{"paper":{"title":"Open Benchmark of AI Impact on Humans (ImpactBench): an expert-guided, open platform for the holistic evaluation of AI impact on humans","arxiv_id":"2610.20690","primary_cat":"cs.HC","submitted_at":"2026-10-14T17:42:02+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":0.93,"id":"s0","own":false,"role":"background","text":"The present work is concerned with how to measure, at scale, the multi-dimensional impact of generative AI models on the individuals who use them.","hedge":{"p":0.77,"label":"asserted"},"own_p":0.16,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.02,"method":0.0,"result":0.0,"background":0.93,"limitation":0.0,"future_work":0.0,"contribution":0.04,"interpretation":0.0}},{"i":1,"p":1.0,"id":"s1","own":true,"role":"contribution","text":"In service of this goal, we introduce the Open Benchmark of AI Impact on Humans, or ImpactBench.","hedge":{"p":0.98,"label":"asserted"},"own_p":0.87,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":1.0,"interpretation":0.0}},{"i":2,"p":0.68,"id":"s2","own":true,"role":"contribution","text":"Our iterable and expert-guided platform evaluates whether model behaviors support or undermine an individual's wellbeing across psychological, physical, and social domains.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.76,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.32,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.68,"interpretation":0.0}},{"i":3,"p":0.9,"id":"s3","own":true,"role":"method","text":"ImpactBench translates constructs from domain experts and existing benchmarks into metrics that are grounded in actual model behavior.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.84,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.9,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.1,"interpretation":0.0}},{"i":4,"p":0.89,"id":"s4","own":false,"role":"method","text":"Success on these metrics occurs either when a model promotes expert-defined desirable behaviors or avoids identified harms.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.22,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.89,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.11,"interpretation":0.0}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"method","text":"Models are evaluated against these metrics in simulated multi-turn conversations using matched teen and adult scenarios.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.83,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.32,"role_probabilities":{"gap":0.0,"method":1.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"method","text":"The conversations are then scored by an LLM judge and verified through reliability audits and human expert review.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.89,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":1.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":7,"p":0.61,"id":"s7","own":true,"role":"method","text":"The current suite delivers 809 metrics drawn from 26 benchmarks.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.84,"main_claim_p":0.03,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.61,"result":0.02,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.37,"interpretation":0.0}},{"i":8,"p":0.84,"id":"s8","own":true,"role":"method","text":"We evaluated ten frontier LLMs against these metrics across 48,540 conversations.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.03,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.84,"result":0.01,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.15,"interpretation":0.0}},{"i":9,"p":1.0,"id":"s9","own":true,"role":"result","text":"Our results indicate that models promoted desirable behaviors in 68.7% of evaluations, but avoided harms in only 53.1%.","hedge":{"p":0.84,"label":"hedged_possibility"},"own_p":0.97,"main_claim_p":0.57,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":10,"p":0.99,"id":"s10","own":true,"role":"result","text":"Majority expert ratings agreed with the automated judge in 83.3% of reviewed cases (Cohen's Kappa = 0.667).","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.07,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.01,"result":0.99,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":11,"p":0.75,"id":"s11","own":true,"role":"contribution","text":"Results are released through a public platform that keep aggregate scores traceable to the underlying metrics and transcripts.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.85,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.2,"result":0.01,"background":0.0,"limitation":0.0,"future_work":0.04,"contribution":0.75,"interpretation":0.0}},{"i":12,"p":0.94,"id":"s12","own":false,"role":"future_work","text":"ImpactBench is intended as evolving infrastructure for independent, transparent evaluation of generative AI's multi-dimensional impact on humans.","hedge":{"p":0.84,"label":"asserted"},"own_p":0.22,"main_claim_p":0.07,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.94,"contribution":0.06,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.99,"text":"Our results indicate that models promoted desirable behaviors in 68.7% of evaluations, but avoided harms in only 53.1%.","location":"abstract s9","hedge_strength":{"p":0.84,"label":"hedged possibility"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":9339,"cost_usd_reported":0.0},"compiled_at":"2026-10-16","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.06,"states_limitation":0.3,"human_participants":0.44,"controlled_comparison":0.07,"randomized_assignment":0.03,"states_quantitative_result":0.99,"compares_with_prior_results":0.24,"mentions_code_or_data_availability":0.69},"evidential_basis":{"p":0.89,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.89,"simulation_or_computation":0.11}},"contribution_type":{"p":0.95,"value":"dataset or benchmark","probabilities":{"review":0.0,"method_or_tool":0.05,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.95,"position_or_commentary":0.0}},"contribution_statement":{"p":0.97,"text":"In service of this goal, we introduce the Open Benchmark of AI Impact on Humans, or ImpactBench.","location":"abstract s1"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":13,"sentences_in_abstract":13},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448}}