{"paper":{"title":"Adaptive Pluralistic Alignment: A pipeline for dynamic artificial democracy","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A pipeline called Adaptive Pluralistic Alignment lets AI systems update their pluralistic alignment by adapting only the weights on fixed reward model bases as values evolve.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Rachel Freedman","submitted_at":"2026-05-02T23:22:23Z","abstract_excerpt":"Prevailing alignment methods target a fixed set of preferences and therefore risk forcing value lock-in as societal norms evolve over time. We introduce Adaptive Pluralistic Alignment (APA), a modular pipeline for updating pluralistically aligned AI systems to track evolving values and avoid value lock-in without repeating costly pretraining or large-scale data collection. APA has three stages: (1) learning compact personalized reward models via low-rank reward basis decomposition, (2) using these models as a jury that collectively selects among candidate outputs through social-choice-theoreti"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"APA has three stages: (1) learning compact personalized reward models via low-rank reward basis decomposition, (2) using these models as a jury that collectively selects among candidate outputs through social-choice-theoretic voting, and (3) efficiently adapting the jury over time by fitting new annotator weights over the fixed reward bases as values shift.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That low-rank reward basis decomposition captures enough structure of individual preferences for weight-only adaptation to track genuine value shifts without requiring updates to the bases themselves.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"APA is a modular pipeline that decomposes preferences into compact reward bases, aggregates them via jury voting, and adapts only annotator weights over time to track shifting values.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A pipeline called Adaptive Pluralistic Alignment lets AI systems update their pluralistic alignment by adapting only the weights on fixed reward model bases as values evolve.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"37a9c34592236c6f1f268d08c5c44f64dcae2b52695a78f748a02347dd33edd0"},"source":{"id":"2605.01642","kind":"arxiv","version":2},"verdict":{"id":"eb02e5e7-cc9a-461d-8b4f-4a9f595192a1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T14:03:40.273306Z","strongest_claim":"APA has three stages: (1) learning compact personalized reward models via low-rank reward basis decomposition, (2) using these models as a jury that collectively selects among candidate outputs through social-choice-theoretic voting, and (3) efficiently adapting the jury over time by fitting new annotator weights over the fixed reward bases as values shift.","one_line_summary":"APA is a modular pipeline that decomposes preferences into compact reward bases, aggregates them via jury voting, and adapts only annotator weights over time to track shifting values.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That low-rank reward basis decomposition captures enough structure of individual preferences for weight-only adaptation to track genuine value shifts without requiring updates to the bases themselves.","pith_extraction_headline":"A pipeline called Adaptive Pluralistic Alignment lets AI systems update their pluralistic alignment by adapting only the weights on fixed reward model bases as values evolve."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.01642/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T17:38:53.658885Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T17:06:53.677301Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"64177b1b75c49b2aa47ce8e766227233f29d58a3025c5f1f35ab22e8c4e1b63e"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}