Definition
Plain language
Teaching an AI to copy recorded expert actions, one situation at a time.
As stated in the literature
An imitation-learning method that trains a policy via supervised next-action prediction over expert demonstrations; prone to compounding error when the agent visits states absent from the demonstration distribution, since flawless experts never demonstrate recovery from mistakes.
Why it matters: It's a simple way to bootstrap skills from demonstrations, but it leaves an agent helpless in situations the expert never stumbled into.
For example, an AI driving assistant trained only on recordings of smooth expert driving has never seen how to recover once it drifts toward the curb.
Heard on the show
“The training method here is called behavior cloning — the agent watches recordings of an expert and learns to predict the next action from each screen.”Episode 155 — Why a Flawless Demo Makes a Worse Computer-Using Agent, And the Fix