News for hackers. Stuff that matters.

A comment on DeepMind says reinforcement learning is ‘enough’ to reach general AI.

Re:DeepMind says reinforcement learning is ‘enough’ to reach... by 21eleven on Thursday June 10, 2021 @10:37AM (#27461271)
RL can provide amazing results (AlphaGo, AlphaStar (Starcraft 2 agent), etc) but it requires a well modeled world to work with.

Games like Go and Starcraft are well modeled worlds. If you want something akin to AGI to operate in the "real world" you will need a high quality data model of the real world for the RL system to work off of.

[ Parent | on HN ]
Re:DeepMind says reinforcement learning is ‘enough’ to reach... by atombum on Thursday June 10, 2021 @11:21AM (#27461750)
Agreed, and that brings up a very interesting discussion of prejudice in modeling the world. Everyone views the world differently and it seems to me of vital importance that any attempts to model the real world for RL are as unbiased as possible. Or more realistically, incorporate as many different biases as possible.
[ Parent | on HN ]