I think the debate about AI and medicine is interesting because AI is not new in medicine. It’s been here with us for a long time, and it’s being used right now in a lot of ways, and I think it’s been very useful. I think the promise is huge. I think it can help in many, many different ways. I don’t think it will replace people like physicians and nurses, but I think it can be an enormous help...
I think the debate about AI and medicine is interesting because AI is not new in medicine. It’s been here with us for a long time, and it’s being used right now in a lot of ways, and I think it’s been very useful. I think the promise is huge. I think it can help in many, many different ways. I don’t think it will replace people like physicians and nurses, but I think it can be an enormous help. The problem is that there are some drawbacks and some concerns that are very real, that need to be addressed first.
There are really two areas where our concerns are pretty important and need to be addressed before it’s ready. Of course, I’ve been assigned to say that it’s not ready. In some ways it is ready, but in others it’s not. That tends to revolve around the idea that when AI is used in a sort of replacement fashion for trained physicians and other clinicians, it falls short because it can make errors. Errors based on the kinds of data that it’s been trained with, errors because of just the way AI works. And, you know, it’s an interesting concept, errors. Guess what? We physicians make errors all the time. And so, do we have a zero tolerance for machines making errors when we tolerate plenty of errors ourselves? That’s something that’s sort of a side question. In San Francisco, where I live, we have self-driving cars. I don’t know if you have them in London or other places, but we do, and they’re pretty good. But when they cause a bruise in one person in San Francisco, it’s headlines. But when person-driven cars hurt people every day, no one seems to be alarmed by that. So, it’s an interesting question, but there are some errors that AI makes, and we have to address how we’re going to think about those, how we’re going to prevent them. And I have a lot of ideas, and I would suggest that maybe you attend the debate, see what you think.
Another one is the idea of confidentiality. When personal information, medical information, gets into a system, an artificial intelligence or really any electronic system, it’s there. And the question is how that can be protected. And we do a pretty good job of that right now, even though AI is part of most of our medical systems. But we have to be careful because of this concept of the black box, and I bet you’ve heard about this. AI is developed by many, many people and many departments in certain corporations, and pretty soon in the development, no one quite knows how it got developed and how it functions and what its drawbacks are. And so it’s immense, these models are huge programs. So once that information is there, how is it going to be controlled and where does it go? That’s the question. It’s like anything else, I think it’s going to have to be deployed before we really know the dangers. But that’s the other danger people talk about in addition to the error issue.
