things like cms to be able to provide ways to encourage the adoption of ai solution amongst the different providers. In the time remaining, i want to ask each of you to speak to this question. Ai is being used in different fields to improve healthcare for patients. We have heard about diagnostics, better care for providers and as we move forward and continue to incorporate iai ai, it will be important for patients to be aware when decisions involve ai. Starting with dr. Michael schlosser , can you talk about what congress will be thinking about in this regard . Well, i would, i would common that say that transparency is common across all use cases. Patients and providers deserve to understand exactly when ai is being used and what data sets were being used to train it. I think that is foundational to an ai strategy. Dr. Nguyen . It is of the most paramount importance that patients always understand who is treating them. There must be transparency around the use of the ai tools. And providers must understand the usage of those tools. Mr. Peter shen . It is not just transparency in terms of how the ai is created, but also how the ai has derived a clinical decision. Being able to educate the users of the ai to understand how is the ai actually making this clinical decision or clinical recommendation. Thank you everyone. Unfortunately, my time has expired. So we will have to look for another opportunity. I yield back. The chair recognizes the Ranking Member for five minutes of questions. Thank you, mr. Chairman. My questions are initially for dr. Chris longhurst. I am concerned that the Rapid Deployment of ai tools means there is the need to share Health Information to train ai models. This raises serious privacy and Data Security concerns, particularly as it results to data transferred outside of the hipaa related or regulated environments. You share my concerns in this respect. In particular, the potential sale of data by third parties, including mobile applications. And that is officially regulated under any official privacy law, including hipaa. Thank you, Ranking Member, for the great question. As dr. Michael schlosser just described in his points, the commitment to transparency is key. Transparency requires privacy, Health Systems and peers who have been subject to hipaa for over two decades understand what that means. But these thirdparty apps are collecting information directly from consumers, and that is not subject to hipaa. That is deeply concerning to us, and as an industry, that there are growing databases, provided in many cases by the patients themselves, that will be inadvertently and unwittingly used for other purposes. In your testimony, you mentioned several examples of ai tools used at the university of california san diego. Do you know what data was used to train those tools . And was the data protected by hipaa . The first two examples i gave her about imaging and sepsis. Those tools were created through occupations that we serve. They were created in a hipaa redacted environment and information never let that environment. The third example is generative ai, using these tools to help respond to messages. This was a general tool that is not accessing our patients to protect Health Information. It is not being trained on patient data. It does exist in our hipaa protective environment. In all three cases, they are subject to hipaa regulations. These thirdparty consumer apps that collect data directly from patients are creating algorithms without that level of transparency or data protection. Let me go to dr. Peter shen, from siemens. You have testified about the vast amount of materials needed to train the ai tools, these are what siemens is creating. Is all of that data needed regulated under hipaa . So, at Siemens Healthineers, we are deeply committed to safeguarding data privacy, and upholding the protection standards set forth by hipaa. I think it is essential to recognize that the data is utilized to train these, it goes through a rigorous process. We actually use methods to remove all personal identifiable information. Or pii, or any protected Health Information as well, which is all done prior to doing any sort of ai algorithm training. It ensures that all of that Data Security and privacy is respected for the patient. You mentioned the importance of strong Data Security. Elaborate a little bit on why it is important, particularly in regards to consumer data. Another great followup question. What is very quite definite questionable, we want to make sure that we have a healthy data set, but at the same time, we have to recognize and respect the data privacy and the patient confidentiality that has been established between the patient and the provider themselves. We work with clinical partners to utilize data for algorithm training, we respect the legal constructs that are already in place with the different providers that are there. On top of that, what we do within Siemens Healthineers, we also make sure that we have double checked the data that we see the data is truly de identified. We go through that extra regular to make sure that it has been removed of any pii. I have only 30 seconds, but dr. Newmantoker, how do you think they can rectify potential biases, to prevent unintended Racial Disparities in healthcare algorithms, in 15 seconds . There are reasonable questions in terms of how to best address differences in human genealogy that may correspond to macroscopic groups. One thing is clear, we should not be converting racial biases into hard and fast determined rules. I think that is a critical feature and it will require that we adopt later data sets that are representative of all the population. That would include oversampling from minorities. The gentleman yields back. We were able to talk a little bit, but dr. Burgess is running for reelection. And there is nobody with more of an encyclopedic knowledge of healthcare policy. More important, just his ability to absorb the facts and his passion for making sure that the Healthcare System works the best and compassionate that it works for people who have the least ability to make it work for themselves. Someone who has become a dear friend of mine and who i have a tremendous amount of respect for. We have another year, but we are absolutely going to have an empty seat at the stable next year from somebody who is so good at what he does, so, dr. Burgess, you are recognized for five minutes. Thank you, chair, for those kind remarks. Probably kinder than i deserve. If you look at the context of the existing ecosystem, if i can use that word in healthcare, it is not new. It is not unique. We have all had some experience with it. And, like anything in healthcare, there is rarely a day that goes by. There was never a day that occurs where someone comes in to me and says, you know, i do not think that we are regulated enough in healthcare. So i want to balance those things. But we do need to be sensitive in finding a balance when we make sure that innovation is not clobbered in the process. Let me ask dr. Michael schlosser and mr. Peter shen both this question. If you can discuss the importance of clarifying the role of ai as a support tool, rather than a primary factor in decisionmaking, and extrapolate on what makes this distinction so significant in practice. Yeah, thank you for the question. I think this is incredibly important. The concept of humans in the loop, which i mentioned in my testimony, is a critical safeguard that will allow us to accelerate the usage of ai and learn more about the capability of these tools. In particular, these new versions, the Large Language Models, but do it in such a way that we still have a trusted physician or clinician between that ai model and the patient who ultimately is protected. The tool becomes an assistant that can provide summarized data, bring in insights, and rely on that physician or clinician to make the ultimate decisions for the patient. That gives us a level of safety that will allow us to continue to experiment and learn how to use ai, going into the future. Very good. Mr. Peter shen . To echo what dr. Michael schlosser said, we believe that Artificial Intelligence is here to be a companion for the clinician. We fully understand the value of the patientdr. Relationship. What we want the ai to do is provide more information, more context, for that clinician to make that informed diagnostic deletion or that more personalized decision. We are not looking for ai to replace what that clinician is doing from a diagnosis or therapeutic standpoint, but actually help inform that clinician to make a more informed decision. We have no time to get into it in the ceiling, but i do also hope for a response in writing about where the technology will learn to long term safety. We have to be concerned about the debt deficit. Look, peter shen, what is your experience . I know chairman guthrie asked you this a little bit ago, but you need to work with medicare services. They make reimbursement and coverage determinations. How has that been working out for you . We have been working very closely with cms to try to determine again, what is the appropriate reimbursement as it relates to Artificial Intelligence. Where we see the biggest concern again is around the adoption of these ai solutions. And what we are hearing from providers and physicians is that they have a strong desire to adopt these ai solutions because of all of the great benefits we have talked about your this morning. The challenge again is the uncertainty of whether they make an investment in ai, whether they will receive any reimbursement are not coming back for that investment. There is inconsistency coming in. This is a safe space. You can talk about cms all you want and we will not tell a soul. Let me just ask you, dr. Michael schlosser went into some detail in testimony about the Large Language Models. You talk about a generative model for the large language model, and i am wondering if it is possible to set down patient interaction, in the language of iambic pentameter, the language of shakespeare. That is what you referenced in your written testimony. Are you asking if it is possible to put down patient discussions in iambic pentameter . Well, using the language of shakespeare, lets focus on whether this can play a role in Drug Development. Absolutely, Drug Development is a complex process involving many moving parts. There are many ways in which ai some of those are predictable at the level of the fbi. As far as collecting the timeliness of the submission, it seems like ai would be a place for that to be organized. Exactly. If something was going to fail, maybe it could save you some time and trouble. Population data, and the administration of the data, we can work around that. It is very clearly things that ai could assist in reducing the burden on. Very informative panel. We are not done with this discussion. Thank you. The chairman recognizes the gentleman from california for five minutes of questions. Thank you for holding this timely hearing and thank you to our witnesses for sharing your expertise and your opinions. Emerging Ai Technologies show an incredible promise to improve and plug gaps in our existing healthcare ecosystem. Many of you have already mentioned that these novel technologies have the potential to expand healthcare access, address outstanding disparities, and support the healthcare workforce. I have been clear in my support for adaptive technologies that increase access and quality of care for all americans, but there is also the potential for harm. If we are not intentional about how to proceed forward, ai should make Healthcare Systems more equitable, not less equitable and ai is only as good as the data it trains on. I worry about the possibility that these technologies may perpetuate or even widen disparities. Ai invasion innovation needs to be developed carefully if we truly want to harness the full potential. I have a question for dr. Newmantoker. Dr. Newman toker, you mention in your testimony that those in rural or underserved communities or those with social determinants of health, or those with worse Health Outcomes, there may be adverse consequences of inadequately regulated Healthcare Systems, can you please respond on the consequences that we have only seen . Obviously we have populations at risk. These are individuals who do not necessarily have good access to healthcare in the first place. On the positive side, we hope that ai will offer the opportunity to deliver Higher Quality care with greater access to expertise. On the downside, there are concerns, obviously a, about whether systems will be accessible to those people at all. For example, there may be Broadband Access problems