I conjecture that this agent could be improved by modelling human behaviour with *imprecise probability*, & training with adversarial optimization of the imprecise component. (Among other…
Posters: davidad. davidad
3 likes recorded by the source corpus.
Topics: AI safety