Meet the Faculty: Todd Gureckis
This entry is part of our Meet the Faculty blog series, which introduces and highlights faculty who have recently joined CDS.
Meet Todd Gureckis, who joins CDS this fall as Professor of Psychology and Data Science. Gureckis is not new to the center. He has been an Affiliate and PhD faculty at CDS, for several years, and he directs the Minds, Brains, and Machines initiative. He now is a core member of the CDS faculty.
Gureckis started out in computer engineering at the University of Texas at Austin, expecting to become an electrical engineer. Near the end of his undergraduate degree, he took a part-time job doing Unix system administration for a psychology professor who studied the mind with tools from AI.
“I learned about this field called computational cognitive science, and decided to switch,” Gureckis said.
He stayed at Texas for a PhD in psychology, completed in 2005, then did a postdoc at Indiana University Bloomington before his first faculty job at NYU.
Neural networks were part of the draw from the beginning, though they were smaller and harder to handle then.
Get NYU Center for Data Science’s stories in your inbox
Join Medium for free to get updates from this writer.
“When we had to write a backpropagation algorithm for a complex network, we usually had to work out the equations to derive the gradients one by one everywhere in the model,” he said. “Now you just do that automatically with PyTorch.”
His research treats the mind as an information processing system and tries to reverse engineer it. By observing behavior, eye movements, or indirect measures of brain activity, he works backward to the algorithms that could produce them. He has always focused on what psychologists call higher-level cognition, and he chooses topics by where the human-AI gap remains substantial. For instance, visual categorization was hard for machines when he started his PhD; convolutional networks changed that. Abstract reasoning came next, and he advises the ARC Prize, a benchmark that agentic language models have begun to crack.
“Physical reasoning is still something that’s a little bit difficult for machines,” he said. “My research always seems to be skating a little bit towards where there are hard things to work on that aren’t quite solved yet.”
That is where his lab is concentrated now. People understand the physical world well enough to stack blocks and catch balls, but their mental model of physics is not the real thing. Gureckis wants to build theories of what he called psychological physics, our internal, imperfect physics understanding that explains why humans are good at some judgments and bad at others, and why they can make those judgments so efficiently. He is looking for students with backgrounds in cognitive science, computer vision and video models, and also in physics.
A second line of work turns the same question toward a practical problem. Online data collection in the social sciences is increasingly contaminated by AI bots. Because his lab spends its time characterizing what humans can do that machines cannot, and vice versa, it is well placed to tell them apart. Gureckis is developing detection methods that rely on implicit signatures such as reaction time and known perceptual limits, which current bots do not reproduce faithfully.
He also played a role in how psychology came to sit inside CDS to begin with. Early in the Center’s history, he made the case to the university that psychology was becoming a data-intensive science and needed a seat at the table. That advocacy led to the first joint hires in psychology and data science.
Students interested in working with him should read the getting involved page on his website and fill out the interest form there rather than emailing. “Students who reach out that way are much more likely to get a rapid response,” Gureckis said. Prospective PhD students can apply through either the psychology or the CDS program.
By Stephen Thomas