or other issues that may constrain their abilities, even though they are broadly available. We have further issues about the ability to use such tools and beyond that, the issue of whether when ai systems potentially fail or make errors , they may be less equipped to deal with those problems. What should Congress Keep in mind, as we look to ai to improve Health Equity and protect against worsening disparities . Go ahead. I appreciate your question. We have talked, on this panel, about transparency. We have talked about submitting to the fda the results of internal testing. It is important for this subcommittee to regularize that those are all self reported. We have recently proposed a series of National Labs that will serve as testing beds for vendorsupplied ai algorithms. I think that should be reviewed in more detail, because having an external testing system will mitigate these Health Equity kinds of issues that come up from algorithms that are based on biased data sets. One of the things i am concerned about, that we have seen in the past with incredible innovations that make it easier for people to make conclusions. For example, back in the day, when the term credit score did not even exist, i was told at the time, i said wait a minute, you were going to use this, not as the backdrop, but as the primary driver of who is going to get access to capital across america and now across the world. They said no, this is just a side tool. No, it is a main tool. The old days of having a review, before the make a decision, is gone. But my point is this. There are proprietary algorithms, the government of the United States does not even have a clue what those algorithms are. Nobody does but the actual proprietor and it is protected. I respect that protection, they have done a lot and invested a lot in that. And it is in fact proprietary, nobody knows what is happening in that black box, that is one of my concerns with ai, especially when it comes not only to quality of life but also who lives or dies, based on the algorithms decision as to what will be the outcome for the cure. Or if someone is even going to get access to healthcare, because they say no, yours is not an emergency situation. Therefore, you are not going to get care for the . 100 agree with you. We see the increasing rate of denials of payer claims, driven in part by ai processing these claims and very rapidly denying them in an inequitable way. We share your concerns. We already have concerns and examples. And i do not think that we in congress will do our job, if we tried to move as fast as ai does, in order to make a better future. I am sorry, i went over my time, mr. Chairman. I yield back. The chair recognizes the gentleman from ohio for five minutes. Thank you. Thank you for being witnesses, this is very important. Ai is something that i applaud our chairwoman and the full committee, through all of our subcommittees, that we have been having hearings on ai and how important this issue is, going forward. Especially when we are talking about, on the healthcare side, this is technology that we do have to put guardrails in place, making sure that we protect the privacy of americans. We need to prevent other countries, especially countries like china, from abusing it. And also, this had been brought up in your discussion today, again, we were trying to help the providers out there, being able to do what they are supposed to do. We have a shortage of Healthcare Providers. And the doctors, when i have asked that question, one panel especially said they have about as many at this table today. Actually talking about the patients. I dont think any of them said more than 50 . When you have a shortage of doctors, nurses, and everyone else, it is important that doing the job makes it essential to get it done. Now, dr. Newmantoker, in 2021, more than 100 drug or biologic applications, including ai and Machine Learning components, would you explain to the subcommittee, if we continue to explore ai, how this could lead to further breakthrough developments . Thank you, congressman, for the question. What i would say is, as we look toward the space of Drug Development, you can imagine, the same way as general healthcare, there are opportunities for ai to help, both in the process itself. That is, the mechanics of working your way through the regulatory process, as well as through the identification process, of actual treatments, for example, we have large data sets that allow us to identify drug therapies that are available to us. I think that will give us the opportunity to break new ground using existing Data Architectures. Thank you very much. Dr. Michael schlosser, i was reading your testimony. I thought it was kind of interesting, the technology that uses ambient speech models, to transcribe the doctor patient text in emergency rooms. You know, we all have the opportunity to visit hospitals. Especially being on the subcommittee, i go through a lot of emergency rooms. I am curious, because of the stress and everything there, how does the technology eliminate that stress to make sure that you are getting the appropriate, absolutely, 100 of what you have got to have transcribed for that patient . Yeah, well, thank you for the question. I will go back to the human in the loop. We always have the physician and their opportunity to review the note as the last step before it ever becomes part of the health record, so that they can ensure that it is 100 accurate, the ai is not yet at the point where it can do that on its own. We are using the data through data sharing agreements to continue to improve the quality and it continues to get better and better. It saves them more and more time to have to do less editing at the end of the event. Right now, we need those physicians to still be vigilant. As you identified, the emergency room is a chaotic environment. That is why we took the technology there in the first place. We feel like it has the most to offer in that space, where precious time is given back to the position, so they can really tune in to what is going on. That will yield quality results. Now, dr. Chris longhurst, you said something interesting, you said that we have to have the ethical review of ai. Can you delve into that ethical review of ai a little bit more please . The subcommittee has been around for a few years, led by several of our internal medicine positions and it includes Health Equity researchers and bioethicists. They are looking for what are the effects of being transparent , in this case we is that ethically appropriate . Or are there challenged questions . When we first amended generative ai, it helped clinicians respond to patient messages. Should we be transparent with our patients . That generated the conversation that resulted in full transparency to our patients about the fact that we were using generative ai, even though there was a human in the loop and those messages work edited by clinicians at the end of the day. Mr. Chairman, might time is expired and i yield back. We now recognize dr. Louise for five minutes. Our Healthcare System is strained by shortages, burnout, and barriers to care for patients. We must Work Together to create a Healthcare System that is sustainable, fair, and always puts the patient first. It is not only to only not only strengthen the workforce, but also ensure that we are addressing barriers to care that affect underserved communities. Communities that need the care the most, due to the disparities of health and the burdens of those disparities on communities. As the Witnesses Today have underscored, innovation in ai has the potential to improve patient care. However, these technologies also suppose potential risks that we must carefully consider and mitigate as the technology continues to develop. As an emergency medical physician i am familiar with the burden that physicians face and the negative impact it can have on patient care. Dr. Michael schlosser, how is ai currently helping physicians cut down on Administrative Burdens to focus on patient care . How does this affect the quality of patient care . Thank you, mr. Ruiz, i agree with your comments about the burnout and the Administrative Burden that our physicians are seeing. As i mentioned in my testimony, we focused on two areas, one is around documentation improvement, which takes up an inordinate amount of our clinicians time. If we can return that time to them so that they can serve on that focus on underserved. We can use just the clinicians we already have, by simply automating and removing some of those tasks. We are also looking at ai as an assistant to our nurses. Our nurses are under the same kind of pressure that our physicians are, so giving them tools that helps them do the job, like the nurse handoff tool. We can automate a piece of their workflow and make it easier for them to spend time bedside with the patients, that will lead to improved experiences for the patients. Transparency uses generative ai, now, what Safety Measures are in place to protect safety when using generative ai, and how safe . Absolutely, yes. Generative ai is still a prototype at transcarent. We use nongenerative ai at our clinic. Safety concerns are to be considered when you are building any ai applications. It really involves safety at every layer. So, when you are building new systems, you need to think about the data layer 1st, and ensure that you do not have a garbage in, garbage out system. You need redundant safety mechanisms. The use of narrow versus general ai systems, as mentioned earlier. You want to build reductive systems that can detect patients asking questions that may indicate they have an emergency. You want those systems in place, insuring, right . You will have a high degree of likelihood that you will catch issues like that. So you want those redundant layers in order to ensure the highest level of Patient Safety. Dr. Newmantoker, what would be the ramifications of ai biases in healthcare, if it is not addressed at this point in the Ai Development . What developments should be lamented implemented to detect bias in ai models used in healthcare . That is a great question, thank you, congressman. Obviously, we do not want to concretize the racial biases that we are seeing, and other demographic biases, that we are seeing today in the form of mathematical algorithms. To prevent that, we need to do work, both on the side of developing the ai tools, using appropriate data sets. We also need to deal, at the backend, with monitoring for these kinds of problems, using sophisticated tools to identify bias, as has been done in a number of recent studies. Furthermore, to monitor for outcomes in healthcare associated with ai, so that we can monitor for measures that matter for patients who are underserved. Thank you. I appreciate that. Thank you all for your insights into the benefits and potential risks. This insight is especially important to ensuring that this Innovative Technology is used to improve patient care, improve access to care, reduce healthcare disparities, reduce the barriers of underserved communities and improve equity, by giving resources and healthcare attention to locations that need them the most, while mitigating bias as a potential risk to patients. I yield back. I now recognize mr. Griffith for five minutes. Thank you very much, mr. Chair. Dr. Chris longhurst, i love this book, the perfect predator, written by one of your colleagues. It is about the saving of one husbands life, Thomas Patterson , who is also a uc san diego individual. By finding the right virus, to attack the antibiotic resistant bacteria that has attacked his body. It is a great story if anyone wants to read it. Are you all using ai to find some of those more therapy viruses . Do you see any potential in that . Great question, and thank you for referring to the book. It is an amazing story. It is something that was explored prior to the advent of antibiotics. Now we have gone back, with antibiotic resistance, to look for solutions. It has started a Movement Across the world. We do have a center at uc san diego, for trying to treat antibioticresistant issues. In general, i think that it will be augmented by the use of ai and i am very optimistic that we are going to see a large language modelgenerated hypothesis about treating new ways, that we may not have previously examined. That was a little off script and i will flip the coin over on the other side. According to reporting the use of Artificial Intelligence has referenced chatgpt and other chat bots, we have talked about how it can help. It can be used to revive old deadly pathogens or even create new ones. At mit, students were able to get Large Language Models to suggest four central pandemic