Research Area: Computational Neuroscience

​machine learning, modeling, artificial intelligence, neural networks, memory, ​visual intelligence, audition, natural language processing, learning theory

Estimating suicide risk from text

by Jennifer Michalowski |

When people reach out during a mental health crisis, a top priority for counselors is identifying those with a high risk of suicide. The distressed person’s language holds critical clues, and a new tool developed by scientists at MIT’s McGovern Institute is designed to pick up on and rapidly evaluate those signals.

The language-processing tool was developed by Daniel Low, a former graduate student in Satra Ghosh’s Senseable Intelligence Group, who is now a research scientist at the Child Mind Institute, where he leads its AI, Risk, and Contemplative Science Lab, and is also a visiting scholar at Harvard University. It uses a custom-built list of words and phrases linked to 49 suicide risk factors, searching text for these and using them to estimate an individual’s risk.

Ghosh, Low, and colleagues report today in the Journal of Psychopathology and Clinical Science that their tool accurately predicts suicide risk from text conversations with crisis counselors. It is already helping to clarify which suicide risk factors matter most in times of crisis. With more validation, it could help with risk assessment in clinical settings and crisis-support situations.

Identifying key risk factors

Suicide attempts are notoriously difficult to predict. Dozens of risk factors have been linked to suicide, and even trained clinicians struggle to identify who will make an attempt among those who have some form of suicidal ideation. Among the factors that can make suicidal thoughts and behaviors more likely are certain psychiatric symptoms and disorders, like depression, borderline personality disorder, and post-traumatic stress disorder, as well as environmental and social stressors, like poverty, incarceration, discrimination, and loneliness.

“You see all these 50 risk factors, and they’re all interacting in ways we don’t really understand,” Low says. “Many different pathways could lead to someone feeling they want to escape their internal pain,” he says—and it’s challenging to know whose path will lead to a suicide attempt or death.

Ghosh and Low wanted to understand which risk factors counselors and clinicians should most look out for during a mental health crisis. To do that, they partnered with the Crisis Text Line, a nonprofit whose trained volunteers provide confidential text-based support to people in distress.

Crisis Text Line, a global mental health nonprofit that provides free, 24/7, confidential mental health support for people in need, provided specialized training and controlled access to this restricted dataset. The researchers analyzed de-identified texts from approximately 16,000 conversations with Crisis Text Line’s volunteer crisis counselors. Based on Crisis Text Line’s assessments, those conversations were grouped into three different risk levels: non-suicidal, suicidal ideation without imminent risk, and imminent risk. It was this imminent risk group—those with a plan for suicide or have an intent to die within the next 48 hours—that the researchers most wanted to understand.

“We wanted to know what type of symptoms predict the highest suicide risk,” Low says. This question has been studied before, he says—but typically through epidemiological surveys that ask a person to recall their symptoms and experiences, often after their mental health crisis has passed. In contrast, he says, “Crisis Text Line gives us an opportunity to assess many different symptoms and potential risk factors as people are having the crises.”

Reading between the lines

Before analyzing the crisis line texts, the research team built a suicide-risk lexicon. They turned to artificial intelligence to generate a preliminary list of words and phrases tied to established suicide risk factors, including factors associated with suicidal ideation, suicide attempt, and suicide death. Then they manually reviewed and curated that list. Their final lexicon includes about 60 words or phrases for each of 49 risk factors, with the relevance of each one confirmed by expert clinicians.

Then they trained a machine learning model to search the crisis conversations for words and phrases in their lexicon and use these to predict suicide risk. Because the lexicon links each word or phrase to a specific risk factor, they could use this data to determine which risk factors are most closely tied to imminent risk among people in crisis.

What they found was consistent with patterns found in previous research, though not always intuitive. For example, depression is a well-known risk factor for suicidal ideation, but their model found that mentions of lethal means and substance use were more likely to be expressed by the highest-risk group than depressed mood or fatigue. Expressions of active suicidal ideation and self-injury were also strong predictors. Intermediate predictors included anxiety, PTSD, and emotional pain.

The predictive model assigns a weight to each risk factor based on its contribution to risk. For example, mentions of lethal means for suicide, like “cut” or “pills,” are weighed heavily, whereas terms related to hopelessness, like “don’t know what to do” or “hopeless,” contribute to a lesser degree. After training their model, the team found they could use it to accurately predict risk severity in new conversations the model had not previously seen.

