Artificial intelligence (AI) is poised to revolutionize many fields of science and technology. One field that stands to benefit significantly is drug discovery, which is a time-consuming and expensive process. AI can predict compounds and some of their relevant characteristics, including their efficacy and toxicity. In doing so, AI can help refine the pool of potential compounds that progress in the drug discovery pipeline, while excluding those that will later likely prove to be too toxic or ineffective (Tran et al. 2023). Essentially, AI can make the early stages of drug discovery more efficient by helping to avoid unnecessary human clinical trials and prevent costly, late-stage failures (Tran et al. 2023). Yet, as drug design AI capabilities burgeon, so does the concern that these algorithms could be used for malicious purposes, such as harnessing AI to instead predict compounds that are both highly effective and highly toxic, posing biosecurity risks. Although concerns about the dual-use potential of AI are warranted, there is great potential for AI’s beneficial application in drug discovery, so eliminating the use of AI in this space altogether is undesirable. We recommend that the Food and Drug Administration (FDA) place a special call for submissions of drug design AI with safeguards in place to prevent dual-use to its Innovative Science and Technology Approaches for New Drugs (ISTAND) Pilot Program. This would allow the FDA to open up a line of communication with drug design AI creators, educate the broader public on the potential for dual-use of these technologies and emphasize the need for safeguards, and select a drug design AI that models responsible AI applications for the field at large.
Read full abstract