pathogens within one hour, asking a series of questions to a large language model. In cambridge, massachusetts, there was open sourced Large Language Models, asked how to revive the spanish flu. Several participants found that reviving it would be feasible, for someone with basic skills. One participant got very close, quoting, to all of the steps needed to obtain the violence. Dr. Nguyen, i was encouraged to see an executive order put out to put out more oversight and security. But this causes great concern. Do you know of any Ai Technologies have security limits on what can be asks and what cannot be asked . Thank you, congressman. In a broader sense, even outside of healthcare, these Ai Technologies have extremely broad and wide capabilities, many of which we do not fully understand yet. While this is outside the realm of the kind of expanse used, ai, especially generative ai, in that field, there is the world of alignment research. That refers to the science of studying the malicious capabilities of these models. And in studying the ways in which we can defend against them. In that world, right, it is very important that the work is being done to find things such as what you are describing, right . Which is, vectors of malicious use. There is still a lot of work to be done in that world. And i think it is really important for us, in the healthcare world, to follow that, acknowledge it, so that we can do this important work. One of the articles i was reading indicated the proponents that said, maybe we need some kind of a test ban treaty, like we do with nuclear weapons. Do you think we should be looking at some kind of limitation on the testing . We want the positives, but the negatives could also be very consequential. That is a very good question. I think it is really difficult to truly limit the progress, you know, of testing the systems and making them safe, without using them, right . And trying to push them to the limits. I think it requires a measured approach. On the other end, the risk of test banning is that countries other than the United States will be testing, right . Dr. Chris longhurst, you mentioned that it was helpful for allowing doctors to provide more care to patients. The question is, do the patients trust the system and find benefit from those automated responses . I know we are expired on time. Anecdotally, we have very good response from patients and we are submitting more quantitative data. Thank you, i yield back. We now recognize the gentleman from michigan. We are now looking at it world with technologies like ai, having the ability to reshape the way we address Public Health and improve Patient Outcome. The way we keep talking about it, there have been many concerns, as some have expressed. Ai poses serious risks, left unchecked, it can harm patients and National Security. As this subcommittee has discussed, the covid19 pandemic has taught us a lot about the fragility of our supply chains and how vulnerable to disruption it is. It is really a National Security issue and we have learned that during covid. How can we keep everyone safe if we cannot access the equipment we need to protect the public . Earlier this month, i joined a public for a discussion on supply chains and what we think that the Public Healthcare supply chain will look like in the future. In addition to finding ways to reduce the overreliance on overseas manufacturers, we all agreed that we need to find ways to strengthen our existing supply chains and improve transparency and increased efficiency. Mr. Peter shen, i think ai has a role to play. How does ai support Healthcare Supply Chain management . I think for Artificial Intelligence, Artificial Intelligence has the ability to drive efficiencies, as it relates to the supply chain. Being able to leverage Computing Power to be able to support the technologies outside the u. S. , but being able to leverage Artificial Intelligence, to accelerate the ability to drive delivery and efficiency of supplies that are needed. We have seen this directly within the solutions that we have made here at Siemens Healthineers. We have leveraged Computing Power to be less reliant on certain components that may not be successful. We saw the success during the pandemic itself, by being able to deliver diagnostic equipment and Therapeutic Technology to patients and providers during that time. Thank you for that. There is a persisting challenge, and you know, over the summer, we saw very real shortages. We are still seeing a high amount, i dont know why i say this summer. We are seeing a shortage of certain cancer drugs. During the flu and rsv season last year, parents had difficulties finding common overthecounter pain relievers such as tylenol and advil for their kids. As well as amoxicillin, which is used to treat common infections. Now, peter shen, how are these ai strategies being used to address drug shortages . Thank you for the question. It is a challenge. That is what is exciting about Artificial Intelligence and the ability to drive efficiencies within the processes established. Within Siemens Healthineers, we leverage Artificial Intelligence not only to provide solutions to take care of patients, but also to improve the processes we have internally within our organization. It allows us to be able to deliver the diagnostic and therapeutic Imaging Solutions to providers, physicians, in a timely fashion. And we make sure that the patients receive the latest technology available, the medical technology to help them with the diagnosis. Thank you, but you need to keep working on it. We saw the need to ensure limited resources in certain communities that needed them the most. Dr. Newmantoker, how can a eyed help us decide how to best allocate our resources . Thank you for your question. Obviously, we have heard a little bit from mr. Peter shen about the architecture of dr. Michael schlosser allowing us to improve allocation resources. One of the Critical Issues is having the right Data Architectures, to be able to get to the point where dr. Michael schlosser can help us in those ways. Often, the key problem is that we do not have the right kinds of information about where the shortages exist and we do not know where the mismatch is between supply and demand. That is one of the critical pieces of the puzzle. If we want dr. Michael schlosser to work properly, we need data sets that can be created for realtime use. I am out of time. I have more questions, for the record. And i think the witnesses for i i recognize the gentleman from florida. Five minutes. I appreciate it. Mr. Peter shen, can you tell us about the role of generative red ai and what the potential is within the healthcare sector . I know it is a general question and i know that others have asked the question, but it is so very important, please. Certainly, my colleagues have talked about generative ai , maybe i will talk about how it relates to patients. Where we see the greatest potential is the ability for the ai to consume more information about the patient themselves. When the patient goes to get the exam done, to get the diagnosis, for example. Leveraging generative ai gives them the ability to know exactly what precise, what precise diagnosis we should be looking for. Not just doing a test for the sake of doing a test on that patient, but doing a test because we are seeking a particular diagnosis with what is happening there. If you think about it, that helps the patient not go through, or avoid going through multiple exams, looking for what the issue is, so that is potentially one area and the other area where generated ai has a benefit, is the interpretation of the images themselves, the ability to be able to take all of this complicated language and convey the diagnosis to laymans term so that the patient gets a better understanding of what is going on in the test results that they have had in that exam. That is great stuff. We appreciate it. Mr. Peter shen, i appreciate that you are talking about predetermined change and control plans. I was proud to lead the effort in the house last congress to authorize the use of that that got enacted last year, can you describe the pcc pathway and explained to us how that can allow for a more efficient regulatory work and why it is so important to make sure that the fda implements this bill effectively . Thank you very much for the pcc effort. The predetermined change control plan allows organizations, like Siemens Healthineers , to include in our initial fda product application, how the software be updated rapidly based on new data as it comes about. Without the need to resubmit back to the fda any new updates, every time the update happens. This really helps accelerate and go in conjunction with all of the Rapid Development around Artificial Intelligence. At pcc itself, we have the description of the methodology that we are using, so that we can provide that transparency that is needed. And what we have talked about today, as it relates to the technology here. And again, we are very pleased with your help, that we are able to move forward and make sure that we are able to make sure that it is part of the fda process going forward. One question for dr. Michael schlosser, can you elaborate on the potential that Large Language Models have in reducing the burden in hospital settings . Thank you for the question, congressman. As mentioned in my testimony, i think there are numerous opportunities where our current Healthcare System will have added tasks to nurses, pharmacists, and other providers that do not directly add value to the patient. Whether acting as data entry analysts or transferring information between different providers or different systems. Large language models are really good at those types of tasks, if we can train them to understand the data. That is part of the challenge. They can search for information, read complex medical charts, find information from multiple disparate sources, synthesize and understand it, and serve it up to the Healthcare Providers in their workflow. Because they are language models, you can interact with them in a natural language way. It is a really powerful tool to make the universe of Healthcare Information simple and easy to access. Very good. Good start. I yield back the rest of my time, mr. Chairman. We recognize ms. Kelly, from illinois. Five minutes. Thank you, chair. Thank you for holding todays critically important hearing, the integration of ai, it is a transformative solution to address longterm disparities and access issues. Many have promoted ai as a means to create a more accessible and equitable healthcare landscape, particularly in minority, underserved, and world communities. Dr. Newmantoker, i am hopeful about the ability of ai to improve Clinical Trial diversity by scanning multiple databases for clinical site placement in populations, with the hopes that diverse Patient Populations can be matched with critical Clinical Trials. What incentives or regulations need to be considered regarding the use of ai to improve Clinical Trial diversity . Thank you for the excellent question, congresswoman. Clearly, diversity in trials is an essential component of eliminating health disparities. We have seen a large number of treatments that we have studied over the course of time, they have only been studied in white man or in very restricted populations with minorities. And i do think the potential of ai to identify locations and places where patients can be recruited is a strong one. In terms of the Regulatory Framework, i do think that some of the existing architectures around critical trial requirements for diversity, i think, are important. I think that we have to further bolster that, as we get deeper into the ai space, in order to make sure that we are having overrepresented groups of minorities, so that we can do proper subgroup analyses within demographic groups. Additionally, this body has worked to decrease the length of time to apply for authorization to medicare population. I am supportive of the use of ai , i am concerned about multiple recent articles on the use of ai and in prior authorizations with high rates of claim denials. Again, dr. Newmantoker, the reliance on ai for critical medical decisions can reduce rigorous testing and validation, imperative to ensure the safety and efficacy of these technologies. Preventing errors, misinterpretations, things that can have severe consequences. In your review of these ai systems embedded in prior authorizations, can you explain why we are currently seeing such disappointing outcomes and what can we do to help mitigate these findings . I do think part of the problem here is, this is, there is a preexisting arms race in the space around claims, with insurers