TRustworthy Artificial IntelligenCE
Fundamentals of Trustworthy AI
Bridging all of our initiatives is our fundamental technical research into trustworthy AI, spanning transparency, collaboration, and evaluation. We develop methods that work in practice, not just in theory.
Research Areas
01
Transparency & Explanations
Uncertainty-aware explanations and adaptive methods to support users in practice
02
Human-AI Collaboration
Algorithmic resignation, purposeful frictions, and mechanisms to improve performance
03
Evaluation
Interactive evaluation methods, soft labels, and stakeholder-informed tuning
Research Questions
How can we develop explainability techniques that provide value in real-world deployments?
When should AI systems resign or defer to human judgment?
How can we evaluate AI systems to uncover failure modes invisible to benchmarks?
How do we prevent skill atrophy and overreliance on AI assistance?