Feedback-Specific Learning and Transfer Limits: Why Skill Learned Under One Kind of Feedback Can Fail Under Another

This conceptual paper introduces Feedback-Specific Learning and Transfer Limits, a structural account of how practice fits behavior to the responses of a training setting. Success in training supports a claim about performance under those conditions; transfer depends on whether the target setting preserves the relevant relationships between action and response. The paper separates competence in a trained domain from demonstrated performance elsewhere. Surface realism does not establish that a simulator preserves the dependencies that matter. Matching those dependencies can support transfer, but actual performance and alternative explanations for failure still require examination. The discussion connects the mechanism to Thorndike and Woodworth’s work on transfer of practice and examines the National Transportation Safety Board’s findings on the 2001 crash of American Airlines Flight 587. The Board identified rudder design and training elements as contributing factors. It found that a simulator exercise could have encouraged a control strategy that would produce a different, surprising response in flight. Part of the Shared Capacity and Learning series within On Human Modeling, the paper examines the scope of skills acquired under particular feedback conditions. Its framework can inform human–AI research on training environments, feedback design, and whether success under one response pattern transfers to another.