generally trying to find ways to reduce expenditures and deny more claims. And providers are trying to increase the revenue that is generated, associated with that. And now we are seeing it escalate into the ai space. I think, when we think more broadly about the issue of regulation here, what we have been talking a little bit about, ai, used in the context of healthcare, with patients, i think what you are alluding to is all of the ai that may exist out in the periphery, around the problem. That is a totally unregulated space. That is a potentially dangerous area. We have no idea even what systems are being used, for controlling the process of healthcare or access to healthcare, or even directly to patients. All of these things exist outside of a Regulatory Framework and i do think that we need to start bringing some of that into a Regulatory Framework. Dr. Chris longhurst, i need to give you a chance to comment. You are shaking your head. I think your questions are incredibly pertinent. We need to think about the diversity of Clinical Trial participants and ensure equity and how this is impacting the Healthcare System and the care of patients. But we also need to think about the workforce and ensure that we are creating diversity in the ai workforce. Previous comment suggested that we needed to train our medical students and other Young Professionals in these new technologies. I think that is absolutely correct. I think we are at risk of making these technologies available only to those who can afford them. I want to use this opportunity to express my support for the bipartisan proposed legislation, creating resources for every american, to experience the Artificial Intelligence act of 2023. It would establish the research at uc san diego and we strongly endorse this proposal, because it would provide the opportunity for academic researchers to develop better methods and knowledge about the systems. It is for Small Businesses, nonprofits, and other organizations. I thank you for that proposed legislation. I yield back in we now recognize metered mr. Johnson for five minutes. Thank you to our panelists for being here. Ai is creating quite a buzz around capitol hill, that is to say the least. Even across the country. It has been met with both excitement and concern. Whether the American People realize it or not, ai is already in many sectors, particularly in the healthcare space. I have always been an advocate for the development of new technology and ai is no different in this regard. I worked with ai when ai literally was just a buzzword, back in, back in the early 80s , when i was in graduate school at georgia tech. I am very familiar with the technology. But simply put, ai is a tool that our medical professionals and scientists can use to not only for the Development Care and therapeutics, resulting in Better Outcomes for patients, but hopefully lower costs to families and the taxpayer. Unlike the vast majority of congress, as i mentioned, i actually have a tech background. In my time in the military, as well as the time spent in the private sector, i worked in information technology. I understand the benefits and challenges of ai. It has been around for decades. Take Electronic Health records, for example. As Congress Continues to incentivize those options, rightfully so, we are settling Healthcare Systems with an immense amount of data from patient notes, imaging, doctors and nurses are expected to utilize all of this information to best treat their patients. That is a lot easier said than done. It is a lot of information. This is a perfect example of how ai can be utilized in reading and deciphering all of this data. They can make these records more digestible and ultimately increase outcomes for everyone. Not to mention, i assume, the Administrative Burden for physicians, nurses, and Healthcare Systems nationwide. Generative ai has gotten a lot of attention, as a result of chatgpt and other publicfacing technologies that have been widely used over the last year. However, generative ai has been used in healthcare for years , through Patient Engagement technologies and clinical support models. So, my first question. Dr. Michael schlosser, do i have that right . What are some of the other promising ways that you can see generative ai integrating into the Healthcare System . Thank you for the question. There are numerous opportunities. I think the one that you highlighted around making the vast universe of data that clinicians and physicians have to access more easy to access is one that we are incredibly excited about. We are working right now with one of our partners about what is essentially an ai system in your pocket so that we can interact with the entire health exchange, through a natural language interface. That will allow you to search for and look for information that otherwise would take a long time and is very burdensome. If you have ever seen a ccd, the output of the hie, this general list of data, the way that information is provided. It is incredibly difficult. These models are capable of doing more than just reading and understanding information. We have taught a model to look at staffing schedules across an entire hospital and we are able to ask it questions like, how do we better balance friday nights . We can deploy our labor workforce in a much more efficient way by harvesting intelligence contained within these large models, to solve complex problems that previously were put on Nurse Leaders or others, that just struggled to have the information and delivered the outcomes we are looking for. What about rural and underserved populations, what can congress do to facilitate the option for these technologies across smaller and rural practices to make the Healthcare System more personalized and ensure that every patient provider has access to the highest Quality Healthcare technology . Another great question. We learned that there is basically a onetoone relationship between having enough Healthcare Providers and the patients that you need to take care of. Ultimately, the providers deliver care. The way we see ai solve this problem is by little it really literally bring them up, so that we can, in a sense, increase the size of our Healthcare Resources through ai. One you can answer this for the record if you would get back to me because my time has expired. How can we facilitate, what can congress do to facilitate more private investment into these technologies to make them more useful to the Healthcare System . If you would get back to me i would appreciate it. I yield back. I recognize the gentleman from washington, dr. Schrier, for five minutes. Thank you over witnesses. Artificial intelligence and Machine Learning are transforming how we study and practice medicine, as we continue to grow the capabilities and make further breakthroughs its important that Congress Keeps up and i think you put this education, last spring, dr. Schrier, i loved visiting the ultrasound and research and Development Headquarters located in my hometown of issaquah, washington. During my visit i learned about and got to see pretty incredible innovations being done. One of them was the ability to diagnose nonalcoholic Fatty Liver Disease in an ultrasound scan that took less than one minute and to be able to catch it early. The implications for morbidity, mortality, are incredible. But, every time we have a new advance , there is this question of cost. As we integrate Artificial Intelligence and these more advanced algorithms and technology, there impacts on costs, there is development impacts, but there is also potential cost savings down the line if you are avoiding liver transplants. Two questions. Just partition your time. If you could comment a little bit on Ai Development and cost. Thank you for the question, of course we were happy to host you at our issaquah facility. As it relates to Artificial Intelligence, as you correctly noted, we are trying to integrate ai tools directly into the types of exams or devices that are touching patients. So here in this case for an ultrasound, being able to integrate the ai and not have it be a separate type of solution. Doing that in itself reduces some of the costs. Rather than trying to have a separate ai solution that has to be maintained or procured or whatnot, we integrate the solutions directly into the medical devices treating the patient. That is one aspect. As we look at ai overall, we do want to look at not just the cost of procuring the ai but what is the downstream cost . What is the benefit, not just to the patient in terms of maybe fewer dates they have to spend at the hospital, or maybe a shorter time to diagnosis or treatment, but also cost savings that could be realized by the provider themselves as well. So the providerdeploying this type of technology is able to be more efficient. Is able to make the diagnosis faster and see more patients because they now have more time to be able to take care of one patient and move quickly to the next patient. These are things we think about as we develop ai alternatives. Thank you. I appreciate that. My next question i want to give it to dr. Longhurst. This is really about the impact of ai on the physician patient experience. Id like to talk about the doctor experience. I can understand how nice it would be to have the latest research pop up as a suggested pathway for a given patient who i am seeing whos maybe already filled out their whole history for me. But, doctors are already burnt out. We have been compared in an op ed to cogs in a wheel, line workers after almost a decade of training postuniversity. And we are being asked to see more patients faster, do more things in a visit and we are burning out. I want to talk with you about, the physician and patient relationship, the trust that is there, how physicians feel when perhaps they are becoming a check on the system where ai makes patient management decisions over them after that kind of training. Can you speak about that . Thank you. Its a privilege to speak with the fellow pediatric graduate from stanford. When i was at Childrens Hospital we were in the process of Electronic Health records. As we know if not the primary cause, has become a primary symptom of burnout. The many hours that pediatricians, physicians in general spend documenting is contributing a National Academic of pajama time or afterhours work. For every day spent in clinic the average physician spends about two hours documenting Electronic Health records to ensure Regulatory Compliance billing and health things. Where it was really important digital structure for collecting data and quality in population purposes, it has introduced these unintended consequences. Im incredibly optimistic about ai, particularly the ai scribes as dr. Michael schlosser described to be a solution that was decrease the burden it was introduced by Electronic Health records. We are seeing incredibly positive results from pilots using these. Unfortunately technologies are still expensive but as they become commoditized and continues to demonstrate outcomes in privacy i think this is going to help us remediate some of the burnout that is happened over the last decade. I yield. I recognize myself for five minutes. Well im only just beginning to learn about how ai can contribute to healthcare i recognize it has Great Potential. I was a cardiothoracic surgeon before was in congress. I do believe ultimate Technology Used properly will help us control cost. I really believe that. As an example, a top priority of mine is legislation that will allow realtime prior authorization decisions by Medicare Advantage plans and ultimately by all healthcare plans. I recognize ai would make real time decisionmaking more feasible and expect it would be used for this purpose. At the same time we are hearing allegations the health plans are making coverage determinations using ai power tools that are ultimately shown to have high rates of errors as was mentioned by the Ranking Member in her Opening Statement. Resulting in patients paying more for healthcare, perhaps medical interventions, basically these are improperly denied claims. Recent media articles have outlined the situation which is unacceptable and i would argue that this should be investigated by congress. I say all of this to remind my colleagues to remind the Companies Using ai to do so responsibly. This statement, dr. Schlosser, as long as ai can properly populate the record to obtain appropriate reimbursement for the providers involved in the case, you will see wide acceptance of it. Even if their record is there, but it doesnt properly reimburse the provider for their care that it will be a struggle. Am assuming it will probably do that. That is a major issue