VC investor suggests AI will lower pharmaceutical costs but some critics dismiss its use in drug discovery as ‘hype’
The pandemic fast-tracked technological deployment in public health and medicine. Everything from contact-tracing apps to services offering remote doctor appointments. But nowhere has the role of technology — specifically artificial intelligence — been more hotly debated than in the field of drug discovery. Advocates point to the pandemic as proof of its value in healthcare even as critics dismiss its use in drug discovery as “hype”, reports Hannah Kuchler. In January 2020, scientists at the pharmaceutical company BenevolentAI used artificial intelligence algorithms to trawl through 50mn medical journals to search for approved drugs that could be repurposed to treat the disease. The scientists and the algorithm narrowed down the search to baricitinib, used to treat rheumatoid arthritis — all in a matter of four days. The Eli Lilly drug tackled both the virus and the body’s inflammatory reaction. The event marked the first time AI had discovered a drug, already in widespread use, that could be redeployed. TechFT sat down with Lee Kai-Fu, an investor and author of AI 2041: Ten Visions for Our Future, to talk about the technology’s role in drug discovery.
EO: What role does AI have in drug discovery?
LKF: AI can help speed up drug discovery in three phases of drug development. The first phase uses AI to narrow down the number of drug candidates. Instead of a scientist filtering a drug from 10,000 candidates down, the AI will do that based on inputs fed in by the scientist. This process combines AI with human expertise to pick the candidates with the highest likelihood of success in clinical trials. The second phase uses AI during the experimentation phase, where robots can automate much of the lab work. Robots replicate simple and routine manual procedures such as opening test tubes, mixing liquids, putting in chemical agents, growing the culture, watching the reaction and producing the result. These are routine and replicable procedures for robots, thereby accelerating the second phase of clinical drug development. The third phase is to use AI to accelerate processes in the clinical trial stage, for example, helping pharmaceutical companies to match patients with the clinical trial. The benefit of these three combinations is to bring down the cost of drug discovery, lowering the bar for pharmaceutical companies to develop cures for rare diseases that were not economical to target.
EO: There is a fair amount of scepticism in the medical establishment about the role of AI in medicine. What will convince them of the future you are charting out?
LKF: I’m very aware of a cognisant mismatch between the medical and AI community. There are many AI people with starry-eyed dreams that they could change the world if only their software was adopted everywhere — who think it’s not happening just because of a different way of thinking [in the medical profession]. But that problem will not happen in drug discovery because the interests are aligned. We’re not proposing a new method of drug discovery. In each of these three phases, everything will be done to the precision and satisfaction of humans, within a framework that already exists.
EO: What are the risks of deploying AI in drug discovery?
LKF: There is a chance the same tools used to discover drugs that cure disease could be used to invent toxins to hurt people accurately. One possible mitigation is not making this open-source to stop it from falling into the wrong hands.