One limitation of lexicons, the researchers note, is that they do not consider the context of terms, and they can miss terms that are similar to those in the lexicon, but not explicitly included. Large language models have reasoning abilities, and Low and colleagues have developed ways of using large language models to detect suicide risk in other projects. However, they say they often use their lexicon in parallel to guarantee flagging certain terms, as well as to maintain data privacy.

Low stresses that while the team used the power of a large language model to develop its lexicon, its prediction model is a simpler, “lightweight” model. Unlike large language models, which require massive computational power, it can be run easily on a personal computer, reducing both cost and privacy concerns. Just as importantly, it is interpretable: Rather than merely generating a risk estimate like some deep learning models can do more effectively, it tells users how it got there. Words of concern can be flagged so users understand the basis for each assessment and act on that information. They are working on similar explainability approaches with large language models.

That’s critical, because the stakes are so high. “This is such a complex space that having a human in the loop is, I think, going to be critical for a long, long time,” says Ghosh, who is a Senior Research Scientist and Director of the Open Data in Neuroscience Initiative at the MIT McGovern Institute. Likewise, the researchers add that any predictive model must be thoroughly validated before clinical use, and might need to be continually refined to keep up with changes in language use or target populations.

Because a reliable lexicon opens doors to new ways of understanding mental health, Ghosh and Low are widely sharing not just their suicide risk lexicon, but also the software package they developed to build it. Researchers can use that tool to efficiently build lexicons for other mental health conditions. Meanwhile, Low says, the suicide risk lexicon is already being used to explore how text data from a variety of sources, from social media to electronic health records, might help researchers and clinicians better estimate risk.

How is the McGovern Institute using AI?

by Jennifer Michalowski |

This story also appears in our Fall 2026 BrainScan newsletter.

***

The tools of artificial intelligence (AI) are enabling McGovern researchers to ask new kinds of questions and solve more complex problems — and that’s leading to new insights into how the brain works and how mental illness might be more effectively managed.

“I can’t think of any facet [of brain research] where there isn’t an AI-based tool or technology that helps us,” says McGovern Institute Senior Research Scientist Satrajit Ghosh. Ghosh thinks today’s AI technology is particularly powerful in the early stages of research, when researchers are figuring out which questions are important and how to answer them. Historically, this has meant deep reading of the scientific literature — but existing knowledge can be overlooked when people don’t know where to look for it.

“The entire scientific workflow — from ideation to research to execution to dissemination — now has AI tools and technologies that can help.”
– Satra Ghosh

Large language models like Gemini or ChatGPT can pull information and ideas from sources across the internet, connecting users with resources they might not consider accessing on their own. “We get caught in our tracks and grooves that are hard to step outside of,” Ghosh says. “There are still biases in these tools, but I think overall they’re in a much more flexible space than humans typically are.”

McGovern Investigator Feng Zhang adds that by accelerating this vital phase of doing science, AI tools are broadening the scope of what an individual lab can explore. “We’ve never been short on good questions — what’s new is that we can chase them the moment they occur to us, and follow them as far as the data goes,” says Zhang, who is the James and Patricia Poitras Professor of Neuroscience at MIT.

Testing ideas faster

Once a plan is in place, researchers can often go from idea to execution much faster with the help of AI. McGovern Investigators Nancy Kanwisher and Josh McDermott say they have begun using AI models to pilot experiments, refining their approach before working with human study participants. They even use AI to generate images or sounds that their subjects will hear or see during their studies, designing stimuli to uncover specific aspects of the brain’s response.

The models they use are far more reliable representations of human perception than researchers were able to build just a few years ago. And with major advances in AI’s ability to generate computer code, models that took months or years to piece together can now be generated in days.

McDermott, who is the Uncas (1923) and Helen Whitaker Professor of Brain and Cognitive Sciences at MIT, uses an array of models to investigate different aspects of auditory processing in the brain. His team has models that recognize speech, models that localize sounds, and models that detect changes in pitch. “I fantasized about doing these things when I was a grad student, and it just wasn’t possible,” he says. “We were always trying to build working models of things, and it never worked. Now it’s really possible, and it’s just opening huge numbers of doors.”

