The Importance of Instruction Tuning for Language Models in Medicine

Instead of fretting about AI replacing jobs humans currently do, Dr. Nigam Shah is urging people to adopt a perspective about the technology that echoes President John Kennedy’s famous charge in his inaugural address: ‘ask not what this technology can do to you, ask what you can do with this technology.’ “If medicine simply automated everything we were doing 200 years ago, we’d have a machine that would do bloodletting. But we didn't fall into that trap,” says Shah, the chief data scientist at Stanford University. Instead, he suggests, people in the healthcare arena should think about what a human and a computer can do together that neither of them could do alone. In this fascinating episode of Raise the Line with host Shiv Gaglani, Shah also issues a call to action to the medical community about training AI for medical purposes. “If you really want to use these things, we have to create the instruction-tuning data so that they produce the output that we expect.” As for predictions of AI being the author of our salvation or doom, count him as skeptical. “I'm quite sure both sides are overblowing it for different reasons, and the truth will land somewhere in the middle. We’ve got to proactively pick the amazing and stay away from all the fearmongering.” There is much to be learned in this engaging conversation about the history of AI hype cycles, how to use AI to maximize productivity and the challenges inherent in AI-human interaction. 

If you like this podcast, please share it on your social channels. You can also subscribe to the series and check out all of our episodes at www.osmosis.org/raisethelinepodcast