House Science Committee Issues Testimony From Oak Ridge National Laboratory
"Thank you for the opportunity to appear before you today. My name is Georgia Tourassi. I am a Distinguished Scientist in the
INTRODUCTION
"As a formally trained physicist, engineer, and applied scientist with three decades of AI experience in biomedical discovery and healthcare delivery, both in academia and a national lab, I have observed the growth, subsequent stagnation, and latest renaissance of AI for health. These experiences have shaped my views on the AI topics you are considering today. In my role as director of the
(i) providing responsible management of national data assets,
(ii) accelerating innovation, and
(iii) supporting a targeted, ethical, and socially diverse research and workforce development strategy.
"With examples drawn from healthcare, I will discuss some of the societal implications of AI and the need for research and development (R&D) investments to ensure ethical use of the data and effective and energy-conscious AI technology. Lastly, I will discuss a vision of how the
THE IMPACT OF AI FOR OUR HEALTH
"The explosive growth of AI is driven by the convergence of big data, massive computational power, and novel algorithms. Together, these three pillars enable AI to accumulate, analyze, and automate the delivery of functional knowledge in many application domains.
"The national laboratories are a remarkable asset for the nation. Over the past 75 years, they have consistently provided the science and technology needed to address national problems. As a
"Through the Department's Strategic Partnership Projects program, we have been given the opportunity to apply our resources to develop novel and promising AI solutions to several challenges in healthcare, as outlined below.
AI for modernization of national cancer surveillance
"Supporting efforts to reduce the cancer burden in
"NCI SEER partnered with
"The collaboration has had a direct impact on cancer surveillance over the first three years. Deployed AI algorithms have demonstrated high accuracy over 12 SEER cancer registries and millions of pathology reports received at the registries and processed with increased efficiency. The high accuracy and time efficiency of AI are expected to reduce the workload burden for cancer registrars while allowing them to realign their focus on abstracting additional complex variables (e.g., new cancer biomarkers, cancer recurrence) that are currently not possible using AI. Furthermore, by leveraging AI for "real time" cancer incidence reporting, we are moving towards real-time eligibility assessment of cancer patients for clinical trials. This is key to NCI's goal of expanding the number and representativeness of patients in clinical trials and other research studies.
"This effort to modernize the national cancer surveillance program exemplifies the benefits of a federally coordinated strategy to leverage AI, high-performance computing, and sensitive health data assets for real-world application. In addition, the specific application domain enables ORNL computational scientists to make fundamental contributions to the
AI for intelligent clinical trials matching
"ORNL recently participated in
"Clinical trials have great potential in advancing the standard of care. However, matching patients with clinical trials remains a challenge, mostly due to the unstructured nature of eligibility criteria as well as the clinical documentation. The approach that we proposed leverages the power of AI technologies such as large-scale knowledge graphs and deep learning to bring together cancer registry data, medical ontologies, and clinical trials data to answer complex questions and provide real-time feedback for patients and clinicians on novel experimental treatments that are available to them.
AI for supporting breast cancer diagnosis
"In an effort to reduce errors in the analyses of diagnostic images by health professionals, ORNL scientists have been working to understand and improve the cognitive processes involved in medical image interpretation. Our work has potential to improve health outcomes for the hundreds of thousands of American women affected by breast cancer each year. Our ORNL team found that analyses of radiographic breast images by expert radiologists were significantly influenced by context bias, or the radiologist's previous diagnostic experiences. So lab reserchers designed an experiment aimed at following the eye movements of radiologists at various skill levels to better understand the context bias involved in their individual interpretations of the images. Using the ORNL supercomputers, our team was able to rapidly train deep learning models necessary to make sense of large datasets of eye-tracking and medical imaging data. These findings will be critical in the future training of medical professionals to reduce errors in the interpretation of diagnostic imaging and will inform the future of human-AI interactions and personalized medical decision support and human augmentation going forward.
