Theodore Schwartz: evaluation

Transcript context

…As you know, for ordinary medical diagnosis — and I do mean ordinary — even GPT-4 does slightly better than human doctors, on average. Are there parts of neurosurgery — and I don’t mean moving around your hand and making decisions on the spot — but where you think an AI model soon will just do better than the humans? I think that in the interpretation of MRI scans — when a patient comes into your office with a conglomerate of symptoms, and you’ve got to figure out what they have — those things in AI are very good at. But often, neurosurgery is different than internal medicine. We’re usually not presented with diagnostic dilemmas or radiographic dilemmas. The dilemmas that we face are physical decision-making in the operating room. That’s really what neurosurgery is all about. Remember, a patient will come to us — they already have an MRI scan in their hand. They’re coming in, and they’re saying, “I know I have a tumor. I know where it’s located. What should I do about it?” The other thing that’s tricky about having an AI make a decision — let’s say you have a tumor that could be observed, it could be radiated, it could be operated upon. So, you have three options. What’s going to factor into whether to do the surgery or not is how good the surgeon that you’re seeing is at doing that surgery. Not all surgeons are created equal. The AI will not necessarily be able to take into account the skill level of the particular surgeon that you’re seeing because that’s unique to that one surgeon and that one surgeon alone. But we could measure that, right, if the surgeons would allow it?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.