Despite widespread public anxiety surrounding artificial intelligence (AI), its role in scientific research and medicine is viewed with a notable degree of optimism. A recent NBC News/Decision Desk poll indicated that 70 percent of adults express more worry than excitement about AI in general.
However, Dr. Dhruv Khullar, a physician at Weill Cornell Medicine and writer for The New Yorker, suggests this optimism is well-founded, particularly within the healthcare sector. AI tools are already integrated into Khullar's practice, aiding in tasks such as note-taking during patient encounters, allowing physicians to engage more directly with patients rather than focusing on computer input. This shift is seen as a significant improvement, especially given the current challenges within the healthcare system, which is often criticized for being unaffordable, inaccessible, inconvenient, and of uneven quality.
AI's impact on healthcare can be broadly categorized. The most immediate and widely adopted application is on the administrative side. AI scribes are increasingly present in doctor's offices, enhancing the patient-physician interaction. Another key area is patient navigation, where AI can help individuals, particularly those with serious diagnoses like cancer or heart failure, manage appointments, understand medication instructions, and move more seamlessly through the complex healthcare system.
Accelerating Drug Discovery
Khullar highlights drug discovery as a particularly exciting frontier for AI. While acknowledging that AI has not yet cured diseases like cancer, it is significantly accelerating the early stages of identifying potential drug candidates. AI can process vast amounts of data to identify molecules with potential biological applications and pinpoint biological targets involved in diseases. Previously, this process could take researchers years; AI can now identify promising targets much more rapidly.
Furthermore, AI assists in designing molecules that could interact with these targets and helps in the lead optimization phase. This involves predicting which molecules are most likely to be absorbed by the body, reach the correct tissues, and possess the necessary properties for a safe and effective drug, a process Khullar likens to finding the 'Goldilocks' set of characteristics.
AI as a Clinical Support Tool
Beyond drug discovery, AI is emerging as a valuable tool for physicians in clinical decision-making. It can serve as a sophisticated second opinion, providing rapid access to information on recent trials or potential diagnostic considerations that a physician might otherwise spend considerable time researching or seeking consults for. This clinical decision support is not intended to replace human judgment but to augment it.
Khullar addresses concerns about the potential for AI to be incorrect, noting that the output of these models requires careful adjudication. He emphasizes that AI is not replacing scientists or doctors but is rather an assistant that requires human oversight, critical thinking, and contextual understanding. Issues such as the manufacturability of AI-proposed drugs and their potential toxicity in ways not predicted by the models still necessitate extensive wet lab research and clinical trials.
Navigating the Future of AI in Healthcare
While the initial stages of identifying drug candidates are being dramatically accelerated, Khullar points out that the latter half of the process—proving efficacy and safety in human biology—will likely remain a more analog, human-driven endeavor for some time. The key challenge lies in maintaining human agency and critical thinking skills as AI becomes more sophisticated and integrated into healthcare practices.
For those advocating for a halt to AI development due to perceived risks, Khullar counters that healthcare is ripe for disruption, and even older AI models can offer substantial improvements. He argues against shutting down AI entirely, particularly in areas like biotechnology, scientific research, and clinical care delivery, suggesting that the benefits far outweigh the risks when implemented thoughtfully.