@fogti Why do you think that? A lot of medicine is about finding a needle in a haystack. A great many of the discoveries so far have been dumb luck. The systems involved are so complicated that humans can barely even spot patterns in them, much less understand how they work. AI can't "understand" biological systems either, but what it can do is to help humans spot patterns.

And that's the difference between AI in medicine and AI in consumer applications: in medicine, AI is typically used to augment humans, not to replace them. As overhyped as they are, even an LLM can work well in that scenario, because it's essentially a search engine. Sure, it might produce a wrong answer, but an expert will confirm it. And it might come up with a right answer a human is unlikely to have thought of. That doesn't have to happen very often to save a lot of lives. And when it comes to drug discovery, the leverage can be huge.

And speaking of drug discovery, I just discovered that the first drug created entirely by AI is entering human trials. https://www.cnbc.com/2023/06/29/ai-generated-drug-begins-clinical-trials-in-human-patients.html

It seems like that's how AI has always been: you have the overhyped bleeding edge stuff that never lives up to the hype, and then you have the slow boring work that by the time it produces something is almost never called "AI" even though it involves neural networks. For example, who calls speech synthesizers "AI" even though all the current state of the art synthesizers use neural networks from end to end? People barely even paid attention to that research, too. Speech synthesizers just magically got better as far as they were concerned.