Hacker News new | past | comments | ask | show | jobs | submit login

Plus it’s not even defined what superhuman AI means. A calculator sure looked superhuman when it was invented. And it is!

Another analogy is breeding and racial biology which used to be all the hype (including in academia). The fact that humans could create dogs from wolves, looked almost limitless with the right (wrong) glasses. What we didn’t know is that wolf had a ton of genes that played a magic trick where a diversity we couldn’t perceive was there all along, in the genetic material, and it we just helped make it visible. Ie a game of diminishing returns.

Concretely for AI, it has shown us that pattern matching and generation are closely related (well I have a feeling this wasn’t surprising to neuro-scientists). And also that they’re more or less domain agnostic. However, we don’t know whether pattern matching alone is “sufficient”, and if not, what exactly and how hard “the rest” is. Ai to me feels like a person who had a stroke, concussion or some severe brain injury, it can appear impressively able in a local context, but they forgot their name and how they got there. They’re just absent.






Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: