Study: AI agents barely learn social norms and rely mainly on being punished themselves
ghadfield · x · 2026-11-20
ghadfield highlights work led by his student @binzeli on agents' normative competence. Key results:
- Agents struggle with norms not available from pre-training.
- They find it hard to identify and follow norms while completing tasks.
- They fail to learn from observing others being punished, relying instead on their own experience of sanctions—inefficient and unlike humans.
Fine-tuning helps only modestly. The authors argue individual trial-and-error is insufficient, and socially intelligent AI may require genuine social learning. This helps explain why agents often seem oblivious to norms in real incidents.
More from coding & agent
- Claude drew an entire flyable game world with pure code — no images, no 3D models — prasenx · 2026-11-20
- Fixing mobile input auto-zoom: a Shadcn refinement workflow with Impeccable skills — bennash · 2026-11-20
- DeepSeek V4.1 Flash pushed from broken to 284+ t/s on RTX PRO 6000, +78.7% over vLLM — SergioPaniego · 2026-11-20
- GTE-MLM sparse embedding models now natively supported in Transformers v5 — tomaarsen · 2026-11-20
- After evaluating Forge for MTG, researcher writes his own engine for RL agents — petrusenko_max · 2026-11-20
- Spring ecosystem ships 127 open-source releases in one day, led by Spring AI 2.1.0 — therealdanvega · 2026-11-20