Research map / Large language models / AI and society
OmniMechanism Design for Human-AI Collaboration with Impressionable Minds
Tianwei Gao, Yifei Pang, Lunjia Hu, Kate Donahue, Lihua Lei, Vahid Tarokh, Zhun Deng
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Theoretical study of mechanism design for human-AI collaboration when human experts can be influenced (impressionable) by AI outputs; abstract is truncated so details are limited.
Problem
Existing theoretical work on optimal human-AI collaboration mechanisms (when to involve humans and how to combine their expertise with AI predictions) largely overlooks the fact that humans may be influenced by AI outputs, which the abstract cuts off before fully stating.
Approach
Proposes mechanism design for human-AI collaboration that explicitly accounts for impressionable human experts whose judgments are shaped by AI predictions; the abstract is truncated so the specific mechanism is not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| post-process learned omnimechanism per cost instead of a single global policy | Ours (NN) · Sarcasm, c=0.0 | utility: 0.792 -> 0.806 | Improved |
| post-process learned omnimechanism per cost instead of a single global policy | Ours (Boosting) · Art, c>0 | utility: 0.777 vs 0.777 | No clear effect |
| post-process learned omnimechanism instead of Global Best at c=0.2 | Ours (Boosting) · Sarcasm, c=0.2 | utility: 0.776 vs 0.776 | No clear effect |
| use neural base predictive distribution instead of discretized boosting (ablation) | COI omnimechanism · Sarcasm, c=0.1 | utility: 0.779 -> 0.792 | Improved |
| use per-input post-processed policy instead of Global Best on radiologist data | Ours (NN) · Collab-CXR Airspace Opacity, c=0.00 | utility: +0.024 | Improved |
| raise human cost to 0.10 or 0.20 | Ours (NN) · Collab-CXR | utility vs Global Best: 0.000 | No clear effect |
| per-input post-processing instead of Global Best AI_2 on house prices | Ours (NN) · AMES house price, c=0.000 | APE: 0.1335 -> 0.1053 | Improved |
| increase human cost c in post-processing | Ours (NN) · AMES house price | revision rate: 0.0278 -> 0.0000 | Improved |
| always show manipulated signal to humans versus no action | Manipulation baseline · Art, c=0.0 | accuracy: 0.768 -> 0.776 | Improved |
| boosting rounds on COI violation | COI-Boost · Art/Sarcasm training dynamics | KL divergence / COI violation | Improved |
Limitations
- Only the truncated abstract is available; full method and results unknown
Closest papers by meaning
- Human Expertise in Algorithmic Prediction (NeurIPS 2024 · similarity 0.89)
- A No Free Lunch Theorem for Human-AI Collaboration (AAAI 2025 · similarity 0.88)
- Learning to Decide with AI Assistance under Human-Alignment (NeurIPS 2026 · similarity 0.86)
- Designs for Enabling Collaboration in Human-Machine Teaming via Interactive and Explainable Systems (NeurIPS 2024 · similarity 0.86)
- Confounding-Robust Deferral Policy Learning (AAAI 2025 · similarity 0.86)
- Position: Towards Bidirectional Human-AI Alignment (NeurIPS 2025 · 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.
