I attended a meeting at the Philippine Medical Association yesterday on updating the PMA-PRC Code of Ethics to include provisions on social media, telemedicine, and artificial intelligence. Let’s discuss ethical issues in AI and healthcare at the #HealthXPH tweet chat 27 Jan 2024, 9 pm (Manila time).
T1. Does the use of AI pose a threat to physician autonomy? If yes, how can this be prevented?
There’s been great interest in reports that AI can beat human doctors in predicting heart attacks, detecting breast cancer, and even passing the USMLE. This has fueled the fear that AI will replace human doctors. But even as we use AI as aids in providing care, tools that help us make complex treatment decisions, what happens when the AI recommendation differs from our assessment? AI will not have “off” days while physicians can get tired, or simply lazy. There’s also cognitive bias when the AI’s recommendation already agrees with the physician’s initial assessment and the physician stops looking more deeply into the patient’s case. We also need to guard against deskilling.
T2. Does your institution have an open data policy? Given systemic and statistical biases from training data sets, how can we ensure that AI will uphold the principle of justice?
AI and machine learning are limited by the quality of data on which it is trained. How much Filipino data do you think are included in AI training data sets? This also reminds me of a clinicopathologic conference in the US where ChatGPT3 was pitted against human experts. The human experts won and were able to diagnose the rare disease. ChatGPT included that rare diagnosis in its differentials but picked the more common disease as the diagnosis because that’s how it was trained.
T3. Do you agree that explicability should be added as an ethical principle in the use of AI? Why or why not?
Use of AI has been referred to as a “black box.” How can a physician trust the AI output when we do not understand how the output was derived. And yet AI in healthcare is valued because it can process millions of data points in ways and at speeds beyond the human brain’s capacity.