AI to improve the health of veterans: MVP-CHAMPION
"ORNL is supporting the
"ORNL has brought its global leadership in computing and big data, as well as its demonstrated ability to analyze protected health information on a large scale, to the task of analyzing the
AI for Medicare and Medicaid: CMS
"A few years ago, ORNL applied its resources to create a Knowledge Discovery Infrastructure (KDI) for the
THE HARD TRUTH OF AI
"These examples demonstrate the potential of AI to revolutionize healthcare delivery. While machines won't replace physicians and nurses any time remotely soon, they do have enormous potential to assist health professionals and other decision makers with time-critical decisions for the smart delivery of healthcare. Furthermore, such studies will inform human-computer interactions going forward as we use AI to augment and improve human performance.
"Still, the hope of AI must be tempered. AI is often exaggerated with hype and unrealistic expectations of universal benefits. This can lead those outside the AI research communities to believe that AI is the silver bullet that will solve any problem and overcome the known, and unknown, challenges facing us now and in the future. The reality of AI is that it is experiencing a renaissance in terms of capabilities, but still faces the necessary reality check that those capabilities have limitations and there are social implications if we fail to transparently address these limitations. With AI's great promise comes an even greater responsibility. As much positive impact as AI can provide to the nation, we must recognize the pitfalls and possible ill-intended uses of this powerful technology. In this section, I will address separately the ethical and societal implications related to (i) sensitive data management and use, and (ii) AI algorithmic development and practice in the real word, both in the context of healthcare.
"The Ethics of Data: Access to large amounts of data is fundamental to AI. Furthermore, access to diverse and seemingly irrelevant data offers the most exciting opportunities in many application domains and certainly in health. For example, the adoption of electronic health records has created great opportunities for biomedical research to understand individual and population-level health outcomes over time. However, liberating and providing access to this data is both a technological and a policy challenge. The different facets of healthcare (clinical services data, pharmacy data, billing claims, insurance) are often siloed, in part due to policy and the intended use of the collected data. To create a richer picture, medical data cannot be siloed and must be combined with other data points, such as those providing context on a patient's living conditions, a well-known "gap" in the data collection system of the healthcare system that has substantial implications for care delivery. To address this gap, there is a big push in the field to include nontraditional datasets, such as socioeconomic data, socio-demographic statistics, environmental data, and even weather data, in predicting patient health trajectories. Broadly speaking, the biomedical research community is looking for ways to understand and describe the human holistically. At the same time, the sheer volume, variability, and sensitive nature of the personal data being collected require newer, extensive, secure, and sustainable computational infrastructure and algorithms.
"Driving sustainable, secure, and energy-efficient infrastructure to handle increasing health data and AI computation: There is a pressing need to consider federal investments in centralizing national data assets in an environment that can provide the compute resources, a secure data management infrastructure, and scalable data analytics capabilities. The
"Though AI is currently accelerating research across many fields and industries, an outstanding issue for sensitive domains, such as health and medicine, is how to preserve privacy while computing with shared sensitive medical data to obtain relevant insights. Such data includes personally identifiable information, personal health information, intellectual property, and other proprietary details, where potential leaks would have obvious adverse consequences. Removing personal identifiers and confidential details is insufficient, as an attacker can still make inferences to recover aspects of the missing data. Inference attacks can also jeopardize AI algorithms over shared data by targeting the shared AI model training process and the trained model itself. Indeed, serious threats are encountered in collective AI endeavors that aggregate data from different sources, since the most vulnerable source establishes the overall security level.
"At ORNL, we fully understand the need to overcome these threats as we work to fully utilize the wealth of information in shared data, including data within the ORNL secure data environment, and to apply our high-performance computing resources for AI to make otherwise impossible scientific discoveries and technological breakthroughs in sensitive domains. This is an underdeveloped field of research in which R&D investments are well warranted to develop new solutions so that the community can responsibly and privately share sensitive data for aggregated analysis, including training shared AI models.
"Aside from the data security concerns, the energy demands of digital infrastructures and compute- and data-hungry AI algorithms pose energy challenges. AI and other computing activities are projected to use over half of the world's energy by 2040.1,2 We cannot easily predict how we can balance AI's energy demands with AI's demonstrated ability to guide management of our energy resources more efficiently.
