Conversational XAI on compas
78FaithfulnessNaive LLM
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Naive LLMModel Variant=Original Model2026.05 | 78 | 77 | 91 | 99 | |
| Unstructured Agentic XAIModel Variant=Original Model2026.05 | 73 | 49 | 62 | 81 | |
| Structured Agentic XAI w/o verificationModel Variant=Original Model2026.05 | 71 | 71 | 82 | 99 | |
| FAXModel Variant=Original Model2026.05 | 69 | 62 | 84 | 100 | |
| TalkToModelModel Variant=Feature-Ablated Model2026.05 | 68 | 24 | 32 | 44 | |
| TalkToModelModel Variant=Original Model2026.05 | 67 | 25 | 34 | 45 | |
| FAXModel Variant=Feature-Ablated Model2026.05 | 67 | 68 | 96 | 99 | |
| Explainer DashboardModel Variant=Feature-Ablated Model2026.05 | 64 | 47 | 52 | 43 | |
| Structured Agentic XAI w/o verificationModel Variant=Feature-Ablated Model2026.05 | 64 | 66 | 87 | 98 | |
| Unstructured Agentic XAIModel Variant=Feature-Ablated Model2026.05 | 55 | 65 | 77 | 94 | |
| Explainer DashboardModel Variant=Original Model2026.05 | 54 | 49 | 51 | 42 | |
| Naive LLMModel Variant=Feature-Ablated Model2026.05 | 44 | 70 | 88 | 98 |