Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Language and Experience: A Computational Model of Social Learning in Complex Tasks
Cédric Colas, Tracey Mills, Ben Prystawski, Michael Henry Tessler, Noah Goodman, Jacob Andreas, Joshua B. Tenenbaum
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add human-written advice to experience | Human learners · 10 VGDL games | median attempts: 4 -> 2.25 | Improved |
| add human-written advice to experience | Bayesian LM model · 10 VGDL games | median attempts: 2.5 -> 1.25 | Improved |
| add human advice | Human learners · 10 VGDL games | delta nAUC: +0.12 | Improved |
| add model-generated advice instead of human advice | Bayesian LM model · 10 VGDL games | delta nAUC: +0.052 | Improved |
| add model-generated advice instead of human advice | Human learners · 10 VGDL games | delta nAUC: +0.035, p=0.26 | No clear effect |
| add language-guided proposals to inference (ablation) | Bayesian LM model · 10 VGDL games | delta nAUC: -0.058 without | Improved |
| add language likelihood (without guided proposals) (ablation) | Bayesian LM model · 10 VGDL games | delta nAUC: -0.030 | Mixed |
| add danger warnings in advice | Humans and model · avoidGeorge, relational, jaws, aliens | deaths: 37% to 67% fewer deaths | Improved |
| add incorrect advice about green blocks | Humans and model · avoidGeorge | exploration: avoided green for first 3 episodes vs about 5 interactions | Hurt |
| add generations of iterated advice passing (2 lives each) | Bayesian LM model · 10 VGDL games | delta nAUC: 0.44 to 0.57 | Improved |
| add advice to pure LM agent | Pure LM agent · 10 VGDL games | first level solved: 3/10 games | No clear effect |
Closest papers by meaning
- When Models Know More Than They Can Explain: Quantifying Knowledge Transfer in Human-AI Collaboration (NeurIPS 2025 · similarity 0.88)
- Language Model Goal Selection Differs from Humans' in a Self-Directed Learning Task (NeurIPS 2026 · similarity 0.88)
- Improving Cooperation in Language Games with Bayesian Inference and the Cognitive Hierarchy (AAAI 2025 · similarity 0.88)
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.88)
- MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systems (NeurIPS 2025 · similarity 0.88)
- Large Language Models Think Too Fast To Explore Effectively (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