"Driving responsible use of sensitive data: The ethical collection and use of human data is an outstanding challenge for the research community. Although we all recognize the scientific value of human data, the debate over data ownership is ongoing in terms of how best to balance the promise of transparent AI innovation with the risks of unethical data handling, intentional or unintentional privacy breaches, and adversarial data use by hostile or malicious actors.
"As scientists, it is our responsibility and as a nation it is our obligation to invest in research and development and related policies to support socially responsible design and use of AI. Data is the new currency in all facets of life and most notably in healthcare, disrupting traditional technology transfer and business models as well as blurring the line between research use and commercial use of patient data. To maintain a strong ethical AI framework, we need to answer this fundamental question: Who owns the intellectual property of data-driven AI algorithms in healthcare? The patient? The medical center collecting the data by providing the healthcare services? Or the AI developer? Clearly, no single entity alone could deliver the breakthrough AI technology. We need a federally coordinated conversation involving all of the STEM sciences, social sciences, economics, law, public policy, and patient advocacy stakeholders. The conversation should also address how to revise and continuously update outdated legislation and regulations to address the emerging complexities of an otherwise exciting technological development.
"The Ethics of
"Driving responsible development of AI technology: AI developers will need to offer solutions that are not only "on-average" accurate but also offer a measure of trustworthiness at the individual decision level. The latter would require a detailed explanation of each individual decision AI makes, as well as deeper understanding of the conditions under which the technology is exceptionally successful or alarmingly flawed. To understand these intricate issues, scientists need to dive deeper into the possible sources of errors and uncertainty, including biases in the data collected to train the AI algorithms and implicit biases in the way AI is embedded with the human. Unless these issues are addressed and properly regulated by the appropriate accreditation bodies in healthcare (or any other relevant domain facing similar challenges), we will not see real-world value of AI outside the anecdotal studies.
"Driving responsible deployment of AI technology: For example, one area of research that deserves attention is how to most effectively integrate AI technology with the human "in the loop." Although there is increasing fear that AI will eliminate jobs, for the most part AI tools have been shown to be effective for narrowly defined, repetitive tasks. AI will not take the jobs of humans in healthcare, at least not any time soon, but it will change those jobs by improving how humans work, making humans more accurate and more efficient. Ultimately, humans and machines will have to work well together. But this synergy won't happen organically, as past health AI experiences have demonstrated. It is important to train healthcare providers in how to use AI responsibly, how to remain vigilant avoiding mistakes of over-reliance when supported by AI, and how ultimately to be knowledgeable users of the technology.
"The
"Driving the training of an inclusive and diverse AI workforce: With respect to addressing the data bias issue, aside from raising awareness and working towards algorithmic advances, an important first step is to acknowledge that socially responsible AI development and deployment starts with an inclusive and diverse AI workforce. When the community of AI developers both reflects the diversity present in the user community and is embedded in that user community from the start, we will be better positioned to anticipate unintended consequences in the real-world use of AI technology. Therefore, I advocate a multi-pronged approach throughout the AI lifecycle. First, we need to invest in foundational statistical and data sciences. During the development phase, scientists should promote a rigorous statistical framework to monitor for potential biases in the collected data. During the deployment phase, AI developers should implement rigorous quality control, monitoring AI performance across subgroups to confirm robust performance or identify performance gaps. We should work to communicate to AI users openly and clearly what they should expect across various settings, and we should educate users so that they are informed consumers of the technology. For example, responsible use of AI technology should become part of our mainstream digital education. We cannot anticipate every blind spot and we should not blame AI for learning from implicit biases in the data. Humans do too! It is unavoidable that we will generate AI code faster than we can deliver appropriate policies and regulations. By being inclusive, diverse, rigorous, and vigilant, we can mitigate many of the aforementioned risks. We will be able to democratically answer the ethical questions which will confront us as developers, marketers, users, and regulators.
"At ORNL, we are particularly proud of our work with area community colleges and universities to develop the next generation of AI-savvy citizens. Every year we provide high school and college students with internship opportunities to learn AI and use the world's largest computer to address pressing biomedical challenges. In partnership with the
TRANSFORMING HEALTHCARE WITH AI: WHY DOE AND WHY NOW?
