Emergent tool use in RL — canonical example.
mediumAnswer
- Baker et al. 2019 (OpenAI): hide-and-seek multi-agent RL discovered emergent tool use — hiders build ramps, seekers box surf, hiders lock ramps, seekers exploit physics glitches.
- Six curriculum stages emerged without engineering.
- Demonstrated that open-ended MARL environments produce increasingly sophisticated strategies.
- Foundation of open-ended learning + emergent capabilities.
Check yourself — multiple choice
- Random
- OpenAI hide-and-seek: 6 emergent strategy stages (ramps → surfing → locking → glitch exploits) — MARL produces increasingly sophisticated tool use without engineering
- Same as OpenAI Five
- Not real
Emergent tool use: hide-and-seek MARL produces increasingly sophisticated strategies.
#multi-agent#exploration
Practise Reinforcement Learning
214 interview questions in this topic.