Research map / Large language models / AI and society
Who Gets Credit or Blame? Attributing Accountability in Modern AI Systems
Shichang Zhang, Hongzhe Du, Jiaqi W. Ma, Himabindu Lakkaraju
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 |
|---|---|---|---|
| use AA-Score instead of stage-level DVEmb for stage effect estimation | stage effect estimator · MNIST MLP (average over 5 settings) | correlation with retraining: 0.9417 -> 0.9456 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · CelebA ResNet, Stage 3 | correlation with retraining: 0.4203 -> 0.8436 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · CelebA ResNet, Stage 1 | correlation with retraining: 0.4406 -> 0.5524 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · CivilComments Gemma-3, Stage 1 | correlation with retraining: 0.7766 -> 0.8576 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · CivilComments Gemma-3, Stage 3 | correlation with retraining: 0.8514 -> 0.9746 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · X-Ray DenseNet-121, Stage 1 | correlation with retraining: 0.9628 -> 0.9326 | Hurt |
| use AA-Score instead of DVEmb | stage effect estimator · X-Ray DenseNet-121, Stage 3 | correlation with retraining: 0.7025 -> 0.7414 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · MNIST MLP, Shift 2 | correlation with retraining: 0.9329 -> 0.9350 | Improved |
| use AA-Score instead of DVEmb | stage effect estimator · MNIST MLP, Mislabel | correlation with retraining: 0.9496 -> 0.9487 | No clear effect |
| skip the top two stages with positive effect on the confounder | ResNet on CelebA (blonde) · CelebA confounder log-likelihood | confounder log-likelihood: decreases by 0.1130 | Improved |
| skip the bottom stage with negative effect on the confounder | ResNet on CelebA (blonde) · CelebA confounder log-likelihood | confounder log-likelihood: increases by 0.0856 | Hurt |
| disperse poisoned examples evenly across all batches rather than concentrating them | AA-Score · CIFAR-10 BadNets ResNet18 | poisoned batch detection | Hurt |
Closest papers by meaning
- Who&When Pro: Can LLMs Really Attribute Failures in AI Agents? (NeurIPS 2026 · similarity 0.85)
- Models That Know How Evaluations Are Designed Score Safer (NeurIPS 2026 · similarity 0.83)
- Unveiling AI's Blind Spots: An Oracle for In-Domain, Out-of-Domain, and Adversarial Errors (ICML 2025 · similarity 0.83)
- The Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory (NeurIPS 2026 · similarity 0.83)
- Embracing Contradiction: Theoretical Inconsistency Will Not Impede the Road of Building Responsible AI Systems (NeurIPS 2025 · similarity 0.83)
- Calibration Is Not Control: Intervention Advantage for LLM-Agent Oversight (NeurIPS 2026 · similarity 0.83)
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.
