Cooperative Inverse Reinforcement Learning (CIRL) — inverts the legibility direction of → Reward autoencoder (goal transparency)

Both proposals appear back to back under "Multi-Agent Approaches" as ways of making agent-human interaction safer through mutual legibility rather than a penalty term (concrete-problems, §"Multi-Agent Approaches:", p. 6), but they run the legibility in opposite directions. CIRL makes the human's goal legible to the agent: the agent infers what the human wants and acts on that inferred goal instead of a fixed objective. The reward autoencoder makes the agent's goal legible to the human: it treats the agent's actions as an implicit encoding of its reward function and applies autoencoding so "an external observer can easily infer what the agent is trying to do," on the premise that actions with heavy side effects are harder to decode back to a clean goal. One proposal asks the agent to understand us; the other asks us to understand the agent.