for providers. The documentation required by the federal government for reimbursement, and honestly in my view has been a problem for a long time. Another issue in our Healthcare System, we will need to think about, is using more ai in how we train and educate to professionals. To use it appropriately. Im going to address this testimony there may be a risk that our future providers will become overly dependent on technology resulting in less welltrained providers in the art of clinical decisionmaking. I will give you an example. Google maps, not direct, have Adult Children in their 20s, they cant navigate anywhere without google maps. They literally dont know what direction they are going. To go around the block, they map it. I would call that an overreliance on technology. Its not a direct correlation, but kind of. So how do we begin to train these professionals on the uses of ai and increase awareness of the pros and cons and is that something that middle school best medical schools are beginning to think about . And if not, should we be . Thats a very efficient example. I too have a lot of trouble navigating without google maps. I understand that. I think our institutions, very rightfully so, focus on the art and science of medicine. At the same time, i think it is very important that institutions leverage these new Ai Technologies to create learning experiences, first to enhance the learning experience and make them more efficient, enabling students to hone in on the most important concepts they need to know in an efficient amount of time. Second, i think it is very important that these institutions train and educate their students on the nuances of these technologies. It will be unavoidable the doctors of the future use these technologies. Whether or not they are trained in them. And so, the most important way to prevent over lines is to educate them on the limitations of the technology. I would agree with that. Dont get me wrong, and a big supporter of technology in healthcare space. Maybe you can talk about real time. Are we seeing medical professions realtime over relying on ai as it relates to the evaluation of, for example, ct scans, mris, xrays . Are the people coming up in properly trained, i would say, i guess, on the positives and negatives of this situation . I mean, thats going to be really important, right . Thats a great question. To echo what he was talking about, whats critical is the transparency around the Artificial Intelligence. Not necessarily how you use ai but how is the ai making the clinical determination. And educating upcoming physicians on how the ai is making that clinical discrimination. Thank you both for the answer to that question. Its really important. And now i recognize ms. Kuster from new hampshire. Five minutes. You so much, mr. Chairman, and thank you to all of you for sticking with us. Todays hearing is an opportunity to understand how Artificial Intelligence can help patients, providers and researchers to fully realize this potential we need to ensure that ai tools are safe and equitable. I want to use this hearing to discuss one opportunity and two concerns i have with my health. For the opportunity i will look to mr. Newmantaker, you describe one potential benefit to ai is it can improve patient diagnoses, could you give us some examples of how ai could benefit Public Health . Yes, so, thank you for the excellent question, as i noted in my testimony, we have recently estimated about 800,000 americans die or are prettily disabled each year from diagnostic error with serious medical illnesses like stroke, heart attack, pneumonia, sepsis, et cetera. There is an enormous potential Health Impact of being able to close that gap. That quality gap, with ai based detection of using Laboratory Data and vital signs for things like sepsis, using videobased interpretation of eye movements for stroke diagnosis, some of the work we have been doing. I think there is to mend his potential in that space. At the same time to deal with some of the concerns raised earlier about cost. When you realign, when you actually improve diagnosis we do is cut down on false positives and false negatives at the same time. By doing that you save lives by catching the cases you missed, and you cut costs by not over investigating the patients that didnt need that investigation. I think its a tremendous Public Health opportunity. Good, thank you. Two concerns. I am worried about bias in the data. Continue with mr. Newmantaker, you stated in your testimony for ai tools to be maximally beneficial they must be properly validated and utilized Gold Standard data sets. What steps can companies and researchers take to ensure that the data that is being used to train ai systems is accurate and without bias . Thank you for the wonderful question. I do believe this is the foundational challenge that faces this whole area of ai in healthcare. The issue of creating Gold Standard data sets, there is not a simple solution to the problem. We actually have to do things in healthcare we dont normally do such as for example, determine what actually happens to her patients downstream after an encounter. So we say, for example, a patient leaves our care and has x diagnosis, but we dont know if thats true. We often dont get the follow. They may go somewhere else or end up in a different Health System. We have to start courting Data Architectures and developing and curating good data sets that can be used at a large scale to train these ai models. I think thats going to take a big effort and one best correlated federally. Helpful. Thank you. In my final concern is about protecting patient data when we are developing ai tools. Mr. Shen, i appreciate Siemens Health and commitment to protecting patient data unfortunately weve seen an increase in healthcare Cyber Attacks which have more than doubled from 2016 to 2021. What steps does your team take to ensure patient data being used to train ai tools is protected from cyber criminals and just plain bad actors . Very timely question and i really appreciate that. Here we take data privacy and patient data privacy as a core component to how we approach the development of Artificial Intelligence. To that respect, as it relates to securing the data, one of the important aspects we do is that any of the data we utilize to train our ai algorithms is fully protected in our big ben office there in princeton, new jersey. Their physical limitations in place, physical barriers that dont allow individuals or bad actors to gain access to that data center. From a cyber standpoint, what are big ben office does is they actually control who has access to the data itself. In terms of controlling internally the audit that is needed in terms of who are the users that can access the Clinical Data to do the algorithm training. So they have the ability to audit the user extras and restrict the user access to the individuals need to be accessing that data. Great, thank you. I yield back. The chair recognizes dr. Dunn for five minutes for questions. Thank you very much. I appreciate the insights from witnesses regarding the role of ai in the clinical setting. I believe there is an important debate to be had about the value add versus the risks of ai in the Doctors Office and hospital. And i agree with her witnesses about the promises of this technology in medicine. I also echo mr. Nguyens caution and consideration that ai used for clinical Decision Making without close physician oversight. It is clear from the advances in ai from narrow ai to the larger generative language models their sweeping indications for the delivery of healthcare and presents clear opportunities and challenges. Im encouraged by the efforts to explore the role of ais role in interpreting radiology and pathology. Although i think the Current Evidence demonstrates this is not ready for prime time i am certain that will be more sophisticated over time. Im especially optimistic about the ability of ai platforms to reduce administered of burdens and simplify clerical tasks. I appreciate the questions that dr. Schrier asked. That is a real problem, as we all address burnout. Physicians are spending a quarter of the time or more on administrative tasks. I would love to have had that when i was practicing. But honestly. I do have some concerns, private practices may struggle with the upfront cost of adopting Ai Technology. I urge the industry to think creatively about ways to provide access to that technology to the full spectrum of provider settings. And to mr. Johnsons concerns with her oral entities and providers would be further disadvantaged if they dont have access to that. If those technologies are only accessible to those with the resources, we may have an even worse situation in rural medicine. Can you briefly comment on the specific challenges that rural private practices may face when trying to adopt Something Like transparent . So, to address your question, i think there always challenges with rural private practices with their simply smaller in size staff and budget. The challenges come in many way shapes and forms for any technology adoption. Ai included, that is the capacity to assess the right tools to adopt and the budget to adopt them. I think it is very, very important to continue to support the development of ai tools amongst the private industry in a safe and nonbiased way because that is actually, in my opinion, the fastest way that ultimately a lot of these practices will experience and be able to use these generative ai tools is when the vendors and the tools that they use begin to incorporate those, when we began adopting ehr most of the private practices had ehrs once it began to be built for the practices. I like to do offline, have someone from your company come and tell me what your pricing mechanism for different practices and how they can adopt that in their office. I and that is something we could do simply. Dr. Schlosser, briefly what are some of the ways you have seen ai improve efficiencies for patients . Improve their experience . One thing all the clinicians and probably everyone in the room will acknowledge his we ask patients the same questions over and over again. We are constantly warming them with delivering their entire Health History at each interaction along a healthcare journey. And then if you change systems are go to a different physician you start the process over. I think the ability of ai to help us wrangle this entire universe of healthcare data that exists across the table disparate ehrs into a longitudinal record of the patient that clinicians and patients can easily access i spent more time as a patient than as a doctor, and i have filled out my history probably 1000 times. It is an amazing experience. Dr. Longhurst, in your testimony you made reference to a study i believe you are a co author, the algorithm is rapley deployed to analyze chest x rays in covid19 patients. Clinician i has been reading chest xrays for 40 years. What specific advantages did you confer to these physicians . Great question. Thank you. The day we rolled this algorithm out, i walk through the Emergency Department and if are attending physicians have used it in one said yes we got a chest xray on this woman whos in for cardiac symptoms. We didnt see a sign of the ammonia, the radiologist didnt call it pneumonia but the ai showed some color and because of that we ordered a test. What did the test show . The answer is it takes 24 hours to come back. It turns out the test was positive, the patient was diagnosed with covid before symptoms. The patient was proactively hospitalized, did not need Critical Care and went home safely. That was a great example of a ai finding a signal we would not have found as a human. That is the kind of technology if used properly. Thank you. I yield back. The chairman now recognizes ms. Trahan. Thank you for holding this hearing. And to all the witnesses. Ai in healthcare has the potential to transform various aspects of the industry by offering new solutions, improving efficiency, and enhancing Patient Outcomes. However congress has the responsibility to make sure we establish appropriate guardrails around ai in healthcare in a way that works best for consumers and maintains patient and provider trust. According to a new survey of more than 20,000 consumers across the globe, consumers are more hesitant about using ai to get advice about medical problems than they are for other uses like billing and customer service. Has the use of ai in medicine as it becomes more commonplace patients have raised logical comments around privacy, transparency, ethical considerations, human oversight, errors and misdiagnoses and access issues. With this in mind i welcome the opportunity