Research
How can AI empower people over the long term?
We study empowerment as the degree of autonomy and self-determination that people and communities have, and ask how AI systems can expand it rather than erode it. Our work combines system building, controlled experiments, field deployments and participatory design across our research thrusts.
Human-centered
People are involved at every step of the AI pipeline, from defining goals and producing data to evaluating and correcting systems.
Use-inspired
Projects start from concrete settings (crowd platforms, research labs, rescue teams, classrooms) where AI can provide measurable value.
Evaluated in the wild
We test our tools with real users and field experiments, and release code and benchmarks so others can build on them.
Accountable
We audit the limits of AI (calibration, fairness, ethical consistency) and design mitigation and oversight mechanisms.
Thrust 01
Human-Centered Automation & Agentic AI
We design and evaluate automation and AI agents that people can specify, supervise, correct and trust before they act on live systems.
- How should people supervise AI agents whose actions they cannot fully anticipate?
- What makes automation inspectable, correctable and trustworthy before deployment?
- When do teams of LLM agents coordinate well, and when do they fail like human groups?
Thrust 02
Future of Work & Worker-Centered AI
We build and field-test AI tools that rebalance power and opportunity for crowd workers, knowledge workers and the people behind AI pipelines.
- How can AI make digital labor platforms fairer and more transparent for workers?
- How do we measure whether an AI tool truly empowers the people who use it?
- How should language models be screened before they join human-computation pipelines?
Thrust 03
AI for Scientific Discovery
We develop AI that helps researchers conjecture, retrieve, reason and run analyses, with humans validating every step.
- Which mathematical context lets a model propose the result a paper actually proves?
- Can language models reason consistently when the same idea is written in different forms?
- How can scientists delegate analyses to AI while keeping every step verifiable?
Thrust 04
Human-Autonomy Teaming in High-Stakes Environments
With partners in human factors and engineering, we study how autonomous systems can become effective teammates where the stakes are highest.
- What do high-risk teams actually need from an autonomous teammate?
- Which AI models should run on an underwater vehicle as visibility changes?
- How can AI assistance preserve, rather than erode, human situation awareness?
Thrust 05
Trustworthy, Metacognitive & Responsible AI
We audit how language models reason, report confidence, portray people and make ethical choices, and design interfaces that help users think critically.
- When does a model's stated confidence deserve a user's trust?
- Why do people attribute understanding and consciousness to language models?
- How can interfaces nudge users toward fairer and more reflective decisions?
Thrust 06
Inclusive AI for Language, Learning & Health
We co-design AI with bilingual speakers, students, patients and other communities that mainstream technology often overlooks.
- How should AI mediate conversations between speakers of different languages?
- How can LLM tools strengthen, not replace, students' information-seeking skills?
- How do embodiment and interface choices shape trust in health and security technology?
Open science
Software, benchmarks and artifacts
We release research code, benchmarks and systems alongside our papers.