"Our national healthcare system faces some pressing challenges. The average life expectancy in
"AI could offer powerful solutions to these challenges. By leveraging our federal data assets, computing capabilities, and AI, we can develop a strategic roadmap to extend health span and rein in healthcare costs by understanding the broad spectrum of all factors impacting well-being. For the first time, we are at the tipping point to map the human genome (i.e., genomic profile), phenome (i.e., physiologic status), and exposome (i.e., physical and social environment) in real time and across the human lifetime. This is an ambitious endeavor that requires transformative scientific advances in data storage, management, and analytics using massive amounts of heterogeneous health data.
"Understanding the human genome-phenome-exposome interplay also demands a multidisciplinary team of biologists, physicists, chemists, engineers, mathematicians, computer scientists, and clinicians as well as the cooperation of key federal, academic, and industrial stakeholders.
"
"The national impact of this investment cannot be underestimated. Studies estimate that every federal dollar invested to map the human genome returned between
"We can draw another lesson from the
CONCLUDING REMARKS
"The rapid adoption of AI has shown great promise across a wide array of domains such as healthcare, transportation, manufacturing, and cyber security, to name a few. It is becoming an integral part of our daily lives, offering solutions to problems as complex as drug discovery and as trivial as guessing what our next word will be when we text. AI can and will transform the national workforce. It can and will cause social and economic changes. Many of these changes will be positive, but AI can also reinforce and perpetuate harmful societal injustices. One thing is clear. As much as the science of AI is about difficult math and powerful computers, the impact of AI is also about the nation and its citizens. AI is a fundamental human endeavor.
"Our nation faces a formidable set of challenges: ensuring our national security in a changing world; increasing the availability of clean, reliable, and affordable energy while protecting the environment; improving human health; and enhancing
"A cohesive national plan for AI is an imperative need to secure the nation's economic competitiveness and well-being. At the same time, we as a nation have the unique opportunity to create a well-defined, federally coordinated roadmap to engage in ethical AI development that delivers benefits across all private and public sectors, perhaps most notably in improving human health. The
"Thank you again for the opportunity to testify. I welcome your questions on this important topic."
* * *
Footnotes:
1 Greenpeace, "Clicking Clean: Who is winning the race to build a green internet?"
2
3 http://www.cdc.gov/nchs/fastats/life-expectancy.htm
4 http://www.nejm.org/doi/full/10.1056/NEJMra1109345
5 http://www.usgovernmentspending.com/year_spending_2017USbn_18bs2n#usgs302
7 http://www.nejm.org/doi/pdf/10.1056/nejmsa1204629
8 Yabroff,
9 https://www.cbo.gov/publication/51580
10 Hood, Leroy, and
11 Goldman, Dana P., et al. "Substantial health and economic returns from delayed aging may warrant a new focus for medical research." Health affairs 32.10 (2013): 1698-1705.
12 https://plato.stanford.edu/entries/human-genome/
[TheHill]
American Action Forum: Understanding President Trump’s Executive Order on Health Care Price Transparency
FEMA: City of Norfolk, Virginia Saves Over $1 Million a Year in Flood Insurance Premiums
Advisor News
- I started using AI, but what now?
- How to support an entrepreneur client with multiple businesses
- Could a narrower focus help grow your practice?
- U.S. debt hits $40 trillion: how to support concerned clients
- Flourish brings private-bank-like cash solution to MassMutual’s network
More Advisor NewsAnnuity News
- MassMutual Expands Lending Options With New Secured Credit Line
- Corebridge Financial Introduces Progressive Trigger, a First-of-Its-Kind Index Annuity Strategy Designed to Turn Missed Years into Future Opportunity
- Insurer proposes $1.1M deal to resolve CA teachers’ annuity fee lawsuit
- IMOs get frustrated with slow carrier adoption of faster technology
- New class-action lawsuit targets Delaware Life over annuity disclosures
More Annuity NewsHealth/Employee Benefits News
- Your questions answered on Medicare open enrollment
- Terry Savage: Questions answered on open enrollment
- New Jersey residents will see Affordable Care Act premiums jump 20% on average for 2027 health insurance
- Health insurance hits municipal budgets, taxpayers hard
- Minnesota seniors face tough choices for health plans
More Health/Employee Benefits NewsLife Insurance News