Cooperative Inverse Reinforcement Learning (CIRL) — interactive counterpart of → Use demonstrations (safe exploration)

CIRL is grouped with the demonstration-based remedies because it shares their goal of letting a human's intentions substitute for a hand-specified reward, but the mechanism is different in kind, not degree. Use Demonstrations (via inverse RL or apprenticeship learning) works from a fixed, pre-collected dataset of expert trajectories, recorded once and then mined offline (concrete-problems, §"Use Demonstrations:", p. 15). CIRL instead sets up an ongoing two-player game in which the human and the agent act simultaneously, and the agent updates its belief about the human's goal from the human's behavior turn by turn (concrete-problems, §"Multi-Agent Approaches:", p. 6). This buys CIRL agents the ability to ask clarifying questions or defer to the human, something that is structurally unavailable to an agent trained once on a fixed demonstration set.