Research map / Probabilistic and scientific modeling / Causal discovery
Explainable Reinforcement Learning through a Causal Lens
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| provide abstract causal explanations rather than relevant-variable explanations | Starcraft II human study · Starcraft II | task prediction score: 8.97 -> 10.90; mean-diff 1.966, p-value 0.034 | Helped |
| provide abstract causal explanations rather than no explanations | Starcraft II human study · Starcraft II | task prediction score: 8.53 -> 10.90; mean-diff 2.400, p-value 0.006 | Helped |
| provide abstract rather than detailed causal explanations | Starcraft II human study · Starcraft II | task prediction score: 10.20 -> 10.90; mean-diff 0.733, p-value 0.735 | No clear effect |
| provide causal rather than relevant-variable explanations | Starcraft II human study · Starcraft II | explanation completeness: 0.873; p-val 0.012 | Helped |
| provide causal rather than relevant-variable explanations | Starcraft II human study · Starcraft II | explanation sufficiency: 1.013; p-val 0.002 | Helped |
| provide causal rather than relevant-variable explanations | Starcraft II human study · Starcraft II | explanation satisfaction: 0.740; p-val 0.029 | Helped |
| use causal explanations rather than relevant-variable explanations | Starcraft II human study · Starcraft II | participant-rated explanation understanding: 0.400; p-val 0.316 | No clear effect |
| use causal explanations to increase trust | Starcraft II human study · Starcraft II | trust | No clear effect |
| use MLP rather than linear regression for structural equations | action influence model · Cartpole | task prediction accuracy: 83.8 -> 86.0 | Helped |
| use MLP rather than linear regression for structural equations | action influence model · LunarLander | task prediction accuracy: 68.4 -> 72.1 | Helped |
| use MLP rather than linear regression for structural equations | action influence model · Starcraft | task prediction accuracy: 94.7 -> 91.4 | Hurt |
| use MLP rather than linear regression for structural equations | action influence model · six RL benchmark domains | task prediction accuracy: Cartpole: 83.8 -> 86.0; MountainCar: 69.7 -> 69.6; Taxi: 68.2 -> 67.9; LunarLander: 68.4 -> 72.1; BipedalWalker: 56.9 -> | Mixed |
Closest papers by meaning
- CausalXRL: Explainable Reinforcement Learning through Causal Graph Reasoning (ICML 2026 · similarity 0.92)
- Learning Nonlinear Causal Reductions to Explain Reinforcement Learning Policies (ICLR 2026 · similarity 0.90)
- Explaining Reinforcement Learning Agents through Counterfactual Action Outcomes (AAAI 2024 · similarity 0.90)
- xWhy: Causal Learning from Explanations (NeurIPS 2026 · similarity 0.89)
- A New Approach to Backtracking Counterfactual Explanations: A Unified Causal Framework for Efficient Model Interpretability (ICML 2025 · similarity 0.87)
- ERCI: An Explainable Experience Replay Approach with Causal Inference for Deep Reinforcement Learning (AAAI 2025 · similarity 0.87)
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.