{"id":"learning-strategies-from-trials","title":"Turning outcomes into transferable strategies","regions":["computation","shared-inquiry"],"question":"How can concrete failures and successes become reusable strategies for a new problem?","formulation":"Move between low-level parameters and a space of abstract strategies, using trials to interpret why a path worked and transfer that interpretation to new tasks.","representations":[{"name":"Observed trial outcomes","affords":"Provides concrete successes and failures.","loses":"Correlation with a successful run does not identify its cause.","epistemic_kind":"reconstruction"},{"name":"An abstract strategy inferred from trials","affords":"Can be tested on another task.","loses":"An appealing explanation may transfer incidental features.","epistemic_kind":"reconstruction"}],"tension":"Interpreting success is necessary for transfer and also a source of invented causal stories.","challenge":"Does the inferred strategy predict which altered tasks will fail?","development":"historical_branch","epistemic_kind":"reconstruction","evidence":["basis-learning-strategies-from-trials"],"next_move":{"text":"Hold out task variations before interpreting trial outcomes; compare transfer predictions with results.","epistemic_kind":"model_proposed"},"open_questions":["How can concrete failures and successes become reusable strategies for a new problem?","Does the inferred strategy predict which altered tasks will fail?"],"formal_status":"open_problem","version":1,"content_hash":"9a89d98270f6a6b02e724a0fc4071a5c2ef535254ec0eef5706d3a9992e9677e"}