Defining and Characterizing Reward Hacking
Authors: Joar Skalse, Nikolaus H. R. Howe, Dmitrii Krasheninnikov, David Krueger
Abstract: We provide the first formal definition of reward hacking, a phenomenon where optimizing an imperfect proxy reward function, $\mathcal{\tilde{R}}$, leads to poor performance according to the true reward function, $\mathcal{R}$. We say that a proxy is unhackable if increasing the expected proxy return can never decrease the expected true return. Intuitively, it might be possible to create an unhackable proxy by leaving some terms out of the reward function (making it "narrower") or overlooking fine-grained distinctions between roughly equivalent outcomes, but we show this is usually not the case. A key insight is that the linearity of reward (in state-action visit counts) makes unhackability a very strong condition. In particular, for the set of all stochastic policies, two reward functions can only be unhackable if one of them is constant. We thus turn our attention to deterministic policies and finite sets of stochastic policies, where non-trivial unhackable pairs always exist, and establish necessary and sufficient conditions for the existence of simplifications, an important special case of unhackability. Our results reveal a tension between using reward functions to specify narrow tasks and aligning AI systems with human values.
Explore the paper tree
Click on the tree nodes to be redirected to a given paper and access their summaries and virtual assistant
Look for similar papers (in beta version)
By clicking on the button above, our algorithm will scan all papers in our database to find the closest based on the contents of the full papers and not just on metadata. Please note that it only works for papers that we have generated summaries for and you can rerun it from time to time to get a more accurate result while our database grows.