From ChatGPT(4):

Inverse Reinforcement Learning (IRL) is a technique in the field of machine learning where an AI system learns the preferences or objectives of an agent, typically a human, by observing their behavior. Unlike traditional Reinforcement Learning (RL), where an agent learns to optimize its actions based on given reward functions, IRL works by inferring the underlying reward function from the demonstrated behavior.

In other words, IRL aims to understand the motivations and goals of an agent by examining their actions in various situations. Once the AI system has learned the inferred reward function, it can then use this information to make decisions that align with the preferences or objectives of the observed agent.

IRL is particularly relevant in the context of AI alignment, as it provides a potential approach to align AI systems with human values. By learning from human demonstrations, AI systems can be designed to better understand and respect the preferences, intentions, and values of the humans they interact with or serve.

(Cunningham law this if you please, it was empty when I came across it and I thought something better than nothing.)