Using a computational model of the auditory system, Josh McDermott has shown how the brain can selectively focus attention on one voice among many in a noisy environment — shedding light on a longstanding neuroscientific phenomenon known as the “cocktail party problem.” Photo: Steph Stevens

For McDermott, one priority is finding ways to develop better devices to help people hear. Computational models can facilitate that by enabling experimentation on a scale that is not possible with human subjects, allowing researchers to explore thousands of different conditions or rapidly test the consequences of modifying a hearing aid algorithm in different ways.

AI models also allow researchers to explore conditions they can’t test in the real world. McDermott’s lab, for example, has studied how hearing might work in a world where sounds behave differently, helping reveal which aspects of perception were shaped by the environment in which our ancestors evolved. And Kanwisher says models help overcome the limitations of experiments with humans, where the tools for monitoring and manipulating brain activity are imprecise and researchers have no control over their subjects’ life experiences.

That’s particularly valuable for studying cognitive functions that are uniquely human. “If you want to study things like language, or mathematical cognition, or theory of mind, it’s pretty hard to do in a mouse or even a monkey,” says McGovern Investigator and Associate Professor of Brain and Cognitive Sciences Evelina Fedorenko, who studies how the brain creates and understands language. AI models, on the other hand, can be trained to mimic these abilities. And what scientists like Fedorenko are learning is that models that behave similarly to humans often process information in ways that are similar to the human brain.

“If you’re in the business of studying sensory systems, like I am, it’s totally changed the way that we do science.” – Josh McDermott

Kanwisher, the Walter A. Rosenblith Professor of Cognitive Neuroscience at MIT, and her colleagues have used AI to explore how and why brains have networks specialized for the processing of certain kinds of objects. Her team and others have tasked AI models with processing visual information, training them with an assortment of images without further information or instruction. Remarkably, such models develop specific responses for certain categories of images: faces, places, bodies, and words — all of which Kanwisher’s team has shown are preferentially processed by specific regions in the human brain.

Kanwisher says that was surprising, because cognitive scientists have reasoned that human brains likely evolved these selective regions to optimize the way perceptual information is later used by the brain. But that doesn’t explain why AI responds to these categories with the same selectivity.

“The AI models don’t use face images to engage social cognition or place images to engage navigation,” she says. “They don’t do a thing with them, and yet they pull out the same categories as important.” That doesn’t rule out previous ideas about human brains — but, Kanwisher says, “It’s proof that you can have these shockingly similar things for very different reasons.” Her group is now exploring how the models’ training experience shapes the development of selective regions — something that cannot be done in humans.

Building better tools

AI is also helping McGovern scientists build physical tools, both to allow researchers to study the brain in new ways and to get better treatments to patients.

“Molecular syringes” (above), engineered by the AI tool AlphaFold in the Zhang lab, can deliver therapeutic cargo into specific cells in the body. These nanosyringes may lead to safer and more effective treatments for a variety of conditions, including cancer. Image: Joseph Kreitz, Feng Zhang

In Feng Zhang and Ed Boyden’s labs, researchers are using an AI tool called AlphaFold to design proteins with specific properties. Zhang’s lab is searching for proteins that can deliver therapies to specific brain cells, while Boyden’s team is developing tools for monitoring signaling in the brain.

AlphaFold predicts the structure of biomolecules, such as how a sequence of amino acids will fold into a three-dimensional protein. This eliminates much of the trial-and-error stage of protein engineering, speeding scientists’ ability to find molecules that interact with biological systems in the ways that they want.

Boyden, who is the Y. Eva Tan Professor in Neurotechnology at MIT, envisions another powerful AI tool: one that predicts how a brain works from its structure. The idea, he says, is to use machine learning to associate detailed brain maps with physiology and circuit function.

Building such a model would require vast amounts of data, but Boyden says the tools and tissue samples needed for large-scale structural analyses are available.

He envisions a kind of “compiler” built from molecular maps of brain cells and tissues — one that could help neuroscientists understand circuit function and eventually predict the effects of medications and other therapies.

Decoding data

Once an experiment is complete, data analysis begins — and AI is enabling scientists to find insights in datasets that, until recently, were too large and complex to make sense of.