to discuss some of those issues today. Many colleagues have brought up valid, ethical considerations around ai including algorithms, potential disk donations and impact invertible puppy nations. As ai advances into healthcare and plays a role in making medical decisions, im curious if there are differences among various different patient demographics in their willingness to consent to ai decisionmaking and whether those preferences may unintentionally skew algorithms. Mr. Newmantaker, how important it is to understand if there are patterns of patients who would or would not consent to use of ai in healthcare based on race, education level, geographic area, et cetera. Thank you, that is a fabulous question. I dont have any specific data about the demographic variability and trust with respect to i specifically, but we have seen over and over the trust issues are equitably distribute. For example in baltimore there is a strong strain of lack of trust of the Healthcare System in the black community. This is a major problem for getting equitably distributed data from patients. So i do believe you have brought up a critical concern the trust gaps are a major issue and they may not be evenly distributed. Thank you. Im going to switch gears. But we will probe that further as we progress. The rapid evolution of ai in healthcare has exposed the need for federal coverage and payment policies that promote innovation and protect patients. Will the fda has moved forward to regulate software, as a medical device, cns has yet to establish coverage and payment for these technologies. Mr. Shen, are federal agencies like the fda and cms position to keep up with the rapid increase of technology and if not, what additional capability resources to those agencies need . Thank you for the question, congresswoman. I think as you correctly pointed out, we work closely with the fda and cms to try to bring forth these new and emerging technologies and make sure they get into the hands of providers and the patients themselves. We we are seeing a challenge here is unfortunately around cms and the reimbursement associated with Artificial Intelligence. Today, unfortunately, there is inconsistency in terms of how this technology is being reimbursed. That inconsistency and uncertainty translates to providers being unsure of whether they should make the investment in Artificial Intelligence. Not knowing whether they will actually get reimbursed or not. We see this is actually inhibiting and creating a bit of an adoption problem. And preventing patients from ultimately benefiting from this technology. Would love to see opportunities where working with this committee here, try to figure out a better way to work with cms to maybe establish some sort of payment that allows the different providers to move forward with investing in Artificial Intelligence. And helping everybody understand what the true value of this technology is. Thank you. I couldnt agree more. And will there is warranted skepticism around the use of ai in healthcare, we are all excited for increased applications of ai and how they will positively impact Patient Outcomes. Dr. Longhurst, how are we already seeing ai used to enhance progress with diseases with no known cure like alzheimers and ms . I looked up, where did he go . He had to step away. Is a neurologist on the panel i will take that one. I think there is tremendous potential for ai to do Early Detection of disease. Chronic disease in particular, such as alzheimers disease. You can imagine if we can make diagnoses 10 years in advance through information coming out of wearables or eyemovement analysis, we will then be able to apply early preventative therapy. I think theres a lot of potential there. Thank you, mr. Newmantaker. I yield back. The chair recognizes the gentleman from georgia, mr. Carter. Thank you for being here. This is obviously a very hot subject on capitol hill. Artificial intelligence. Particularly in the healthcare world. We are very concerned about it. Look, im a big believer in telehealth, i represent rural area and i am seeing how it has benefited us in the rural areas. As you know, all of you know that we have a doctor shortage here in america, particularly in our rural areas. Telehealth has been a great savior for us. I have always said there is a big difference between knowing something and realizing something. And during the pandemic, i think we realized just how important telehealth can be. I think it was an article in the paper, in the New York Times that said the telehealth had advanced more in one day than i had in the last 10 years and it probably has. So i want to kind of focus on telehealth. Mr. Nguyen, can you talk about how its using it in your telehealth solution and allowing doctors to be more efficient with their time so they can see more patients . Certainly. Thank you for the question. I think it is been repeated on this panel, the common phrase youll here is a very important benefit of ai is enabling the doctors and nurses to do the doctoring and nursing. What transparent does is, we really believe in freeing up the time of the doctor to spend with the patient and reducing time required for Administrative Burden. The way we use ai is, in our clinic, when a patient comes to the clinic, a ai assistant gathers information and synthesizes the information for the doctors. That enables the doctors to come into the visit and see all the information organized and spend time on that diagnosis and treatment and really create the plan with the patient. That frees up time and capacity to see more patients, including patients in rural areas since we serve 4. 4 million americans across the u. S. Mr. Shen, let me ask you, ive heard that sometimes there is bias in ai and that i can actually be good. That is somewhat baffling to me. But nevertheless, can you explain to me how that might be and how bias can sometimes help improve the utility of ai in healthcare . Thats an interesting question, congressman. I think whats important here that will emphasize is as we train the ai algorithms, the have to be trained with data that is respective of the Patient Population theyre going to be serving. It is important we work with our different clinical collaborators to find the right type of patient data to train these ai algorithms. Again that are applied toward the Patient Population and making sure that Patient Population is reflected in the data that is training the algorithms themselves. Okay, i have one last question, its kind of for all of you or any of you if you will, that is, we have a Doctors Office hearing congress, and im a healthcare professional pharmacist by profession. I have served in the state legislature on healthcare and one of the things i noticed is that a lot of our healthcare costs have increased because of defensive medicine. Doctors running unnecessary lab tests really to protect themselves from litigious patients or situations. But how is that going to impact the practice of medicine if a physician doesnt use ai, and then something happens and then all of a sudden they are sued because you didnt use something available that you shouldve used . It seems to me like this could potentially increase healthcare costs as well. I see the savings, yes, but i am also seeing, and try to deal with on a state level and on a federal level. And i will open it up, whoever wants the comment. I will take a stab. I think we need to remind ourselves that healthcare decisions are made by physicians and practitioners. They should be the ultimate decider when it comes to coverage, when it comes to you need to be admitted to the hospital, what treatment do you need . These need to be made by trained healthcare positions. You are trained healthcare professional, not a lawyer. Ive got to feel like from a lawyers perspective, they are going to take a different approach. Thats why think its important we understand that as a community, as an industry, the we are not turning over decisionmaking. These are tools. Tools in the tool belt and we need to review them as such. Not as an authoritative decision that someone should be held accountable to. Anyone else . Quickly . I will just say if we can prove ai systems save lives, and people should be using them. And if we cant, we should be relying on clinician judgment. I dont know that we will ever get away from relying on clinician judgment. I agree with you, i think it is unlikely, certainly in my lifetime. Good. Thank you. And i yield back, mr. Chairman. The chair recognizes the gentleman from indiana, mr. Pence. Thank you for being here today. Incorporating Ai Technologies into Healthcare Systems may improve and streamline diagnosis and Treatment Options in addition to using the Administration Burden healthcare facilities. Personal and medical data and information however is typically the foundation of ai delivery in healthcare. The trust and safeguarding of personal information between patients and their providers is critical for people receiving the highest quality of care. In the ecosystem of electronic apps and wearables there are areas where healthcare data is not clearly protected. I had a hospital in my district in Hancock County that was on 60 minutes a number of years ago. That is why this committee needs to consider a federal data policy lot to set the foundation of protections and how such that is collected, used and shared. We should do that before we can look at regulating ai in healthcare and find the balance in simultaneously encouraging private innovation. Our increasingly Digital World leaves americans in the dark about who has access to their information. It is alarming to me how little consumers and patients know about how many personal details of their lives are collected, shared with third parties and monetize without their informed consent. Monetized with no recompense to the provider of the information. Patient trust and those is paramount. Dr. Schlosser, as we introduce new Ai Technologies in healthcare, patients deserve to have control over when their information is collected, who has access to their data, the right to remove their data, and where the data might be shared. Heres the question. Should Healthcare Organizations that collect protected medical information be transparent with patients on how their data is stored, who has access to their data, and, for todays hearing, identifying that ai is part of the process . Yes. So, i would agree with everything you just said. I think we fully support the idea that we should be transparent with our patients and we currently are through a rigorous consent process. As to how the data is being used, how its being protected and how it may be stored and shared. I think is the use of ai expands, that will be increasingly important so patients can know where the data is going and how it might be used. I will just add that ai is entirely dependent on the data. If we want the benefits of ai we also have to do this in a way that enables us to use the data to train and finetune algorithms. There is a really important balance we need to strike ensuring we are transparent and keep the patient data private but we dont create too many barriers to actually using that data train algorithms to achieve alone welcomes we have been talking about. Yeah, it comes back to a previous life of mine, garbage in, garbage out. The wrong algorithm or the wrong collection point of the data can skew the outcome in a big way. In finance we say you pay off the National Debt with wrong numbers. Would anyone else like to answer that . Yes, sir. Yeah, i appreciate your question very much. I think as dr. Schlosser has said, your ai strategy is your data strategy. I would point out that in the ecosystem of treatment, payment and operations, all of us, Health Systems, providers, insurance companies, are provided by hippa laws around data privacy. Where i think the greater risk lies is when these Consumer Health apps and others that are accessing data either directly from patients, from Health Systems with patient consent, the 21st century cares act, and other mechanisms. I think you are right, theres a lot of healthcare data floating around not subject to hipaa because of these mechanisms. I think it is a risk and something that should be looked at legislatively. Your concern is teaching . Is that what you are referring to . I think there are number of risk of those benefits being used to generate algorithms without transparency to target appetizing other types of uses to patients without their awareness. I go to the doctor and googled all the answers. Thank you very much. Chair, i yield back. Mr. Joyce is recognized for five minutes. Thank you for holding todays hearing. To the witnesses, being with us here today, we