AITP Lab · NYU CDS · Purdue CS (Aug 2026)

AI that learns and thinks with people.

We study humans and machines in mutually informative ways, and build AI thought partners: systems that learn and think with people rather than instead of them.

Research thrusts → Publications Contact

The AI Thought Partner Lab (AITP Lab) works at three scales (individually, together, and collectively) across three lines of inquiry. We measure how human and AI minds share structure, design interactive systems that turn that shared structure into productive partnership, and diagnose when partnership fails because the structure breaks. The lab is led by Ilia Sucholutsky, currently a Faculty Fellow / Assistant Professor at the NYU Center for Data Science, and starting August 2026 as an Assistant Professor in the Department of Computer Science at Purdue University.

Six programs, three lines of inquiry

What we work on

01 Measure Alignment between human and AI minds How human and AI minds come to share structure across representations, concepts, values, and perception, and how to use that structure as a lever for collaboration. Explore → 02 Measure Data-efficient learning A multi-year program on the fundamental limits of learning from very few examples, for machines and for people. We introduced less-than-one-shot (LO-shot) learning. Explore → 03 Diagnose Cognitive science for AI diagnosis Classical psychology paradigms, including IATs, serial reproduction, rational analysis, psychophysics, and theory-of-mind probes, adapted as measurement instruments for AI. Explore → 04 Diagnose Collective intelligence What emerges when humans and AIs interact in groups: teams, networks, and populations. Coordination, cultural dynamics, and failure at scale. Explore → 05 Design Safe and trustworthy thought partners How partnership breaks (bias, overreliance, manipulation, miscalibrated trust) and how to build AI that resists those failure modes. Anchors our DARPA ITM work. Explore → 06 Design Applied thought partnership Where the framework lands in the world, across education, music, mathematics, and robotics, with each application chosen to feed back into the core science. Explore →

Selected recent work

Highlighted papers

Phil. Trans. R. Soc. A · 2026

Revisiting Rogers' Paradox in the context of human-AI interaction

Collins, Bhatt, Sucholutsky.

Nature Human Behaviour · 2024

Building machines that learn and think with people

Collins et al. The AITP framework paper.

NeurIPS Spotlight · 2023

Alignment with human representations supports robust few-shot learning

Sucholutsky & Griffiths.

PNAS · 2025

Explicitly unbiased large language models still form biased associations

Bai et al.

Nature Human Behaviour · 2024

LLMs surpass human experts in predicting neuroscience results

Luo et al.

Preprint · 2026

Language Model Teams as Distributed Systems

Mieczkowski, Collins, Sucholutsky, Vélez, Griffiths.

All publications → Google Scholar GitHub

Updates

Recent news

May 2026 Our paper Revisiting Rogers' Paradox in the context of human-AI interaction is published in Philosophical Transactions of the Royal Society A.
May 2026 Co-organizing The Cognitive Science of AI Alignment workshop at CogSci 2026.
Apr 2026 New preprint: Language Model Teams as Distributed Systems (Mieczkowski et al.) reframes multi-agent LLM workflows using distributed-systems primitives.
Mar 2026 New preprints: Human-AI Synergy Supports Collective Creative Search and Why Human Guidance Matters in Collaborative Vibe Coding.
Jan 2026 Ilia is serving as Area Chair for ICLR 2026, ICML 2026, and NeurIPS 2026.
Dec 2025 Awarded a Google Gemini Academic Program Award and a Tinker Research Award.
Sep 2025 Co-PI on a new $5M DARPA "In the Moment" (ITM) program on algorithmic trust at scale.
Apr 2025 Community paper Getting aligned on representational alignment accepted at TMLR.
Mar 2025 Explicitly unbiased LLMs still form biased associations published in PNAS.
Sep 2024 Ilia starts as a Faculty Fellow / Assistant Professor at NYU CDS.
Aug 2024 Collins et al.'s Building Machines That Learn and Think with People (Nature Human Behaviour) introduces the AITP framework.
Dec 2023 Alignment with human representations supports robust few-shot learning is a NeurIPS Spotlight.

Looking to join or collaborate? Prospective PhD students apply through Purdue CS (applications open Fall 2026 for Fall 2027). Postdocs, visiting researchers, and collaborators are welcome to reach out directly.

Contact the lab →

Active funding includes the DARPA “In the Moment” (ITM) program (Algorithmic Trust at Scale, co-PI; 2025 to 2027), a Google Gemini Academic Program Award (2025), a Tinker Research Award (2026), and a Microsoft Accelerate Foundation Models Research award (2023).