3 ms·
I will say what I’ve said before when this has come up. For declarative languages, this could be good. This is the “centaur” period (human + computer was bette
by EGreg 6y ago
I will say what I’ve said before when this has come up.
For declarative languages, this could be good. This is the “centaur” period (human + computer was better vs just a human or just a computer) that Garry Kasparov touted, before the computer-only part became better. (Chess was much easier to do play against itself adversarially though, harder to simulate how humans would react to your website so you can’t train it MCTS style).
BUT...
For imperative languages with side effects, we have a long way to go. We would need a completely DIFFERENT model of AI, perhaps closer to Cyc than GPT