Research map / Label-efficient and robust learning / Recognition and long-tail
‘Why Didn’t You Allocate This Task to Them?’ Negotiation-Aware Task Allocation and Contrastive Explanation Generation
Read the paperCheck your idea against 108,967 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.
Proposes AITA, a centralized task allocator that uses a simulated negotiation process to produce negotiation-aware explicable allocations and generates negotiation-tree contrastive explanations when a human contests with a counterfactual, evaluated via human studies and simulations.
Problem
Centralized task allocation for human teams is often contested by members with incomplete knowledge of teammates' costs and team performance; allocations need to be explicable and contestable with contrastive explanations.
Approach
Computes an allocation via backward induction over a negotiation (allocation enumeration) tree so that all agents would accept it at subgame-perfect equilibrium, and when contested by a counterfactual, explains via a negotiation tree rooted at the counterfactual showing it leads to a no-better outcome for the challenger.
Key results
- Majority of participants (84.2% cooking, 86.4% class project, 55% paper writing) selected AITA's allocation as fair.
- Negotiation-tree explanation rated most convincing in relative setting (ANOVA F(2,24)=8.96, p=0.001) while understandability was statistically equivalent across explanations.
- Explanation length stays small (max 8 in simulation vs. theoretical max 31); pessimistic noise increases explanation length while optimistic noise reduces it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Propose the negotiation-aware AITA allocation instead of the offered counterfactual options | AITA · Cooking, Class Project, and Paper Writing | participant selection as fair: 84.2%; 86.4%; 55% | Helped |
| Generate the optimal counterfactual using the human’s negotiation tree rather than a sub-optimal counterfactual | AITA counterfactual generation · Cooking and Paper Writing | counterfactual selection: 15.8% vs 0%; 37.5% vs 7.5% | Helped |
| Use a negotiation-tree explanation instead of vacuous and verbose explanations | AITA explanation · Cooking and Class Project | convincing power: F(2, 24) = 8.96, p = 0.001; p = 0.003 | Helped |
| Use a negotiation-tree explanation instead of vacuous and verbose explanations | AITA explanation · Cooking and Class Project | understandability: F(2, 24) = 0.04, p = 0.96; p = 0.68 | No clear effect |
| Increase random noise in a human’s estimates of teammate costs | AITA explanation process · Project management simulation | mean explanation length | No clear effect |
| Increase pessimistic noise, causing humans to underestimate teammates’ costs | AITA explanation process · Project management simulation | mean explanation length | Hurt |
| Increase optimistic noise, causing humans to overestimate teammates’ costs | AITA explanation process · Project management simulation | mean explanation length | Helped |
| Increase the number of teammates whose costs a human knows completely | AITA explanation process · Project management simulation | mean relative explanation length | Helped |
Surprising findings
- Random noise in costs does not monotonically increase explanation length because optimistic and pessimistic noise counteract each other.
- Even with zero noise, mean explanation length is about 1 because humans with limited computation pick counterfactuals from a restricted neighborhood.
Limitations
- Assumes AITA knows all true individual and performance costs upfront.
- Fairness assessed via informal subjective perception with small participant pools (38 and 40).
Experimental setup
Datasets: project management scenario from Certa et al. 2009 (simulated)
Metrics: fairness perception (option selection rate) convincingness rating (1-5) understandability rating (1-5) explanation length
Baselines: vacuous explanation verbose explanation
Builds on: model reconciliation explanations sequential bargaining game (Baliga and Serrano)
Closest papers by meaning
- PieArena: Ranking and Profiling Language Agents in Realistic Negotiation Scenarios (NeurIPS 2026 · similarity 0.84)
- Understanding Individual Agent Importance in Multi-Agent System via Counterfactual Reasoning (AAAI 2025 · similarity 0.84)
- On scalable oversight with weak LLMs judging strong LLMs (NeurIPS 2024 · similarity 0.83)
- Explaining Decisions of Agents in Mixed-Motive Games (AAAI 2025 · similarity 0.83)
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.83)
- When Debate Helps: Proposal Supply and Verification-Aware Readout in Multi-Agent Reasoning (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.