Alfred P. Sloan Foundation Award Will Let NICO Researchers Listen In on Scientific Brainstorms

A two-year award backs a NICO team led by a first-time NICO principal investigator with NICO co-directors Daniel Abrams and Brian Uzzi: tracing ideas from the moment they are spoken to the collaborations they become.

Put fifty early-career scientists in a room for three days, sort them into small groups, and something reliable happens: new teams form. Researchers at the Northwestern Institute on Complex Systems (NICO) have put a number on it. Sitting in on a single small-group session together raises the odds that two scientists later collaborate by roughly sevenfold.

What nobody has measured is the part in between: the brainstorming conversation itself. Somebody floats a half-formed idea. Someone else pokes a hole in it. A third person splices it onto something said twenty minutes earlier. Years later, two of them publish a finding together. Which of those moves mattered?

A new grant from the Alfred P. Sloan Foundation will help a NICO team find out. The two-year award of $325,207, running September 2026 through August 2028, supports The Dynamics of Scientific Brainstorming: From Observation to Evidence-Based Facilitation — an effort to measure what happens inside scientific brainstorming meetings and to test which interaction patterns predict collaborations, proposals, and funding. It comes through the Sloan Foundation's Technology program, which aims to leverage advances in technology to benefit the research community.

“Workshops and conferences shape the trajectory of science, helping researchers find collaborators and develop their ideas among peers,” says Joshua M. Greenberg, director of the Sloan Foundation’s Technology program. “Through this project, the NICO team will provide a fine-scale empirical account of how research ideas surface, evolve, and become collaborations during scientific brainstorm sessions.”

The project is led by Joshua Stadlan, Research Associate Professor at NICO, in his first role as principal investigator. NICO co-directors Daniel Abrams and Brian Uzzi serve as co-PIs, and NICO core faculty Matthew Groh is consulting on the project's use of large language models.

Closing in on the room

The work grows out of a line of research led by Abrams, the Bette and Neison Harris Chair in Teaching Excellence Professor of Engineering Sciences and Applied Mathematics. With Emma Zajdela — who earned her applied math PhD with Abrams and is now a postdoctoral research associate at Princeton — and longtime partners at the Research Corporation for Science Advancement (RCSA), the group showed that assigned small-group interaction at conferences causally increases later collaboration (Physical Review Research, 2022).

They then took the question online. Formal small-group sessions carried more of the collaboration-forming weight in virtual settings, while in-person meetings were far better at connecting strangers — 40 percent of initially unacquainted pairs became acquainted in person, versus 22 percent online (PNAS Nexus, 2025). More recently, Northwestern PhD candidate Ruoming Gong built a pipeline that turns session audio into labeled speech intervals — who spoke, for how long, and after whom — and a model of turn-taking that runs on those intervals alone. PhD candidate Olga Lew-Kiedrowska has used that pipeline across dozens of sessions and hundreds of speakers to study participation, turn transitions, and interruptions at the level of the group.

Each step moved closer to the brainstorming. This project goes inside it.

Why Scialog

Scialog ("science dialog") conferences, run by RCSA, are unusually well suited to study brainstorming. Roughly fifty early-career scientists gather for three days with a smaller number of senior facilitators. Participants are surveyed beforehand about who they already know, and an algorithm assigns them to discussion groups that pair people who have not met and who bring different training. They then form their own teams, write short collaborative proposals, and some receive seed funding.

That structure leaves a complete paper trail: who attended which session, who knew whom in advance, which teams formed, and which proposals were funded. The team's corpus spans 42 conferences across 15 initiatives from 2015 to 2026, including audio and video from nearly 500 brainstorming sessions with more than a thousand participants.

What large language models make possible

Turning hundreds of hours of overlapping, jargon-dense scientific talk into structured data was, until recently, prohibitively tedious hand work. Large language models change the arithmetic. The team will use them to label what each statement in a transcript is doing — proposing, critiquing, extending, synthesizing — and to identify the distinct "idea units" a session produces, so an idea can be followed as it travels through the discussion and into proposal text.

