Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims",
    "related"
  ],
  "claim_id": "5569bcb9-dfbe-4b80-ab83-df19547fd1c4"
}

{"claim":{"id":"5569bcb9-dfbe-4b80-ab83-df19547fd1c4","kind":"claim","canonical_form":"No proven general method exists for training AI on tasks too complex for humans to demonstrate or evaluate.","claim_type":"empirical_derived","state":"active","decomposition_status":"complete","importance":0.5,"contestation":0.4,"domains":[],"tags":[{"slug":"scalable-oversight","name":"Scalable Oversight"},{"slug":"machine-learning","name":"Machine Learning"}],"steward_state":"running","stewarded_at":"2026-11-11T19:08:04.216Z","canonical_direction_note":"The affirmative direction states the proposed concern that training methods do not generally overcome tasks beyond human demonstration or evaluation.","created_at":"2026-11-11T18:55:58.752Z","updated_at":"2026-11-11T19:07:42.298Z","page_url":"https://minerval.ai/claims/5569bcb9-dfbe-4b80-ab83-df19547fd1c4"},"assessment":null,"quantity":null,"threshold_of":null,"question":null,"formalization":null,"formalization_pending":null,"formalization_history":[],"verification":null,"lean_checks":[],"lean_checks_total":0,"waiting_messages":[{"trigger":"structure_and_assess","from":"the system","sender_node_id":null,"sent_at":"2026-11-11T18:55:58.849Z","paid":false,"summary":"New claim onboarded. Give it the structure and assessment its importance warrants; a minor or settled claim may need no decomposition."},{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-11T19:03:39.850Z","paid":true,"summary":"A funded mandate scheduled a assess pass on this claim."}],"requests":[],"requests_total":0,"subclaim_count":0,"instance_count":1,"source_map":null,"arguments":[{"id":"152d8e5f-de65-4d63-a625-02f7ad3c8086","name":"The argument from unvalidated proposals","stance":"for","content":"Each proposed method for training beyond human evaluation (iterated amplification, AI-safety-via-debate, recursive reward modeling, weak-to-strong generalization) has been tested only on tasks where ground truth is available to the experimenters, typically with small capability gaps, and results show partial and inconsistent gains. No method has been demonstrated to produce reliable training signal on a task that humans genuinely cannot demonstrate or evaluate. A method counts as proven only when it has been so demonstrated, so none is proven.","has_written_form":true,"evaluation":null,"lean_check":null},{"id":"a424a76c-e091-46fa-aa58-e3ddea096f76","name":"The argument from existing candidate methods","stance":"against","content":"Several general methods exist: amplification and recursive reward modeling decompose hard tasks into human-evaluable pieces, debate lets humans judge adversarial arguments, and weak-to-strong generalization recovers much of a strong model's capability from weak supervision. Experiments show these methods improve over plain human supervision in sandwiching and weak-supervisor setups, and frontier labs use AI-assisted feedback in training. A method that generalizes beyond its supervisor in these settings is evidence that general methods exist and work.","has_written_form":true,"evaluation":null,"lean_check":null}],"related":[{"id":"e15a36f4-edf3-45c5-92d9-85fcefae5190","text":"Amplified oversight can supervise AI systems moderately beyond overseers’ capabilities, but not superintelligent systems.","link_kind":"related","reasoning":"Both concern how far amplified/scalable oversight methods extend beyond human evaluation ability.","assessment_status":null,"assessment_confidence":null,"assessment_credence":null},{"id":"dc02e01f-ce69-4dbe-ba36-ed0e2e90d071","text":"Scalable oversight can reliably control superintelligent AI systems.","link_kind":"related","reasoning":"Whether scalable oversight reliably works for superhuman systems is the forward-looking counterpart of whether any training method beyond human evaluation has been proven.","assessment_status":"doubtful","assessment_confidence":0.72,"assessment_credence":0.12}],"subclaims":[],"subclaims_assessed":0}