AI tools have been invaluable, for instance, in creating detailed wiring diagrams, or connectomes, which many neuroscientists see as critical guides for brain research. That work, which involves tracing neurons across millions of microscope images, relies heavily on computer vision.

Without AI, piecing together the circuitry of an entire brain would take lifetimes. Remarkably, scientists now have access to a fully mapped connectome for the brain of a fruit fly, whose neurons make more than 50 million connections with one another.

A consortium of scientists, including McGovern Investigator Sven Dorkenwald, created a complete connectome of the fly brain (above) using computer vision algorithms and hundreds of human proofreaders. It is the largest, most complete neuronal map ever produced by scientists. Image: FlyWire Consortium, Sven Dorkenwald

McGovern Investigator Tomaso Poggio and his team are using AI to interact with that connectome by running simulations of circuit function, testing for a motion-detecting circuit that Poggio first proposed in theoretical work he did in the 1980s. Until recently, Poggio, who is the Eugene McDermott Professor Emeritus of Brain and Cognitive Sciences at MIT, had no way to find it.

Personalizing mental health

Other McGovern researchers are deploying AI to find ways to improve the diagnosis and treatment of brain disorders. They are finding that AI can pick up on signals that can be difficult for clinicians to detect — sometimes years before diagnosis.

Working with Satra Ghosh, John Gabrieli’s team found that a machine learning model could predict from clinical data which children and teenagers would later be diagnosed with bipolar disorder. With other collaborators, Ghosh has found that AI can recognize autism-related behaviors in parents’ videos of their toddlers, and he trained an AI model to detect changes in voice and speech patterns associated with conditions like Alzheimer’s and Parkinson’s disease. With further development, tools like these could facilitate earlier diagnosis and intervention for people who need it.

Researchers in Gabrieli’s lab think AI can also enable a more personalized approach to treating mental illness. It’s difficult to predict who will respond to a particular medication or therapy, leaving doctors and patients with a trial-and-error approach to treatment. Gabrieli’s team is betting that machine learning algorithms can find patterns hidden within complex clinical datasets that can enable more informed treatment decisions.

By merging machine-learning approaches with clinical and neuroimaging data, Gun Ahn, a graduate student in John Gabrieli’s lab, is pioneering precision-based methods that could help heal mental illness more quickly and effectively than current strategies.

In a recent success, Gun Ahn, a graduate student on Gabrieli’s team, followed up on a study in which the group found that a daily mindfulness practice meaningfully reduced anxiety for about 50 percent of autistic adults. No single factor separated those who benefitted from those who did not, so the team used machine learning to analyze demographic data and patient questionnaires. Their model was able to predict, with 80 percent accuracy, who responded to the practice.

This finding could help patients decide whether mindfulness is worth trying, particularly if it can one day be paired with predictions about alternative treatments. Gabrieli’s lab is taking a similar approach to predict patients’ response to medications for attention-deficit/hyperactivity disorder, hoping to help people bypass ineffective options and quickly find what works for them.

Tailoring interventions to a patient’s personal biology and experience has long been a goal in psychiatry — but the knowledge clinicians need to achieve this has been elusive. “For about 10 to 15 years, people have been talking about the incredible value that precision psychiatry would have,” says Gabrieli, who is the Grover Hermann Professor of Health Sciences and Technology at MIT. “This is, to me, the most promising scientific approach that could be widely usable.”

AI could even help clinicians move beyond formal diagnoses to understanding and treating people more holistically. Ahn imagines a model trained broadly, with data from a wide swath of patients.

Moving the field forward

Science isn’t done once the data has been analyzed: sharing knowledge is essential, too. Poggio, for example, integrated AI into the writing of a new book, “Brains Minds Machines: The Mystery of Human Intelligence, The Enigmas of the Artificial,” even crediting Gemini and ChatGPT as coauthors. He says the tools helped him organize his thinking and offload tedious tasks.

Looking ahead, most McGovern scientists say AI technologies are advancing so rapidly, their future impacts on neuroscience are difficult to predict. Could AI one day replace human researchers altogether? Some think it’s possible — at least in some areas.

“The scary prospect is that AI is going to put humans out of business. We’re definitely not close to that now, but nobody can predict the future,” says McDermott. “Physically doing experiments may be something that continues to be the domain of humans. But deciding what experiments to run and what they mean, and analyzing all the data? It’s unclear whether humans are still going to be the most efficient way to do that.”

