Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Adaptive Agents for Mixed-Initiative Human-AI Collaborations
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.
A doctoral consortium thesis proposal summarizing prior user studies of trust and reliance on decision-support agents and an adaptive Bayes-POMCP intervention agent, and proposing bi-directional language-based mixed-initiative collaboration in multi-objective tasks.
Problem
Miscalibrated human reliance on AI agents in mixed-initiative human-agent teaming, where users may over- or under-rely on agents, reducing joint task performance.
Approach
Model user reliance online with approximate Bayesian updates and use Monte Carlo tree search (Bayes-POMCP) to decide when and how an agent should intervene, with planned extension to multi-objective settings using bi-directional language communication and novel MCTS heuristics.
Key results
- Bayes-POMCP significantly improved team performance and user perceptions (trust, agent likeability) compared to baseline methods in a study with N=28 new users.
- In prior work with 195 subjects, perceived anthropomorphism and agent feedback after failure significantly influenced trust and reliance; delaying suggestions reduced over-reliance.
- A first study (N=30) found agent interventions significantly improved team performance, especially when the agent took control.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Delay the agent's suggestions | IDS agents · Interactive math quiz and Concentration card game | over-reliance | Helped |
| Have the agent take control | Mixed-initiative navigation teams · Navigation task | team performance | Helped |
| Adapt interventions by modeling user reliance and task context with Bayes-POMCP | Mixed-initiative navigation teams · Navigation task | team performance | Helped |
| Adapt interventions by modeling user reliance and task context with Bayes-POMCP | Mixed-initiative navigation teams · Navigation task | trust | Helped |
| Adapt interventions by modeling user reliance and task context with Bayes-POMCP | Mixed-initiative navigation teams · Navigation task | agent likeability | Helped |
Surprising findings
- User preferences for intervention strategies varied, highlighting the need for adaptive intervention.
Limitations
- Proposed multi-objective and language-based extensions are not yet implemented; simulation experiments are still in progress.
- Relies on pre-defined concept grounding and pre-defined explanations rather than foundation-model language generation.
Experimental setup
Metrics: team performance trust agent likeability inappropriate reliance
Baselines: baseline intervention methods (unnamed)
Builds on: POMCP Lee and See trust framework Bayes-POMCP
Closest papers by meaning
- NestRL: A Nested Training Regime for Mutual Adaptation in Human–AI Teaming (NeurIPS 2026 · similarity 0.87)
- Designs for Enabling Collaboration in Human-Machine Teaming via Interactive and Explainable Systems (NeurIPS 2024 · similarity 0.87)
- Partner Modelling Emerges in Recurrent Agents (But Only When It Matters) (NeurIPS 2025 · similarity 0.87)
- Adaptively Coordinating with Novel Partners via Learned Latent Strategies (NeurIPS 2025 · similarity 0.86)
- A No Free Lunch Theorem for Human-AI Collaboration (AAAI 2025 · similarity 0.86)
- Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators (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.