{"event":{"id":"evt-sparrow-20220922","dedupe_key":"retrieval_augmented_dialogue_agent_research:2022-09-22:deepmind-introduces-sparrow-with-rules-human-feedback-and-evidence-retrieval","event_type":"retrieval_augmented_dialogue_agent_research","title":"DeepMind introduces Sparrow with rules, human feedback and evidence retrieval","summary":"Sparrow combined reinforcement learning from human feedback, explicit natural-language behavior rules and evidence retrieval for factual dialogue.","occurred_at":"2022-09-22T00:00:00.000Z","published_at":"2022-09-22T00:00:00.000Z","observed_at":"2026-10-02T15:12:29.000Z","reconstructed_at":"2026-10-02T15:12:29.000Z","ingest_type":"live_world_scan","locations":null,"status":"verified","confidence":0.97,"metadata":{"actors":["DeepMind"],"what_happened_at_time":"A dialogue-agent research system used human feedback and explicit rules while retrieving supporting evidence for factual responses.","evidence_at_time":"DeepMind's September 22 announcement and the September 28 arXiv paper document the system; the two source dates are kept distinct.","affected_layers":["layer-agents","layer-models","layer-governance","layer-software","layer-human"],"change_kind":"agent_retrieval_and_safety_maturation","q2_continuity":"Q2 showed model-driven action and deployment controls; Sparrow strengthens agent-like information retrieval and explicit behavioral governance.","retrospective_inference":"Dialogue agents were beginning to combine model generation with external evidence and explicit control structures, without persistent delegated autonomy.","uncertainty":"The system remained a research proof of concept, its evidence was not always sufficient, and rule violations remained possible.","date_normalization":"occurred_at/published_at follow the September 22 DeepMind announcement; the associated preprint retains its separate September 28T19:04:43Z publication clock."},"created_at":"2026-10-02T16:08:16.248Z","updated_at":"2026-10-02T16:08:16.248Z"},"artifacts":[{"id":"art-sparrow-20220922","canonical_url":"https://deepmind.google/blog/building-safer-dialogue-agents/","artifact_type":"dialogue_agent_research_release","title":"Sparrow dialogue-agent research","summary":"DeepMind introduced Sparrow, a dialogue agent trained with human feedback, explicit natural-language rules and evidence retrieval for factual claims.","creator_entities":["DeepMind"],"released_at":"2022-09-22T00:00:00.000Z","content_hash":null,"metadata":{"historical_scope":"2022-Q3","date_note":"The research announcement is dated 2022-09-22; the associated arXiv preprint was first submitted 2022-09-28T19:04:43Z and is retained as a separate source clock."},"created_at":"2026-10-02T16:08:16.240Z","updated_at":"2026-10-02T16:08:16.240Z","relation":"evidenced_by"}],"signals":[{"id":"sig-q3-agent-tool-memory-retrieval","statement":"Conversational agent research increasingly combined large language models with external information retrieval, conversational memory, human feedback and explicit behavior constraints.","signal_type":"agent_precursor_maturation","direction":"increasing","confidence":0.94,"epistemic_status":"supported_inference","mapping_mode":"retrospective","reconstructed_at":"2026-10-02T15:12:29.000Z","created_at":"2026-10-02T16:08:16.251Z","updated_at":"2026-10-02T16:08:16.251Z"},{"id":"sig-q3-ai-risk-operationalization","statement":"AI governance became more operational through structured risk-management guidance, explicit dialogue-agent rules, human feedback and deployment safeguards, while remaining largely nonbinding.","signal_type":"governance_maturation","direction":"increasing","confidence":0.93,"epistemic_status":"supported_inference","mapping_mode":"retrospective","reconstructed_at":"2026-10-02T15:12:29.000Z","created_at":"2026-10-02T16:08:16.252Z","updated_at":"2026-10-02T16:08:16.252Z"}]}