Others are more skeptical. “It’s hard for AI to generate truly creative new ideas,” Boyden says. “I haven’t seen that happen yet.”

For now at least, today’s tools need oversight and guidance. To get the most out of them, users need experience and expertise. “They still have hallucinations from time to time,” Poggio warns. “One has to be careful.”

Nonetheless, today’s tools are tremendously empowering. They’re not just getting more powerful; they’re also getting more accessible. “These models are becoming exciting because they’re becoming usable by everyone, not just specialists,” Zhang says. “And when more people can do more, the whole field moves faster.”

How visual learning happens in the brain

by Jennifer Michalowski |

The wiring and rewiring of the brain never ends. Neural pathways are constantly being reshaped as we interact with the world and learn new things. At MIT’s McGovern Institute and York University in Toronto, scientists are combining detailed analysis of brain activity with computational modeling to better understand that change.

McGovern Institute postdoctoral fellow Lynn Sörensen, Investigator James DiCarlo, and York University assistant professor Kohitij Kar worked together to compare what happened when monkeys and an artificial neural network with brain-like architecture were trained to visually identify the same objects. As the model’s performance improved, it reorganized itself in ways that closely paralleled changes the team detected in the brains of monkeys.

Their work, reported today in the journal Nature Communications, shows how changes in visual processing support animals’ ability to learn to discriminate new kinds of objects. By modeling these changes, the researchers hope to better predict how training reshapes perception, which could one day inform educational strategies for a wide range of learners.

Subtle changes

Learning about a new object calls on many parts of the brain. Visual-processing areas work together to make sense of information taken in through the eyes, then communicate with other brain areas to give the visual information meaning and guide behavior. Multiple parts of this system likely change during learning, and the research team wanted a clearer understanding of how that change is distributed.

Neuroscientists have debated how much change occurs in the brain’s visual-processing areas when an animal learns to recognize new objects. Some suspected that visual-processing pathways remain largely unchanged during learning to avoid broadly disrupting visual perception, but others have reported changes in activity within dedicated visual-processing areas with this kind of learning in humans and other primates.  

To take a closer look, the team focused on neural activity in a key component of the brain’s visual object-processing network, the inferior temporal (IT) cortex. By the time visual information reaches the IT cortex, key object features are clearly represented—so much so that it’s possible to “decode” what object a monkey is seeing and even predict what errors it’s likely to make in identifying it, simply by analyzing patterns of neural activity there.

The team recorded neural activity in the IT cortex from two groups of monkeys as the animals looked at and identified images of objects. Some of the monkeys were untrained, so the images they saw had little meaning to them. Others had already learned to identify similar objects, so they could usually discriminate between elephants, chairs, and other select objects, even when those objects were presented at different sizes, from different angles, or against different backgrounds than the ones they had seen before.

The broad pattern of activity in the IT cortex was largely similar in trained and untrained monkeys, suggesting that learning had not dramatically rewritten this high-level visual representation. Still, the group found subtle but reliable differences in the way neurons in the IT cortex responded to images in monkeys that had learned to recognize the kinds of objects they were shown, compared to the untrained monkeys.

Modeling learning

The group turned to computational models to investigate how those modest changes might contribute to learning. Sörensen trained a suite of artificial neural networks whose internal components had been mapped to monkey IT cortex to identify the same categories of objects the monkeys had seen. The models were designed to learn using gradient descent, meaning they continually improved their accuracy by adjusting their parameters in response to errors.

Only some of the primate-like models showed learning behavior that matched that of the monkeys. In those that did, the IT-like stage changed in ways that resembled the learning-related changes the researchers had observed in the IT cortex of trained monkeys.

While gradient descent is commonly used to train artificial intelligence, it is generally considered biologically implausible as a direct model of how the brain learns. The researchers say the strong match in learning effects between the monkeys and their model demonstrates that these kinds of artificial neural networks can offer insights into biological learning at a useful level of abstraction, even if the brain does not learn in the same way.

“This shows that you can actually build in silico versions of future experiments,” Sörensen says. “I think that gives us this playground of asking ‘what if’ questions—and potentially predicting new things that go beyond the experimenter’s intuition.”

