AI research institute on interaction for AI assistants (ARIA)

The National Science Foundation awarded a $4 million Cooperative Agreement to Brown University on October 1, 2025, to establish the AI Research Institute on Interaction for AI Assistants (ARIA), operating under the Computer and Information Science and Engineering program (CFDA 47.070) through September 30, 2030.

ARIA will accelerate development of next-generation AI assistants for mental and behavioral health by integrating research in computer science, neuroscience, cognitive science, philosophy, law, and education. The institute organizes work around three research pillars—Grounding, Instructability, and Alignment—each addressing specific technical challenges in trustworthy AI interaction. Under Grounding, ARIA will develop efficient learning models, new learning algorithms for causal reasoning, computational theories for model architecture trade-offs, and evaluation metrics for trustworthy AI. Instructability will establish new trust-centered paradigms, theories of human-AI interaction, methods for describing AI internal processing, and computational frameworks for AI assistant development. Alignment will operationalize human values and develop methods for human-AI alignment across diverse domains and sectors. The institute will also build workforce capacity through K-12 through postgraduate educational pathways.

Place of performance is Providence, Rhode Island. Brown University serves as prime recipient with sub-awards to seven collaborating institutions awarded between December 5, 2025, and February 12, 2026. Carnegie Mellon University will explore cognitive science principles for AI assistants focused on human well-being. UC San Diego will lead Alignment research on specification and operationalization of human values. Dartmouth College will lead use-inspired research for mental health applications and engage mental health community partners. UC Berkeley will develop mathematical models based on mammalian brain organization and explore active exploration and caregiver roles in learning. Princeton University will advance trustworthy, context-aware AI assistants for mental and behavioral health. Santa Fe Institute researchers will conduct collaborative projects on learning methods, abstract reasoning, and evaluation of internal world models. New York University faculty will apply linguistics, cognitive science, and neuroscience to evaluate and improve AI interaction skills. University of New Mexico doctoral researchers will study collective decision-making methods aligned with diverse stakeholder viewpoints and will engage through the Trustworthy AI Grand Challenge Team. Data & Society Research Institute will provide qualitative and technical researchers to assess AI alignment in clinical settings. Colby College will advance research at the intersection of AI and mental health, with emphasis on undergraduate engagement and workforce development.

Abstract

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. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are planned for this award.