appreciate your time and testimony. Artificial intelligence has made a Significant Impact on our day to day lives in the benefits that industries and individual drives through its use are numerous. As this technology continues to explode onto the scene, it has been especially prevalent in healthcare. But, like Many Industries where ai is seeing a dramatic increase in usage, there are and will be certain risks associated with it that we must contend with as policymakers. While that should not demean the potential efficacy of its day today uses, applications and functions, ai remains a tool. A tool that utilizes vast amounts of data and with its integration into healthcare space, we must be vigilant to ensure that sensitive patient information is safe, secure, and is protected. As we move forward, Congress Must have that unique task of analyzing and further understanding ais evolution and applicability when it comes to healthcare. Will president bidens executive order on Artificial Intelligence might layout the administrative policy initiatives, it is still the responsibility of congress to legislate. It is paramount the congress has a firm grasp and a clear comprehension on how ai interacts with existing regulations so we can ensure ai first does no harm. But instead, continues to positively reshape the healthcare industry. Dr. Nguyen, patients that live in rural areas like the district i represented pennsylvania often face barriers that impede their access to healthcare. Do you believe ai has the potential to squash those impediments and if so how can we incentivize further adoption of Ai Technologies . Thank you for the question, congressman. Absolutely. The distribution of care to rural areas and that barriers to access our wellknown and great. Ai can squash those barriers and closes gaps in a few different ways. First there is always a supply and demand, when you think about distribution of resources across rural areas. Making clinicians more efficient. Making clinicians more available means there are more clinicians available to see those patients in rural areas. Many of the barriers come from a lack of health literacy, lack of access to the Healthcare System. Ai also can help patients in rural areas level the Playing Field by assisting them in better understanding their care, better navigating the care system, and better understanding how to find the best care for themselves. The incentives that congress can encourage include the development in education, of ai skill sets across the Healthcare System, but specifically the clinicians were going to practice in rural areas and healthcare leaders were going to lead systems in the rural areas. That education is very important, as i mentioned in my statement, it doesnt come naturally to many of our institutions. Thank you, dr. Schlosser, welcome to another Johns Hopkins trained physician. I took that to congress, you took it another direction. The fda has been regulating some forms of ai under existing authorities for drugs and biologics as well as medical devices. Do you think the current regulatory structure is sufficient to keep up with the innovation in ai uses in healthcare . Yes, and thank you for the question. I actually have some experience, i was a medical officer for the fda a number of years ago and i would say that the rate at which ai is changing healthcare is likely going to require us to think a little differently about how we regulate medical devices. The current approach, which really is based on laws from 1974, i think, never really anticipated the kind of technology we are talking about. Dealing with models that can learn over time, its something were going to have to Work Together, i think, to figure out what the regulatory pathway looks like. There is incredible potential here but i feel the movement and progress of ai is a little bit outpacing the current regulatory approach. Thank you. And chairman, before i yield, i ask unanimous consent to enter into the record a letter of support from the American College of surgeons. Seeing no objection, so ordered. Thank you, and again, thank you to the witnesses for being here and the yield. Thank you, thank you for being here. I have heard a couple of you speak before, i will start with you, dr. Schlosser, since you are from tennessee or hca. Lets talk about the pharmaceutical supply chain, it can be very difficult and complex to trace and as a pharmacist i am responsible for knowing every step in that process in the supply chain. From the manufacturer to the dispensing of the drug because of the pedigree. How can it be used to help pharmacists in their role optimizing medication to providing patient care and improving Health Outcomes . How can we use that and be reimbursed for that . Think for the question. I think there is to me this opportunity. We actually already use a variety of Artificial Intelligence in our pharmacy processes. We have done this for years. We have heuristic models which are basically rulesbased models constantly surveilling patient charts, looking for opportunities that can be served up to the pharmacist, be they drug interactions or substitutions or places where we could be more efficient or provide more effective treatment. I think those models are only going to improve and get better with the advent of the more advanced Artificial Intelligence algorithms. Think the same is true for pharmacy supply chains. The ability to get predictive in understanding the demands and needs of her patients on a hospital or even unit basis, and then be able to go upstream and ensure we have the adequate supply to meet those demands and if we dont can we make preemptive steps to ensure we can maintain adequate supplies, its another area we are already working on and think it will be great benefit. Very good. You have already told us about how hca Hospital System sees value in adoption of Ai Technologies without additional payment in that. You have talked to us about removing the Administrative Burden and its reduction of time spent the does not involve direct patient care. So this is my question. And this is your chance to tell me what to do. What recommendations do you have for this committee in creating payment models in ai for Healthcare Services applications and addons . Some of my panelists have had a chance to weigh in on this already today. I do think is this Technology Advances and becomes more meaningful and essential part of healthcare delivery, that cms is going to have to find an approach to reimburse for this technology. In my personal opinion, it is not that different than the initial purchase they try to take around chronic disease management. How you reimbursed for sort of the ongoing work, in this case an algorithm would do, to help prevent complications, to help reduce costs . The last comment i want to make here is we have a great opportunity with ai since we are at the beginning , to take a Business Case model approach to how we deploy the technology. And not have it just be another technology that we deploy that adds more cost and then we try to add more reimbursement and therefore drive up the cost of the Healthcare System. But instead be really thoughtful about, how can these technologies make us more efficient and more effective and decrease the overall cost of the Healthcare System as we deploy them. So, you will continue to exist. Dr. Longhurst, how do you think ai and medical liability intersect . Vent has the question. Thank you. As was previously mentioned, ai is a tool, whether being used for diagnosis or Treatment Options or others, ultimately liability for treatment of patient rests with the treating physician. We have had perhaps, not ai tools for a long time, but we have certainly had medical Decision Support tools for a long time that suggest potential drug interactions or dose range errors or other things. Liability has always rested with the treating physician to see these alerts, manage them but make the best decision for the patient. I think the real question about liability and ai comes if you take the human out of the loop. If there is a step taken towards making diagnoses without clinicians, it makes all sorts of other questions about licensing the tools. Exactly, do you see a scenario where litigation increase of doctors dont use ai . Has the question. A recent survey of patients asked, would you see a doctor that was not using ai . The predominant answer from patients was, i would be concerned if my doctor was not using the latest tools. I think as was recently described by mr. Newmantaker, if these are shown to be best practice, if they can decrease mortality, if they can increase survivorship, then they will become a best practice and it should be used in every case. Thank you. I yield back. The chair recognizes the chair lady from iowa, ms. Millermeeks. Thank you very much. I want to add in my own personal expense with Electronic Health records as a doctor and its much more than two hours of my time after seeing 35 to 40 patients in a day. Just as an example, a global postop no charge after i finish completing might medical record, it took an additional six clicks to put in no charge. It takes me 10 seconds to write andc. It certainly does lead to burnout. I would also like to follow up with something dr. Longhurst said and also dr. Schlosser. Certainly the fda has not kept up with medical devices utilizing Artificial Intelligence, not only generative but repetitive Machine Learning. At the university of iowa, dr. Ebert off in the phd had one of the first medical devices approved by the fda, i like to submit for the record an article on effectiveness of Artificial Intelligence screening and preventing vision loss from diabetes, a policy model. That would lead to reimbursement. What is great about this is it increases access by having a device that can be put into any persons office, whether it is an eye care writer or family practitioner. The second letter is a letter of support from Johnson Johnson the does talk about privacy, equity, bias, and transparency in the systems since those things have been brought up. Thank you, we will accept the documents list at the end and make sure and give my friend a chance to review. I know i did it first because im older and i forget things. Mr. Shen, we have heard a lot about ai over the past few years and the potential risk contributed to unregulated ai integration. However the fda has been regulating softwarebased medical products since the 1970s and we know that ai integration into healthcare has already raised the status quo of care. We have seen it in digital pathology, drug organization, integration in Patient Engagement, personalized risk prediction. Can you give examples were gaps exist in current regulation that congress can address to ensure continued innovation that will drive better, more Personalized Care for patients without burdensome regulation . Thank you for the question. I think what we are seeing with the fda, and will continue to work closely with the fda to make sure they stay current with the rapidly changing technologies that are there, i think the challenge we are seeing and it was acknowledged here on this panel today, is really how Artificial Intelligence continues to change and what becomes a algorithm today might be a different algorithm tomorrow or might need to be adjusted or increased in terms of accuracy or whatnot after its being used in a clinical setting. So i think this is were areas like pc cp that we have worked in here with this committee obviously, these are important aspects that the fda needs to consider and actually not water down in terms of its ability. Because that actually will then inhibit us from being able to continue to develop and innovate in this particular area. Thank you. Dr. Schlosser, many hospitals and Hospital Systems are facing significant staff shortages. We are seeing ai as a meaningful tool to help alleviate some Administrative Burdens driving providers away from the medical profession. A recent report from Goldman Sachs notes that shifts in workflows triggered by these advances could expose the equivalent of 300 million full time jobs to automation. What steps can Congress Take to facilitate better ai integration to Health Systems to streamline processes that will allow Healthcare Professionals to focus more on patients and less on billing, coding, et cetera. Is a great question. Thank you. I think this is incredibly important that we do streamline our ability to use ai to tackle this serious workforce problem we have that is only going to continue to get bigger. We think the gap between supply and demand for nurses alone will continue to increase over the next decade. I would see the easy answer is, lets not put too many burdensome