Most of the changes that allowed for learning in the model occurred outside of the IT cortex. “This tells us that there is a lot between the area we recorded from and the final behavioral readout that needs to change during this process,” Kar says. He adds that the team’s model will be useful as researchers look more deeply into how downstream brain areas contribute to learning.

The researchers stress that their study allowed more granular measurements of brain activity than would be possible in humans, and because monkeys’ brains are organized similarly to our own, their experiments have direct relevance to human learning. They say understanding the impact of plasticity in monkeys’ IT cortex could help researchers design new learning strategies for humans.

“Our prior conceptual working model of you—or a monkey—learning new objects was that your brain makes changes to synaptic connections that are largely downstream of your visual system, so you don’t destroy your visual system,” says DiCarlo, who is also the Peter de Florez Professor of Brain and Cognitive Sciences and Director of the MIT Siegel Family Quest for Intelligence. “You wouldn’t want your whole visual system to become an elephant detector [just because you’ve learned to identify an elephant]. But this study went beyond that to say actually, when you learn ‘elephant,’ your IT does change a little bit to make it a little more relevant to elephants.”

That likely has consequences for recognizing other visual features, too. Subtle changes in the IT cortex that support elephant recognition might also make you better at identifying things other than elephants, DiCarlo says. Likewise, the same changes might make it a little harder to identify something else.

These kinds of consequences may be difficult to predict intuitively, but become obvious with computational modeling. For instance, the team’s models revealed that after learning to recognize new objects, the IT cortex contained more information about objects’ locations. By providing insights like these, models could aid the design of more effective training strategies for visual tasks, including for people with altered sensory processing, who may learn from visual information in atypical ways.

 

 

Two McGovern faculty appointed to named professorships at MIT

by Julie Pryor |

MIT has appointed McGovern faculty members Sven Dorkenwald and Josh McDermott to named professorships that will provide additional support for their “outstanding research and educational careers.” Named professorships at MIT are prestigious endowed faculty chairs that provide crucial financial support for both junior faculty and senior scholars, enabling them to pursue bold research and global challenges.

Dorkenwald, who recently joined MIT as an assistant professor of brain and cognitive sciences and an investigator at the McGovern Institute, has been selected to hold the  Silverman (1968) Family Career Development Professorship for a three-year term beginning July 1, 2026. A trailblazer in the field of computational neuroscience, Dorkenwald reconstructs maps of neuronal circuits to investigate how they support complex computations. He is recognized for his leadership in connectomics—an emerging discipline focused on reconstructing and analyzing neural circuitry at unprecedented scale and detail. Jeffrey Silverman ’68 is a life member emeritus of the MIT Corporation. His generous gift to the institute empowers early career professors to pursue high-risk research. 

McDermott, a professor of brain and cognitive sciences and an associate investigator at the McGovern Institute, has been selected to hold the Uncas (1923) and Helen Whitaker Professorship for a five-year renewable term beginning July 1, 2026. McDermott’s research operates at the intersection of psychology, neuroscience, and engineering to study how people hear and interpret sound. Groundbreaking discoveries from the McDermott lab are informing new treatments for hearing loss, and paving the way for machine systems that emulate the human ability to recognize and interpret sound. The Uncas (1923) and Helen Whitaker Professorship chair was established in 1980 through a gift from the late Helen Whitaker, the first woman elected to life membership of the MIT Corporation. It is designed to support distinguished faculty whose work spans multiple disciplines to solve complex, real-world problems. 

 

A different reality

by Jennifer Michalowski |

This story also appears in our Spring 2026 BrainScan newsletter.

***

Schizophrenia, a complex and variable psychiatric disorder, changes people’s perceptions of reality. People with schizophrenia may hear, see, or sense things that aren’t there, and they often hold firm to mistaken ideas about the world despite strong evidence to the contrary. As if these changes aren’t disruptive enough, they are usually accompanied by cognitive difficulties and disorganized thinking.

Scientists at the McGovern Institute’s Poitras Center for Psychiatric Disorders Research are looking for clues into the origins of the disorder and its symptoms so they can help guide the development of new treatments. Encouragingly, they are beginning to uncover the brain changes that reshape reality for people with schizophrenia.

