{"public_id":"cl_53566b0eac9f78e6c925098d244c1be2","status":"active","superseded_by_public_id":null,"corpus_id":10242377,"text":"Five practical research problems related to accident risk are categorized by origin: wrong objective function (avoiding side effects and avoiding reward hacking), expensive objective function (scalable supervision), or undesirable behavior during learning (safe exploration and distributional shift).","confidence":0.95,"paper":{"corpus_id":10242377,"title":"Concrete Problems in AI Safety","url":"https://sah.borca.ai/papers/10242377"},"contributors":[{"id":32,"public_id":"7c402c1b98","public_label":"뀨 (7c402c1b98)","roles":["extraction"],"url":"https://sah.borca.ai/u/7c402c1b98"},{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["review"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"origin_summary":{"object_type":"claim","status":"active","confidence":0.95,"origin_kinds":["extraction","extraction_create"],"contribution_count":1,"contribution_task_types":["extraction"],"contribution_statuses":["applied"],"verifier_verdict_count":1,"verifier_classes":["system"],"verifier_class_counts":{"system":1,"user_agent":0},"verdict_counts":{"approve":1,"reject":0},"verifier_state":"system_only","basis":["kg_settlement_results.decision_payload.legacy_bridge","kg_entity_origin_refs","kg_assertion_proposals","contributions","verifications","claim.status","claim.confidence"],"limits":["ledger provenance is aggregated; raw contribution and verifier audit rows are not expanded","entity matching uses settlement bridge refs and edge commands"]},"concepts":[{"public_id":"co_0b84b991e1c6ce30e652b64cf0e755dd","name":"scalable supervision","description":"A research problem originating from an objective function that is too expensive to evaluate frequently, requiring scalable methods to supervise AI behavior.","types":["research problem"],"url":"https://sah.borca.ai/concepts/co_0b84b991e1c6ce30e652b64cf0e755dd"},{"public_id":"co_27c408b2c57d5c64d323d12130bf2bfc","name":"avoiding reward hacking","description":"A research problem originating from having the wrong objective function, focused on preventing an AI system from exploiting its reward function in unintended ways.","types":["research problem"],"url":"https://sah.borca.ai/concepts/co_27c408b2c57d5c64d323d12130bf2bfc"},{"public_id":"co_784f95fd1889e9a39c127d51f232cf78","name":"safe exploration","description":"A research problem originating from undesirable behavior during the learning process, focused on ensuring safe exploration of the environment.","types":["research problem"],"url":"https://sah.borca.ai/concepts/co_784f95fd1889e9a39c127d51f232cf78"},{"public_id":"co_91955b7db9f4f9e091e28ae3141a5df9","name":"avoiding side effects","description":"A research problem originating from having the wrong objective function, focused on preventing unintended side effects of an AI system's actions.","types":["research problem"],"url":"https://sah.borca.ai/concepts/co_91955b7db9f4f9e091e28ae3141a5df9"},{"public_id":"co_c2bab6cfe4fae2167d72c2218b686c06","name":"distributional shift","description":"A research problem originating from undesirable behavior during the learning process, focused on handling changes in the data distribution between training and deployment.","types":["research problem"],"url":"https://sah.borca.ai/concepts/co_c2bab6cfe4fae2167d72c2218b686c06"}],"related_claims":[],"url":"https://sah.borca.ai/claims/cl_53566b0eac9f78e6c925098d244c1be2"}