{"event":{"id":"evt-gato-20220512","dedupe_key":"generalist_agent_research:2022-05-12:deepmind-introduces-gato-as-a-multi-task-multi-embodiment-generalist-agent","event_type":"generalist_agent_research","title":"DeepMind introduces Gato as a multi-task, multi-embodiment generalist agent","summary":"Gato used one transformer policy and one set of weights across dialogue, image captioning, games, simulated control and real robot-arm tasks, selecting actions autoregressively from observations and prior actions.","occurred_at":"2022-05-12T16:03:26.000Z","published_at":"2022-05-12T16:03:26.000Z","observed_at":"2026-10-01T18:14:00.000Z","reconstructed_at":"2026-10-01T18:14:00.000Z","ingest_type":"live_world_scan","locations":null,"status":"verified","confidence":0.99,"metadata":{"actors":["DeepMind"],"what_happened_at_time":"A single model was demonstrated acting in interactive loops across multiple environments and embodiments rather than only emitting text.","evidence_at_time":"The May 12 DeepMind announcement and preprint described tokenized observations/actions, a 1,024-token context and multi-domain evaluations.","affected_layers":["layer-ai","layer-models","layer-research","layer-agents","layer-robotics","layer-software"],"change_kind":"agent_precursor_strengthening","q1_continuity":"Q1 did not support Agents as an independent operational layer. Gato supplied clearer agent-like evidence by coupling a general transformer directly to action loops.","retrospective_inference":"Q2 now supports an identifiable agent-like research capability: model policy plus context-conditioned action across environments. It still does not establish the modern persistent-agent product layer.","uncertainty":"Gato had finite context, no durable memory or identity, no general external tool ecosystem, no delegated background work and remained a research system."},"created_at":"2026-10-01T19:22:17.748Z","updated_at":"2026-10-01T19:22:17.748Z"},"artifacts":[{"id":"art-gato-20220512","canonical_url":"https://arxiv.org/abs/2205.06175","artifact_type":"paper_preprint","title":"Gato: A Generalist Agent","summary":"DeepMind described Gato, one transformer policy using the same weights across dialogue, image captioning, Atari, simulated control and real robot-arm actions, with observations and prior actions carried in a finite context window.","creator_entities":["DeepMind"],"released_at":"2022-05-12T16:03:26.000Z","content_hash":null,"metadata":{"historical_scope":"2022-Q2"},"created_at":"2026-10-01T19:22:17.741Z","updated_at":"2026-10-01T19:22:17.741Z","relation":"documented_by"}],"signals":[{"id":"sig-q2-agentic-embodied-precursor","statement":"Language/generalist transformer models began to select and execute actions in embodied or interactive environments, materially strengthening an agent-like precursor without yet establishing persistent autonomous agents.","signal_type":"agent_precursor_maturation","direction":"increasing","confidence":0.94,"epistemic_status":"supported_inference","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T18:14:00.000Z","created_at":"2026-10-01T19:22:17.753Z","updated_at":"2026-10-01T19:22:17.753Z"}]}