AI is strong in games but weak in the physical world, one post argues
The post urges caution in robotics, arguing that cognitive AI techniques were pushed forward without solving alignment—getting AI to behave as intended.
TLDR
One post argues that AI techniques are strong in constrained domains such as chess, Go and math, where games allow exact simulation and math allows formal verification. It contrasts those strengths with what it describes as weak performance in the open-ended physical world. The author says cognitive AI techniques were advanced without solving alignment and urges extreme caution about repeating that approach in robotics.
Combined views
37.1K
5 Sources, first seen 19d ago