UNDER THE MICROSCOPE
Artificial Intelligence (AI) hugged the headlines recently with Pope Francis warning: “Yet at the same time, it (AI) could bring with it greater injustice between advanced and developing nations or between dominant and oppressed social classes.”
He further insists that AI be human-centric, which is also applicable to its uses in medicine. AI has tremendous potential in medicine. It can be used to analyze medical and health data to provide insights and improve health outcomes and patient experiences.
It can diagnose diseases, develop personalized treatment plans and assist clinicians in decision-making. AI can predict an individual’s risk of certain diseases and suggest preventative measures. In public health care, it can reduce waiting times and improve efficiency in healthcare systems.
But it also requires human supervision that can and should override AI-driven decisions should they be wrong, which is also possible, nay, will happen since AI is driven by past human knowledge that may be erroneous in the first place. We have seen this in times past when past research concluded that coconut oil, which is a medium chain fatty acid, was deemed to be bad for the cardiovascular system. Current knowledge now states otherwise.
Medicine is always an evolving science. We need to keep adding more recent and validated data to for machine learning in order for AI to come up with the real diagnosis and treat accordingly. Humans, in the form of specialists in their particular medical fields, must exercise ultimate control and not let AI run amuck in life or death decisions.
Currently, AI is being used to interpret images in radiology (x-rays, ultrasound, CT-scans, PET scans), histopathology and cytology in the laboratory and fundoscopy (examination of the interior of the eye). This straddles several medical disciplines-radiology, pathology and ophthalmology. I’ve written about the use of AI for public health purposes (A New Approach to Cervical Cancer Screening, March 12, 2024).
AI has beneficial applications for public health. It can be used to screen chest x-rays at a much faster speed than humans can, and that can free up radiologists from very routinary work so they can concentrate on more complex tasks that evade AI at the moment. The same can be said of cervical cancer screening, which involves a tedious and time-consuming search for a few cancer cells in Pap smears. Actually, AI can not only do it faster but also see cancer cells that may evade fatigued human vision. Of course, it can only flag abnormal smears which are then reviewed by pathologists who will sign out the cases.
Other applications of AI are in medical transcription, where it can translate human voices in dictation for incorporation into patient medical records. AI can synthesize data from patient interviews and laboratory tests to write notes directly, even offer diagnoses and in difficult cases, consider other (differential) diagnoses.
In medical research, AI can be used to mine the voluminous databases with greater efficiency and accuracy than humans and it can render insights that are difficult to attain with more traditional data-analysis methods.
However, AI is not a panacea. It can be brittle, work in a narrow domain, and have built-in biases that disproportionately affect marginalized groups. Expertise in AI is also closely linked to commercial applications, which may lead to conflicts of interest. Large scale AI models require resources that big companies possess for them to be at the forefront of AI systems research and development. That comes at a cost for the eventual application of AI and restricts access to a few, producing monopolies that can charge large sums for its use.
For AI use in public health, cost is the overriding concern. The cost for a chest x-ray or Pap smear should be at a minimum for it to be employed in large-scale screening and early detection of cancers and tuberculosis, both public health problems of highest concern. AI use in generating diagnoses, medical records and health insurance coverage, assisting caregivers in making claims and payors in adjudicating them, can result in more efficient use of doctors and nurses who are often overwhelmed by voluminous paperwork and medical record documentation. Considering we have an acute shortage of both, this will go a long way in providing better health care in the public sector.
For AI in health applications to be beneficial to all, it will have to take into account human values. Inputs into AI have to be scrubbed clean of biases that will affect its output. The choice of studies to input into AI programs matters because if these are race-specific, for example, more harm may come to patients of other races.
Pope Francis scored an important point for developing AI for the benefit of all mankind, not just a few rich nations that are on the forefront of AI development. Here’s hoping those developers are listening.