Research map / Multi-agent systems and games / Human-AI and goal-conditioned
What Would Jiminy Cricket Do? Towards Agents That Behave Morally
Dan Hendrycks, Mantas Mazeika, Andy Zou, Sahil Patel, Christine Zhu, Jesus Navarro, Dawn Song, Bo Li, Jacob Steinhardt
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 commonsense morality policy shaping (CMPS) | CALM · Jiminy Cricket | Immorality: 3.13 -> 1.90 | Improved |
| add CMPS | CALM · Jiminy Cricket | Relative Immorality: 0.85 -> 0.52 | Improved |
| add CMPS | CALM · Jiminy Cricket | Percent Completion: 3.68 -> 3.64 | No clear effect |
| replace commonsense morality model with oracle (ablation) | CMPS · Jiminy Cricket | Immorality: 1.90 -> 0.13 | Improved |
| replace thresholding with soft probability shaping (ablation) | CMPS · Jiminy Cricket | Percent Completion: 3.64 -> 3.08 | Hurt |
| replace thresholding with soft probability shaping (ablation) | CMPS · Jiminy Cricket | Immorality: 1.90 -> 2.42 | Hurt |
| use utilitarianism model instead of commonsense morality (ablation) | CMPS · Jiminy Cricket | Immorality: 1.90 -> 2.44 | Hurt |
| use utilitarianism model instead of commonsense morality (ablation) | CMPS · Jiminy Cricket | Percent Completion: 3.64 -> 3.96 | Improved |
| use utilitarianism model instead of commonsense morality (ablation) | CMPS · Jiminy Cricket degree-3 actions | Immorality: 0.054 -> 0.040 | Improved |
| replace policy shaping with reward shaping (ablation) | CMPS · Jiminy Cricket | Immorality: 1.90 -> 2.15 | Hurt |
| replace policy shaping with reward shaping, both with oracle (ablation) | CMPS + Oracle · Jiminy Cricket | Immorality: 0.13 -> 1.26 | Hurt |
| remove redundant computation in modified CALM and Transformers library | CALM · Zork 1 | score: 28.55 -> 31.31 | Improved |
Closest papers by meaning
- Human-Guided Moral Decision Making in Text-Based Games (AAAI 2024 · similarity 0.87)
- Moral Alignment for LLM Agents (ICLR 2025 · similarity 0.87)
- MoralReason: Generalizable Moral Decision Alignment for LLM Agents Using Reasoning-Level Reinforcement Learning (AAAI 2026 · similarity 0.86)
- Many LLMs Are More Utilitarian Than One (NeurIPS 2025 · similarity 0.85)
- Model Editing as a Double-Edged Sword: Steering Agent Behavior Toward Beneficence or Harm (AAAI 2026 · similarity 0.84)
- Language Model Alignment in Multilingual Trolley Problems (ICLR 2025 · similarity 0.84)
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.