regulations between us and our ability to deploy ai to support our healthcare workforce. We are not talking about ai directly influencing patients or providing diagnoses, we are talking about it removing Administrative Burden. That is an area where with the right responsible ai platforms we should be able to move quickly to adopt technologies and free up the workforce to handle this increasing demand. Thank you. I saw dr. Nguyen and shaking their heads. I have another question. With that i yield back. Thank you. If youll give us the documents you submitted for the record so we can review those, we appreciate it. The chair now recognized the recognize the gentleman from california. Thank you tour witnesses. An interesting hearing. Mr. Shen, i want to ask you about some your interactions with the fda. We are really in a crossroads when it comes to devising a Regulatory Framework for Artificial Intelligence. We can either follow the lead of entities like the European Union who believe ai is its own unique discipline and there needs to be a separate bureaucracy spun off to issue licenses with respect to the use of ai. Or we can follow the lead of countries like the uk who has pointed out that because of the risk of ai is so contextual, that the existing authorities are best equipped to regulate within their sectorial spaces with a bunch of technical help and resources. I was curious, we have to choose, we are at a crossroads. We go one way or the other way, there is really no middle ground. Which of those two paths do you think we should follow . Is it easier to teach the fda what it doesnt know about ai or to teach a brandnew agency everything that the fda knows about ensuring Patient Safety . A very good question. Certainly a tricky topic. What we have to remember is that , at least in our industry from a vendor perspective weve been working closely with the fda for many years and we work, we have direct dialogues with them around this topic on a weekly basis. I think the other thing that is important to remind ourselves here, especially in the context of Artificial Intelligence, is that a i cant not just be considered a separate type of technology, but this technology is also being embedded into the medical devices themselves as well. For instance, ct scanners or mri scanners, they have a i built in to the system that allows for better image quality or faster exams for the patient. A lot of pitch to the patients are happening already with the Ai Technology built in to the medical devices themselves. We have to consider that, especially when we consider how we move forward. I think that is good point. I also am heartened by your comment that you feel the existing regulatory relationship with the fda is doing a good job at both ensuring Patient Safety and catalyzing innovation. And so, i think that is a pretty powerful argument, you know, for maintaining that relationship and powering the fda to regulate in that space. Mr. Newmantaker, thank you very much for your testimony. You said something i found incredibly interesting. You said the best we can expect from ai is that it repeat the existing human biases that exist in the date it was trained with. And i found that a fascinating statement. I dont like the use of the word bias because its very human word and when you apply it to Machine Learning, algorithms, there is no such thing as bias, they are all biased. Machine learning is all about bias because your training it to generalize. We call a bias when you talk about kind of maintaining our social standards. We say, for example, it would be wrong to consider someones race when making a hiring decision. We can all agree that is true. But that also means scrubbing the data that we used to train ai that makes those recommendations for things that could be used as proxies for race. That is the difficulty we have had so far. You were talking about how important it is in the medical context of maintaining high quality data sets to avoid those as these. Biases. How do we ethically navigate the space of patient consent . If you have a chest xray, detecting covid pneumonia, from a chest xray, if you were a patient and you come in and get a chest xray, not consented for the use of that xray to be used to train a Machine Learning algorithm. You have the right to say i dont want the data used . If you do, the problem is, that is introducing bias into the algorithm because, from a statistical sense, you are biasing the algorithm of the algorithm because who knows what else the group of people who withhold consent have in common. Right, so a statistician would say its a serious problem. How we navigate that space and protect the patient data and at the same time avoid biasing the algorithms . Thank you, thats a great question. I come from the world of Clinical Research where there is always the opportunity to refuse to participate. And im generally of the mind that that should always be the case, that if patients wish to opt out they do. It does create a certain bias, a volunteer bias of those who want to dissipate. But i think that is a bias we can accept. As far as the issue of replicating the human biases, as i mentioned in my testimony, i believe we have, they existed two different levels, but the most important piece is where our biases are causing us to behave differently as clinicians. So if i dont order the same test in a black patient that i order in a white patient for the same circumstance in the same condition in the same appropriate is, that is the kind of bias i dont want to replicate in my ai systems. I think that is why well curated Gold Standard data sets are so critical. I agree, i am a ai optimist so i would argue against your statement. The best we could expect is replication of existing biases. I think its a golden opportunity to remove biases. I say im out of time. To very much. I yield back. We now have a vote on the floor, but we only have one member left to ask questions so we can hopefully complete this now. We will get started on that. The next member to speak is the gentleman from texas, mr. Crenshaw. Im glad were doing this hearing. I think healthcare utilization of ai might be the least of our worries. I also worry we are not always talking about this in an accurate way. We are not properly differentiating between advanced algorithms and ai. We are just saying ai. From our witnesses, that is for everybody, america,to properly regulate it, and im going to ask you what you mean by that. Im genuinely curious how and what we do. But, we have to talk about it accurately first, we mean Machine Learning, that you cant actually look under the hood and change, that is where it gets scary, we are talking about advanced algorithms. What people call facebook and instagram, listen and watch my actions and make predictive analysis based on that. I have used that like today and call it a. I. , it is not a. I. , you can change that algorithm, you can change how that works, the programmers can go in there and change it. A. I. You cannot change, so i think we need to be really accurate about what we mean by a. I. , it is meant to mimic a person, and that can be really amazing especially for healthcare. So, things we have to talk about is what data inputs go into that Machine Learning. Is it everything . That is how you get chatgpt and what kind of person is it mimicking . A good person or a bad person . This stuff gets scary really fast. Talking about healthcare, it is obvious you want to make the data inputs, does that need to be a law, is that one of the regulations youre talking about . So i want to stop there and ask you. A couple said we need to regulate it, but im curious what you mean by that. Dr. Tucker, maybe you can start. Sure, thanks very much. I think your question is very important, in terms of regulatory oversight, i think there are certain gaps with respect particularly to diagnosis in this space, so i believe that for example, if we take about system checkers for diagnosis, where there is a legal advice, that it is the incumbent upon us to pay more attention to that Consumer Health space that has been brought up previously. Kind of like a digital watermark, i have talked about that before with respect for a. I. , it should be known that wherever this output is, it is a. I. And not a person. Not just that, but when people are making decisions about how to act as the Healthcare System and it is based upon some kind of algorithm behind the scenes, there should be an accountability in that framework and right now there isnt any accountability to everything that exists outside the proper confines of say the hospital setting or clinic, before you get to the Healthcare System. Theres a lot going on. Can i give you example of what you mean by that . Lets say somebody types into the symptom checker that they are dizzy and it says dont worry, it is nothing, little rock crystals in your ear. This is hypothetical by the way . No, it is not hypothetical, there are a lot of symptom checkers that have been looked at and the accuracy is often quite low. And what essentially happens, at the bottom, there is a legal disclaimer that says this is just a toy. If you want real medical advice, ask your professional. That is not how patients are dealing with that and i do think some of those Decision Makings, you are having a stroke and if you need to be sent to an Emergency Department, this a. I. System unregulated is saying dont worry, just stay home. That is a real risk to the Public Health. There is broad agreement that we would never want a. I. To operate independently, maybe not never, there might be star trek mode at some point, but definitely for the future you would definitely have a doctors blessing. Because theres Amazing Things that can happen, when we are seeing this technology coming out of china, of course there is the competition problem, too. But, apparently diagnosing pancreatic cancer at 99 success rate, that is amazing. Doctors should still look at that afterthefact, but its just Amazing Things we can do with this technology, also amazing risks that can happen especially when we are talking about the more generalized generative a. I. Which is basically mimicking a person. And again, the question, congress has to ask whether it is healthcare or any other conversation, what kind of person is it mimicking . We dont know the answers to that and we have not talked about it enough in this congress. I yield back. Thank you, that concludes all members questions, and we thank our witnesses for being here, but we have the documents for the record that some of the members have asked for and others have been submitted, and i can send into the record that the documents included on this documents list, without objection that will be in order and i remind members, some said they were going to submit questions for you, youve got 10 Business Days to submit questions for the record, and we ask that the witnesses respond accurately. And members submit their questions by the close of business on december 13th. Again, we appreciate everyone if you are here, for your time, this is something that we are still very curious and we have very engaged members and we want to understand it, so we can act appropriately without to protect, but without impinging the great things that can come from this. That is what we are focused on and again, i appreciate it and without objection, the subcommittee is adjourned. Cspan is your unfiltered view of government, we are funded by these Television Companies and more including cox. When youre connected, youre not alone. Cox, supporting cspan as a Public Service along with these other television providers, giving you a front row seat to democracy. Cspan now is a free mobile app featuring your unfiltered view of what is happening in washington, live and on demand, keep up with the biggest events with live streams of hearings from the u. S. Congress, white house events, the courts, campaigns and more from the world of politics, all at your fingertips. You can also stay current with the latest episodes of washington journal and find scheduling information from c span tv networks and cspan radio, plus a variety of compelling podcasts, cspan now is available on the apple store and google play. Scan to download for free today, or visit our website, c span. Org, your front row seat to washington, anytime, anywhere. Since 1979, in partnership with the cable industry, cspan has provided complete coverage of the halls of congress, from the house and Senate Floors to congressional hearings, party briefings and committee meetings. Cspan gives you a front row seat to how issues are debated and decided with no commentary, no interruptions, and completely unfiltered. Cspan, your unfiltered view