Forecasting the Legal Implications of AI
New deep-learning technologies entering the health-care space bring a host of unknowns for patients and physicians.
Artificial intelligence (AI) represents a transformative technology that has the potential to disrupt the field of radiology, potentially more than any other health-care technology in recent memory.
In addition to the changes AI will bring in day-to-day practice, the legal, regulatory, and policy landscape will also be transformed.
AI technology has provoked a mixture of excitement, fear, and anxiety in both the public domain and in the health care industry. This early uncertainty is likely to translate into inconsistent regulatory and legislative responses to health-care AI technologies at the federal, state, and local levels. An appropriate example of this is the robust and oft-conflicting regulatory responses to the use of AI in the transportation industry. In the first five months of 2017, 33 states have introduced legislation related to autonomous vehicles, compared to only six states in all of 2012. Beyond legislative and executive action, many courts will be faced with tortured attempts at applying traditional legal principles to new-age quandaries created by AI.
As the ACR frames its position and strategy with respect to AI technologies, the AI Advisory Group is taking into account the following legal and regulatory considerations and implications.
Data from millions of individual patients will be necessary to develop and train AI tools. What legal and ethical obligations are owed to the patients whose data are used to create these tools?
AI tools in health care will allow for earlier and more accurate predictions regarding patient behavior and health care outcomes, raising challenging privacy questions. For instance, should AI algorithms using medical or genetic data be employed by insurers to predict and price insurability of an applicant? Should reproductive specialists use AI to predict the intelligence, sexuality, or health of a blastocyst for embryo transfer? And can a patient who values the spontaneity of life prohibit such forward-looking predictions of their health status? These are just a few possible scenarios.
A human must attain a certification or license before interpreting a chest X-ray. How do AI tools become certified to perform the same medical task? For instance, IBM reported that Watson was ready to take the Radiology Boards. Would passing the boards be enough to certify Watson? Or should AI tools be held to a higher standard of care? This issue of how we certify AI tools must strike the right balance between encouraging innovation and growth in this industry while protecting patients from ineffective or inconsistent diagnostic solutions.
As radiologists cede more clinical decisions to machine-based algorithms, do they become inherently less responsible for the final interpretations? The answer is almost certainly yes. As a result, this is likely to mean a gradual shifting from professional liability (holding the radiologist responsible for harm caused by a medical decision) to product liability (holding the manufacturer of the AI tool responsible for harm caused by a medical decision).
As AI substitutes begin filling human roles, there will be a shake-up in the needs of the labor market. Certain health care jobs will be eliminated, others will be created, and many job descriptions will simply be rewritten. Ultimately, it is likely that AI will increasingly shift health care investment from payroll to capital expenditure. If in doubt, just look to the transformation of Amazon shipping warehouses in the last decade. The impact of AI on income and staffing levels of radiology health-care providers in the near term, however, is less clear.
Patients will benefit most from artificial intelligence if radiologists serve a leading role in guiding the technologies that best enhance medical imaging diagnosis and treatment. – James A. Brink, MD, FACR
This column is intended to provide an initial, non-exhaustive framework by which to start considering the legal, regulatory, and policy implications of radiology advancements in AI, based in part off of the foundational work of the "One Hundred Year Study on Artificial Intelligence" at Stanford University.1 We expect iterative improvements and expansion of this framework as experience and advancement dictates.
An impact of AI on radiology within the next couple of years — or even the next decade — is uncertain at best. But one thing is clear: Advances in AI are thrusting us into a brave new world, whether we like it or not. Our field could choose not to act, sitting passively and letting the uncertainty of the future loom uncomfortably over our profession like a final verdict yet to be announced. But, thankfully, inaction is not the favored posture of the College. Modeled after the College's proactive response to recent payment reforms (which enabled us to help shape CMS decisions in ways favorable to our membership and patients), we are similarly committed to being proactive with respect to AI.
To that end, the ACR recently launched the ACR Data Science Institute™ (DSI), which will put radiologists at the forefront of developing and enhancing tools to effectively guide the introduction of AI in clinical imaging practice. The DSI will build on our decades of expertise in modality accreditation, appropriateness criteria, and practice parameters to develop standards and a validation process for AI applications in medical imaging.
ACR DSI volunteers and staff are moving quickly to collaborate with industry, regulators, and health-care stakeholders to improve patient care in the following ways:
• Defining appropriate AI use cases for medical imaging
• Setting standards for AI interoperability
• Evaluating the diagnostic performance of AI algorithms
• Enhancing machine-learning tools for effective application in clinical practice
• Addressing regulatory, legal, and ethical issues that accompany AI in medical imaging
"Patients will benefit most from artificial intelligence if radiologists serve a leading role in guiding the technologies that best enhance medical imaging diagnosis and treatment," said James A. Brink, MD, FACR, chair of the ACR BOC. Advised by the AI Advisory Group, the ACR DSI promises to do just that.
James A. Brink, MD, FACR, Chair
H. Benjamin Harvey, MD, JD, legal community representative to the ACR Artificial Intelligence Advisory Group