Key Takeaways
- Liability around AI depends on how the tools and algorithms are utilized in clinical practice, and whether the technology is considered the standard of care.
- Data security and privacy concerns should prompt policymakers to update HIPAA regulations to include rules around the usage of AI.
- In addition to physician liability concerns, hospitals, and manufacturers may also be considered liable for errors depending on where in the care pathway errors occur and how patient care is ultimately affected.
The integration of artificial intelligence (AI) into healthcare promises a revolution in the delivery and management of medical care. Initially conceptualized as a tool for handling large datasets more efficiently than humans, AI’s role has evolved significantly. Today, AI assists with data management and plays a crucial role in diagnostic procedures, treatment protocols, patient monitoring, and personalized medicine. Machine learning algorithms, a subset of AI, are particularly adept at recognizing patterns in complex medical data, which can potentially lead to earlier diagnoses for certain diseases.
The Growth of AI
Several factors have facilitated the rapid advancement of AI:
- Increased computing power enables the processing of vast amounts of data at high speeds, making real-time data analysis feasible in clinical settings.
- The digitalization of health records, imaging data, and genomics has provided the data sets needed to train medical AI models.
- Breakthroughs in AI predictive analytics can help forecast patient outcomes, suggest personalized treatment plans, model the spread of a virus, and manage healthcare resources.
AI tools have the potential to enhance healthcare delivery and reshape standard practices by providing more precise, efficient, and impactful interventions. Some generative AI models can match or even surpass the accuracy and speed of doctors in diagnosing complex conditions. For example, AI has excelled in dermatology by accurately identifying skin cancer types from image data, and in ophthalmology, where AI algorithms have been shown to detect diabetic retinopathy at rates comparable to top-tier specialists.
Beyond diagnostics, AI is also making strides in patient interaction. AI-driven chatbots and virtual health assistants are increasingly able to handle certain patient communications in a scalable way, offering support and guidance to patients 24/7. These machine abilities presage AI’s potential to revolutionize healthcare practices.
AI As An Add-on
While AI has the potential to improve healthcare, many patients will prefer the reassurance and familiarity of human interactions with their doctor. This suggests a future in which AI complements rather than replaces human clinicians.
For instance, doctors may treat patients with an AI assistant in attendance. This could take the form of AI systems integrated into diagnostic equipment or even humanoid robot nurses that assist with routine tasks. AI assistance may enable healthcare professionals to focus on direct patient care, while AI performs background analysis on data and lab results.
“To maximize its benefits and mitigate risks, the policies and regulations governing its use need to evolve to ensure safety, efficacy, and fairness in AI-powered medical care.“
Bob Hansen, JD
Legal Implications of AI in Healthcare
The potential of AI in healthcare, however, is not without challenges. The rapid deployment of healthcare AI tools raises legal, ethical, and operational questions. To maximize its benefits and mitigate risks, the policies and regulations governing its use need to evolve to ensure safety, efficacy, and fairness in AI-powered medical care.
AI & Standard of Care
The promise of AI to improve medicine and patient outcomes is widely covered in the press, yet questions remain: when AI contributes to a patient injury, who and what will be held liable for the consequences?
As AI technologies demonstrate efficacy and safety in clinical trials and real-world applications, their adoption will likely redefine the standard of care in various medical specialties. AI tools proven to improve patient outcomes could become a benchmark for clinical practice, similar to the adoption of other medical innovations in the past.
As these tools become ubiquitous, significant legal implications arise, particularly in the realm of medical malpractice. At some point, if a doctor fails to use proven, reliable AI diagnostic or treatment tools considered standard of care within a specialty, this omission could potentially be the basis for a medical malpractice lawsuit (e.g. if an AI-based system is known to improve diagnostic accuracy for a disease and a physician does not use the technology, resulting in a missed or delayed diagnosis, it could be viewed as negligence).
Consequently, the legal framework surrounding medical malpractice may evolve to consider the use of such AI technologies a requirement rather than an option.
However, no software product is perfect. AI-based technologies have shortcomings, which underscores the need for continuous evaluation (and re-evaluation) of AI-generated work, continuous professional development, and adherence to AI medical standards as they emerge.
Liability Considerations For Doctors
It is safe to assume that doctors will remain at least partially responsible for patient care and treatment decisions, even if AI contributes to the physician’s decision-making. Modern medicine leverages many state-of-the-art diagnostic technologies, from medical imaging to the latest in blood analyses, yet physicians generally remain liable for medical malpractice due to misdiagnoses. Unless AI is implemented in a medical device (e.g., a surgical robot) where strict liability may be asserted, liability may lie with the physician who relied on the AI tool.
A recent survey of medical malpractice cases revealed that when physicians adhered to erroneous software recommendations, patient malpractice claims alleged that the physician should have ignored the software recommendations and independently reached the correct diagnosis or treatment decision. Accordingly, when using AI for advice or consultations, physicians should be aware of the potential for generative AI “hallucinations” (answers that sound correct but are not), critically evaluate the outputs of such tools, and make independent judgments and decisions.
As a result, physicians must understand how much and how far AI tools can be trusted and remain the decision-makers regarding patient care. On the other hand, if AI tools are expected to become a new standard of care (e.g., machine vision, and imaging), physician liability may be reduced.
Liability For Hospitals
Generative AI technologies and tools hold promise for hospital operations and patient care by improving efficiencies and reducing costs. Examples of AI usage at the hospital level include managing patient records, tracking and dispensing medications, recognizing infection patterns, and more.