That yields two pictures of a conversation, and each has a Northwestern origin story. One is a linkograph, a design-research method that maps how contributions link back to earlier ones; the team encountered it at a Wednesdays@NICO seminar by Max Kreminski, incoming Assistant Professor of Design Tech at Cornell Tech, entitled "Tracing and Shaping Paths in Design Space." The other is a people–idea network, borrowed from Stadlan's HIV transmission modeling experience at Northwestern's Center for Computational & Social Sciences in Health (COMPASS), where people and the places they go to meet partners form a bipartite network. Swap places for ideas, and you have a map of who engaged which idea, and when.

Uzzi, the Richard L. Thomas Professor of Leadership, brings another framing: his work on how teams assemble (Science, 2005) and on atypical knowledge combinations in high-impact research (Science, 2013) established that novelty often comes from recombination. This project asks whether that recombination can be caught in the act, in the minutes it happens.

The team

Zajdela also connects the project to ARCH, an open-source collaboration platform built by the Santa Fe Institute, where Zajdela was appointed as a Siegel Research Fellow. NICO Research Assistant Professor Evey Huang and research specialist Max Chalekson, formerly a NICO graduate researcher, support the project through their work on a multimodal pipeline that reads video, audio, and language together to extract theory-grounded behavioral signals from a corpus of team recordings. 

The work and data collection rest on a long partnership with RCSA Senior Program Directors Andrew Feig and Richard Wiener, who have collaborated with the NICO group across this entire line of research and built Scialog into the setting that makes it possible.

From radar to contact tracing to conference rooms

Investigator Stadlan came to this question sideways. He earned his PhD in mechanical engineering at Tufts, starting out on state estimation for robotics and sensor systems, while working at The MITRE Corporation, where radar algorithms led him — through MITRE's independent research and development program — to experiment with new strategies for COVID-19 contact tracing. His dissertation modeled face-to-face interaction at academic conferences, and he has surveyed attendees about the conversations that never make it onto a schedule. He recently published a proximity-sensor dataset from a wedding cocktail hour: 95 badge-wearing guests, 7,213 contact events across 2,760 pairs in 58 minutes. (The wedding was his own.) He first joined Northwestern as a postdoctoral fellow in the Feinberg School of Medicine Department of Medical Social Sciences advised by NICO core faculty Michelle Birkett, and continues to collaborate with her Center for Computational and Social Sciences in Health as core faculty. Moving between fields that way — autonomous systems, epidemics, social policy — he kept arriving at the same lever: the social structure around a problem. Modeling how scientists actually talk to each other at conferences turned that instinct into research questions similar to the ones Abrams and Uzzi have tackled at NICO.

Who it's for

The findings are aimed at the people who convene scientists — conference and workshop organizers, facilitators, and the funders who pay for these gatherings — who now design meetings largely on instinct and precedent. The same measurements bear on a question the project leaders have been pondering in discussions of AI agency: what is lost when research is automated end to end, and what collectives contribute that no single member, human or model, could supply on their own. Both are claims about the deliberative part of science — the part they aim to measure. Measuring it means being able to point at a recording and say which human scientist interaction made all the difference.

The Alfred P. Sloan Foundation is a not-for-profit, mission-driven grantmaking institution dedicated to improving the welfare of all through the advancement of scientific knowledge. Established in 1934 by Alfred Pritchard Sloan Jr., then-President and Chief Executive Officer of the General Motors Corporation, the Foundation makes grants in four broad areas: direct support of research in science, technology, engineering, mathematics, and economics; initiatives to increase access and opportunity in graduate science education; projects to develop or leverage technology to empower research; and efforts to enhance and deepen public engagement with science and scientists. sloan.org | @SloanFoundation

This work is supported by Alfred P. Sloan Foundation grant G-2026-79578