Kanga the marmoset places her hand on the lever and looks at Dodson, a fellow marmoset working with her on a task. As it becomes apparent that Dodson is ready to pull his own lever, neurons in Kanga’s dorsomedial prefrontal cortex ramp up their firing. The activity reaches its peak as Kanga decides to pull the lever, in sync with her partner. As a reward for their coordinated effort, both marmosets earn a sip of liquid marshmallow fluff.
This type of neuronal computation underlies the “evidence accumulation model,” a major theory of how perceptual decisions are made: The brain gathers evidence and executes a decision once the evidence reaches a certain threshold. The marmoset study, which was published last month in Neuron, demonstrates that the model also applies to social decisions.
This result wasn’t a given; making a social decision relies on the changing behavior of another animal, and the actions of the decider can influence what the other animal does, says study investigator Monika Jadi, associate professor of psychiatry and neuroscience at Yale University. “It’s a very recurrent system,” she says.
Support for the evidence accumulation model has come largely from highly controlled experiments; the fact that the same activity pattern appears in a social and less constrained task “implies that this is a generalizable computation,” says Timothy Hanks, associate professor of neurology at the University of California, Davis, who was not involved in the work.
Social, perceptual, foraging and other decisions are “categories we’ve created,” but there may not be anything “acutely different” about them, says Cory Miller, professor of psychology at the University of California, San Diego, who was not involved in the study. “I love this line of work; I think it’s super powerful.”
A
So the team developed the lever-pulling task for Kanga and Dodson instead. It’s a “really nice fusion” of studying natural behaviors while maintaining some experimental control, Miller says. The marmosets stop and start the task whenever they want, and they can repeat it many times.
“It’s not exactly an ecological situation, because they’re not going to go out in the forest and find a couple levers, but at least they’re freely playing the game,” says Peggy Mason, professor of neurobiology at the University of Chicago, who was not involved in the work. “That was a real strength.”
The marmosets’ performance worsens when they can’t see each other, and they struggle to time their lever-pulling correctly when the other lever is programmed to move on its own, according to a 2025 paper from the team.
In this most recent work, the researchers added neuronal recordings: They captured activity in the dorsomedial prefrontal cortex using high-density neural probes encased in a lightweight, untethered head implant, leaving the marmosets free to do the task unencumbered.
Activity in the dorsomedial prefrontal cortex ramped up slowly when there was weak evidence about the other marmoset’s actions, and activity ramped up faster when the evidence was strong. Activity peaked right before the animal decided to pull the lever.
The brain gathered evidence only when a marmoset was looking at its partner, computational modeling showed. “The social gaze is reducing the complexity of the problem,” Jadi says.
Although these findings may seem intuitive, together they demonstrate that for social decisions, “what matters are the dynamics unfolding over time” rather than individual actions, says Inbal Ben-Ami Bartal, associate professor of psychology and neuroscience at Tel Aviv University, who was not involved in the project. “Studies like this help bridge the gap between classic decision neuroscience and the complexity of real-world interactions.”
T
he activity patterns in the dorsomedial prefrontal cortex only correlate with decision-making, Mason says; future research will need to stimulate or inhibit the region during the task to determine if it drives a decision or merely reflects it after it has been made.Additional follow-up experiments could examine if the same evidence accumulation mechanism is at play for decisions that involve trust and long-term relationships, as well as how the marmosets’ performance changes when they work with a stranger versus a monkey they know, Ben-Ami Bartal says.
The group is designing the next iteration of neural implants that will enable them to record from multiple brain regions at once and capture how these regions communicate during the task, says study investigator Anirvan Nandy, associate professor of neuroscience and psychology at Yale.
Overall, these first few papers have been “an early exploration,” Jadi says. “With this paradigm, there are just so many more, deeper questions you can ask neurally.”
