How Can Crowdsourcing Improve Reward Learning?

Crowdsourcing is a promising way to try to collect annotations for this because it's pretty scalable. And then once you have that, you can use your favorite reinforcement learning algorithms or planning algorithms in order to actually optimize for the behavior that you want. It seems if we could have such a fully general reward function, if this was trained on a very large amount of data, maybe then reward functions could become actually pretty useful. Yeah.