The same survey of medical malpractice cases mentioned above revealed that when patient harm results from a malfunction of software embedded within medical devices (e.g., implantables, surgical robots, and medical monitors), malpractice plaintiffs may assert claims against physicians and hospitals for negligent use, installation, programming, and maintenance of such devices.
Due to these risks, hospitals may be slow to implement AI-aided devices and systems out of caution to avoid malpractice complaints. However, given AI’s potential life and cost-saving benefits and the likelihood of altering the standard of care in some regards, it’s important that hospital policies remain open to adopting AI while proactively identifying and mitigating the unique risks that come with the technology.
Liability For Manufacturers
Manufacturers can be found liable for products that cause harm to patients under the doctrine of product liability. The same survey of med-mal cases demonstrated that when patient harm results from defects in software, malpractice plaintiffs typically bring product liability claims against the developer. However, there is an ongoing debate about whether AI software is a product, traditionally a physical thing, subject to product liability risks or something else entirely. This determination likely depends on how directly the AI tool contributed to patient injuries, e.g., in the case of doctors, nurses, or hospital staff.
However, courts may struggle with assigning liability when generative AI is involved somewhere in the diagnosis-treatment-product-care causal chain due to the novelty and complexity of AI technologies.
For example, if an injury is alleged to be due to an error in a generative AI tool, the root cause, and thus liability, may be hard to identify. Generative AI models are typically generic, with the functionality based on the training dataset and the prompts applied to the model. Unless a flaw can be identified in the training dataset (e.g., bias, erroneous data, or improper training methods) or the error can be shown to have been an AI hallucination, blame may lie with the user or party that prompted the AI model, potentially shifting liability back to the physician or hospital.
Developers of medical AI tools would be well advised to work hard to solve the challenges of hallucinations and ensure training datasets are accurate and devoid of bias that could lead to erroneous outputs.
Data Privacy & Security Concerns
The use of large datasets of information that drive AI tools and algorithms opens up the technology to risks related to handling Protected Health Information(PHI), and the potential of inadvertently violating Health Insurance Portability and Accountability Act (HIPAA) standards. For example, AI chatbots can risk exposing patient data by transmitting PHI outside of the protected healthcare environment, which could lead to unauthorized disclosures if proper business associate agreements are not in place.
There are also challenges related to de-identification of datasets. Traditional methods might not suffice in the era of AI, as these systems can potentially re-identify patients from datasets that were previously thought to be anonymized, thus placing patient privacy at risk. Additionally, there are legal complexities concerning the control and use of patient data by private AI companies, which may not always align with the stringent requirements of HIPAA.
To ensure that AI technologies’ promises of improving medical care are fulfilled, updates in regulatory frameworks, like HIPPA, may be required to address and effectively manage the challenges to patient data privacy raised by AI technologies.
AI & Informed Consent
The hype surrounding AI, in general, may shape patient views and fears about AI, highlighting the importance of educating patients so that they understand the risks and benefits of technologies that are being used for diagnostic and treatment purposes.
Appropriate informed consent regarding AI use includes:
- Explaining how the AI program or system works
- Explaining the healthcare provider’s experience using AI
- Describing risks versus potential benefits of the technology
- Discussing the human versus machine roles and responsibilities in diagnosis, treatment, and procedures
- Describing safeguards in place, such as cross-checking results between clinicians and AI programs
- Explaining issues related to confidentiality of patient information and data privacy risks.
Further, the rapid advancement and evolution of AI necessitates a dynamic approach to informed consent, including continuous updates to the consent process to incorporate new information about AI tools as they develop and become more integrated into clinical practice.
Regulatory & Policy Framework For AI In Healthcare
The World Health Organization (WHO) has issued new guidance aimed at the ethics and governance of large multi-modal AI models, proposing over 40 recommendations for governments, technology companies, and healthcare providers. This guidance is designed to ensure the appropriate use and equitable deployment of AI in healthcare, promoting benefits while safeguarding against risks.
In the United States, the FDA has been active in shaping the regulatory landscape for AI in medical devices. The agency’s new guidance proposes a flexible approach to the oversight of AI/ML-enabled medical software, facilitating innovation while ensuring safety and efficacy. This includes a Predetermined Change Control Plan, allowing for ongoing adaptations of AI functionalities in medical devices without the need for re-approval for every change, provided that these changes adhere to predefined safety and performance guidelines.
The American Medical Association (AMA) has also been involved, establishing its first policy recommendations on augmented intelligence. These recommendations aim to guide the incorporation of AI into healthcare settings, addressing ethical, educational, and policy challenges to ensure that AI tools are implemented responsibly.
Each of these efforts reflects a broader recognition that AI’s tremendous potential comes with complex challenges that require thoughtful regulatory responses to ensure that benefits are maximized without compromising patient safety or ethical standards. These evolving guidelines and recommendations are crucial for supporting the safe and effective integration of AI technologies in healthcare practices globally.
Final Thoughts
AI technologies are likely to positively transform many aspects of healthcare; however, with such transformations come new and unique risks for which healthcare professionals must remain aware. Managing such risk will require updated and potentially new regulations and medical standards. The remaining question is whether policy and regulation updates can keep up with the pace of developments and adoption of AI technologies in medicine.
Robert Hansen, JD
Bob has 25+ years of experience as a patent attorney and is currently a partner & founder at the Marbury Law Group ($12M law firm). He has experience advising dozens of startups on intellectual property law and business strategies. He is also the founder of Swim Magazine. Prior to Marbury, he was an M&A manager at Raytheon Co. Mr. Hansen holds a B.S., M.S. in nuclear engineering from the University of Florida, and a J.D. from George Mason University.