Research map / Large language models / AI and society
Multi-Round Human–AI Collaboration with User-Specified Requirements
Sima Noorani, Shayan Kiyani, Hamed Hassani, George J. Pappas
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 |
|---|---|---|---|
| allow more counterfactual harm (epsilon_CH 0.05 -> 0.30) (ablation) | Online calibrated prediction sets · Crowdsourced shape counting | human GT loss rate: epsilon_CH 0.05 (Alg A) vs 0.30 (Alg B) | Hurt |
| tighten complementarity target (epsilon_COMP 0.70 -> 0.50) (ablation) | Online calibrated prediction sets · Crowdsourced shape counting | human GT gain (recovery) rate: epsilon_COMP 0.70 (Alg C) vs 0.50 (Alg A) | Improved |
| AI avoids a CH error (keeps the human's correct label in its set) | Alg A · Crowdsourced shape counting, human held GT | GT loss rate: 35.3% -> 5.1% | Improved |
| AI includes the missed true label (no COMP error) | Alg B · Crowdsourced shape counting, human lacked GT | human GT recovery rate: 6.4% -> 35.8% | Improved |
| loosen CH target in LLM-simulated medical task (ablation) | Online calibrated prediction sets · DDXPlus LLM-simulated collaboration | rate of abandoning a correct initial guess | Hurt |
| tighten COMP target in LLM-simulated medical task (ablation) | Online calibrated prediction sets · DDXPlus LLM-simulated collaboration | human GT recovery rate | Improved |
| apply online threshold updates under non-stationary human behavior | Online calibration algorithm · DDXPlus and crowdsourcing study | running CH and COMP error rates vs targets | Improved |
Closest papers by meaning
- A No Free Lunch Theorem for Human-AI Collaboration (AAAI 2025 · similarity 0.88)
- OpenDeception: Learning Deception and Trust in Human–AI Interaction via Multi-Agent Simulation (ICML 2026 · similarity 0.87)
- Towards Unifying Evaluation of Counterfactual Explanations: Leveraging Large Language Models for Human-Centric Assessments (AAAI 2025 · similarity 0.87)
- Position: Towards Bidirectional Human-AI Alignment (NeurIPS 2025 · similarity 0.86)
- RedDebate: Safer Responses Through Multi-Agent Red Teaming Debates (ICML 2026 · similarity 0.86)
- Human Expertise in Algorithmic Prediction (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.
