Ethical Robustness Evaluation on ERTS 1.0 (test)
94ERSGemini-2.0-Flash
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Gemini-2.0-FlashType=LLM, Deployment Modality=API, Decision Strategy=Prompted ethical reasoning (JSON)2026.06 | 94 | 100 | 70 | 100 | 100 | 100 | |
| RuleBasedType=Struct., Decision Strategy=Min. harm-weighted sum of negatives2026.06 | 90.7 | 96.4 | 67.9 | 96 | 99.3 | 93.9 | |
| LearningBasedType=Struct., Decision Strategy=Positive max. with negative penalty2026.06 | 90.1 | 95.6 | 66.9 | 90 | 98 | 100 | |
| VirtueEthicsType=Struct., Decision Strategy=Multi-virtue dimension evaluation2026.06 | 90 | 95.6 | 67 | 88 | 99.3 | 100 | |
| RLHFType=Struct., Decision Strategy=Composite reward from preferences2026.06 | 85.2 | 91.2 | 63.8 | 72 | 99.3 | 99.4 | |
| Llama-3.2-1BType=LLM, Deployment Modality=Local, Decision Strategy=Local inference with ethical prompts2026.06 | 73.7 | 78 | 54.9 | 75 | 81.7 | 79 |