The year's most important article for AI in biotech isn't any of the AlphaFold ones, it's this one by Matthew Herper. The inaccurate one phrase summary: The bottleneck to finding new treatments isn't basic research, it's clinical trials.

To expand a bit, we don't know how to make useful clinical trials for new treatments for complex diseases that aren't horrendously slow and expensive. You can make them cheaper by making them less useful (either unsafe or uninformative) or by aiming at simpler diseases or lower thresholds of improvement, but we are very very bad at testing the kinds of new treatments we want to have, and this might be the factor slowing down medical applications (that is, why all those exciting headlines never quite become things your doctor can use).

The bad news is that this isn't a sexy AI problem - it's not the sort of thing you can come out of left-field with no medical or organizational know-how, throw a new training idea and heaps of computer power to and push the frontier forward. It's messy, it's slow, you need to work with and for the people already on the field. The good news is that this is a great AI problem: it's a hellishly complex inference problem at its core, and we're getting very good at those when we bother to attack them.

And, of course, it's one of those rare trillion-net-present-value-plus-change-the-world-for-the-better problems.

Riffing on something I wrote and talked about elsewhere, a recession — and global tech is entering one — is the perfect moment to start working on disruptive technologies, because nobody expects hyperfast growth right away, so you have the time to build something truly new. Large-scale superhuman clinical trials is one of the perfect areas for this: get the technology working right (which at this stage isn't about building laboratories but about putting together the cutting edge of AI and inference with the cutting edge of clinical trial design and operation - it's discipline engineering, not biological engineering), be ready to show concrete improvements as soon as the VC money starts flowing, and you'll have a hugely valuable thing on your hands.