Accounting for Human Learning when Inferring Human Preferences
This addresses the challenge of accurately inferring preferences for AI systems interacting with humans in unfamiliar environments, though it is incremental as it builds on standard IRL techniques.
The paper tackles the problem of inferring human preferences using inverse reinforcement learning (IRL) by relaxing the assumption that human demonstrators are stationary, instead modeling them as learning over time. The result shows that in small examples, this approach can lead to better inference than assuming stationarity, with evidence that misspecification can cause poor inference.
Inverse reinforcement learning (IRL) is a common technique for inferring human preferences from data. Standard IRL techniques tend to assume that the human demonstrator is stationary, that is that their policy $π$ doesn't change over time. In practice, humans interacting with a novel environment or performing well on a novel task will change their demonstrations as they learn more about the environment or task. We investigate the consequences of relaxing this assumption of stationarity, in particular by modelling the human as learning. Surprisingly, we find in some small examples that this can lead to better inference than if the human was stationary. That is, by observing a demonstrator who is themselves learning, a machine can infer more than by observing a demonstrator who is noisily rational. In addition, we find evidence that misspecification can lead to poor inference, suggesting that modelling human learning is important, especially when the human is facing an unfamiliar environment.