Observational Learning by Reinforcement Learning

The man at risk of being stung is actively trying to prevent it. Blindly mimicking those movements, when not at risk of being stung, wouldn't have an effect, positively or negatively, on the ability to avoid stings. But if the man slammed his head into a post trying to avoid the wasp, copying would obviously be an issue. As long as the mimic is aware of harmful actions, learned either by experience (should be preferred) or given rules (cargo-culting risk), and stops copying the subject immediately (since it may have malicious intent or need emergency assistance), it should be safe, at least as safe as any of us can hope to be.