4 ms·
What's fascinating (and admittedly somewhat worrying) about Self-play is that an agent can accidentally become adept at tasks other than intended via transfer l
by indescions_2017 9y ago
What's fascinating (and admittedly somewhat worrying) about Self-play is that an agent can accidentally become adept at tasks other than intended via transfer learning. The "wrestling spiders" in OpenAI's demo quickly mastered the art of Sumo Wrestling. And whatever skills they learned in resisting an opposing force to stay standing on a platform, were immediately applicable to myriad different domains. In this case, being subject to hurricane force winds, and not as any normal spider may, be hurled into the sky!
It's more difficult to see how Go playing skills can translate to other domains. But for tasks in robotics, cybersecurity or fintech the power of self-play trained transfer learning becomes more apparent.
- visarga 9y agoIt is clear that these "self-play" scenarios depend on simulation - unless there is an appropriate stage for self play to take place on, there can be no play. The question is - how do we stand with simulation for robotics, self driving cars, etc. My bet is that simulation is going to be the crowning jewel in the AI field, replacing static datasets and supervised learning with "dynamic datasets" and rewards. It would help with data sparsity as well (where can you find an image of a donkey riding an elephant for the new ImageNet? - but you can sim that or any possible combination). Not to mention that humans are fallen head over heels with simulation as well - VR headsets and games in general. I see a great future for simulation with both AI and humans. It will be our common learning/playing/research sandbox.