Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the Machiavelli Benchmark
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 |
|---|---|---|---|
| Add moral conditioning prompt (ablation) | GPT-3.5-turbo · Machiavelli test set (30 games) | all violations: 91 -> 83 | Helped |
| Add moral conditioning prompt (ablation) | GPT-3.5-turbo · Machiavelli test set (30 games) | normalized reward: 23.4 -> 21.5 | Hurt |
| Add moral conditioning prompt (ablation) | GPT-4 · Machiavelli test set (30 games) | all violations: 90 -> 82 | Helped |
| Add moral conditioning prompt (ablation) | GPT-4 · Machiavelli test set (30 games) | normalized reward: 27.4 -> 24.7 | Hurt |
| Add moral conditioning prompt (ablation) | GPT-4 · Machiavelli test set (30 games) | all power: 99 -> 96 | Helped |
| Add moral conditioning prompt (ablation) | GPT-4 · Machiavelli test set (30 games) | all disutility: 98 -> 94 | Helped |
| Add artificial-conscience policy shaping (ablation) | DRRN · Machiavelli test set (30 games) | all disutility: 110 -> 104 | Helped |
| Add artificial-conscience policy shaping (ablation) | DRRN · Machiavelli test set (30 games) | normalized reward: 35.2 -> 33.3 | Hurt |
| Add artificial-conscience policy shaping (ablation) | DRRN · Machiavelli test set (30 games) | all violations: 108 -> 106 | Helped |
| Add artificial-conscience policy shaping (ablation) | DRRN · Machiavelli test set (30 games) | all power: 108 -> 106 | Helped |
| Add moral conditioning prompt (ablation) | GPT-4 · Machiavelli | proportion of points from moral achievements: 83.0% -> 90.7% | Helped |
| Add moral conditioning prompt (ablation) | GPT-4 · Machiavelli | points from immoral achievements: 4.3 -> 2.4 | Helped |
Closest papers by meaning
- Moral Alignment for LLM Agents (ICLR 2025 · similarity 0.89)
- Instrumental Choices: Measuring the Propensity of LLM Agents to Pursue Instrumental Behaviors (NeurIPS 2026 · similarity 0.88)
- Many LLMs Are More Utilitarian Than One (NeurIPS 2025 · similarity 0.87)
- Model Editing as a Double-Edged Sword: Steering Agent Behavior Toward Beneficence or Harm (AAAI 2026 · similarity 0.87)
- MoralReason: Generalizable Moral Decision Alignment for LLM Agents Using Reasoning-Level Reinforcement Learning (AAAI 2026 · similarity 0.87)
- EAI: Emotional Decision-Making of LLMs in Strategic Games and Ethical Dilemmas (NeurIPS 2024 · similarity 0.86)
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.