Funding Source
NSF, 2433429
Project Period
10/1/25-9/30/30
Principal Investigator
Ellie Pavlick, PhD (Brown University); Suresh Venkatasubramanian, PhD (Brown University)
Other Project Staff
Lisa A. Marsch, PhD (Behavioral and Mental Health Lead)
Project Summary
This project establishes the AI Research Institute on Interaction for AI Assistants (ARIA). ARIA will accelerate the development of next-generation AI assistants for mental and behavioral health, a field where trust, empathy and personalization are critical and where current AI systems fall short. ARIA embraces the integrated study of human cognition and machine cognition, treating them as inherently complementary scientific endeavors essential to achieving successful interaction with AI assistants. ARIA brings researchers in computer science, neuroscience, cognitive science, philosophy, law, and education together with mental health practitioners and civil society groups to forge new opportunities for synergistic scientific inquiry that advances technology and improves human well-being. The institute will grow a future-ready workforce through interdisciplinary education pathways from K-12 through postgraduate training, helping shape a generation that understands the technical and ethical dimensions of AI.
ARIA’s research activities are centered around three pillars: Grounding, Instructability, and Alignment, that are interconnected and motivated by challenges of developing effective AI assistants for mental and behavioral health. In grounding, ARIA will develop new models for efficient learning and generalization, new learning algorithms leading to rich, causal models, computational theories for navigating the inherent trade-offs between learning algorithms or model architectures, and new evaluation metrics for tracking progress toward the goal of trustworthy AI assistants. In instructability, ARIA will design new paradigms centered on establishing trust in AI, new theories and models of how humans interact with AI, new methods for describing AI?s internal processing, and new models and training procedures for integration into a computational framework for the development of AI assistants. In alignment, ARIA will advance current best practices for human-centered design, establish precise definitions for what it means to be aligned, develop computational and experimental methods to operationalize these definitions, and develop cognitively and computationally sound metrics of alignment in complex ethical and social contexts. Across all activities, ARIA promotes integration between academia and industry, and between research and continuing education, enabling a holistic approach to AI conceptualization, development, and evaluation.