Genetic clues

Researchers who want to study the root causes of a disease often turn to genetics for clues—and the genetics of schizophrenia are complicated. Hundreds of different genes seem to shape people’s risk of developing the disorder, most of which nudge risk only slightly. For most people, it seems to be the cumulative effect of these genes and how they intersect with other risk factors, like stress and prenatal complications, that determine who develops schizophrenia and who does not.

Gene variants that substantially impact the risk of schizophrenia are expected to reveal more about the underlying biology of the disorder than genes whose individual impact is minor. But these variants are rare, and it took a massive study to find them. In 2022, scientists at the Broad Institute’s Stanley Center for Psychiatric Research reported that after analyzing the DNA of more than 24,000 people with schizophrenia, they had identified mutations in 10 genes that dramatically increased the risk of the disorder.

“I think this is exciting, because for the first time, you can actually have an animal model based onhuman genetics findings,” says McGovern Institute and Stanley Center Investigator Guoping Feng. “You can put these mutations in animal models to try to understand how this mutation affects brain development, circuit formation, circuit function, and behavior.” Feng is also the James W. (1963) and Patricia T. Poitras Professor of Brain and Cognitive Sciences at MIT.

Woman and man sit at desk looking at brain image on computer screen.
Guoping Feng (right) and his postdoctoral researcher Tinting Zhou (left) examine a mouse brain carrying a genetic mutation associated with schizophrenia. Photo: Steph Stevens

In work supported by the Poitras Center, the Stelling Family Research Fund, and the Yang Tan Collective at MIT, Feng’s lab has engineered three strains of mice that carry ultra-rare schizophrenia-associated mutations. Their first significant findings come from mice with a mutation in a gene called Grin2a. People who inherit a dysfunctional Grin2a gene, which neurons need to detect and respond to a signaling molecule called NMDA, are 20 times more likely to develop schizophrenia than people in whom Grin2a is intact.

Tingting Zhou, a postdoctoral researcher in Feng’s lab, says the team had to think carefully about how to assess mice for schizophrenia-like symptoms. You can’t ask mice about hallucinations or delusions. Instead, Zhou designed an experiment that tested how well mice use new information to update their beliefs about the world—a process that is thought to be impaired in people who experience delusions.

To illustrate how failure to update beliefs can skew someone’s ideas about reality, Zhou describes a situation in which a person watches a stranger reach for something in their pocket, fearing that person intends to harm them. Then, the stranger’s hand emerges with a lollipop. The new information should alleviate concern—but a person with schizophrenia might hold on to their original belief, convinced the lollipop-holding stranger is a threat.

In Zhou’s experiments testing animals’ belief-updating abilities, mice had to keep up with changing information to earn as many treats as possible. Those with the Grin2a mutation were slow to adapt when experimenters adjusted the relative values of their choices. “Once the animal learns something, it’s very hard for them to update the information,” Zhou explains.

Zhou and Feng linked this behavioral difference to abnormally low activity in a part of the brain called the mediodorsal thalamus. The mediodorsal thalamus acts like a switchboard in the brain, routing and coordinating information between different parts of the cortex to support thinking, decision-making, and flexible behavior. Studies with patients have implicated this region in schizophrenia as well, showing that it has fewer cells and is less active in people with the disorder than those without.

A slice of mouse brain dyed purple showing two pink blobs towards the center.
The mediodorsal thalamus (pink) is less active in people with schizophrenia and mouse models of the disease. Image: Guoping Feng, Tingting Zhou

Feng’s lab and others are now looking for belief-updating deficits in other genetic models of schizophrenia. “The goal is to look at whether this is a converging mechanism…then you can start to look at what other [brain] regions are involved,” he says.

In mice with Grin2a mutations, the researchers were able to restore normal belief updating by activating neurons in the mediodorsal thalamus, offering hope that manipulating the same circuitry might benefit patients. “It will not be easy,” Feng says, “but at least you have something you can work on. Previously, it was just very hard to imagine how to develop a new therapeutic for schizophrenia.”

Internal noise

It’s not just the genes associated with schizophrenia that differ across affected individuals. The symptoms of the disorder vary, too. People experience some combination of delusions, hallucinations, disorganized speech, and cognitive problems—but none of these are experienced by everyone with the disorder. This heterogeneity complicates the diagnosis, treatment, and study of schizophrenia. For this reason, some researchers are focusing their efforts on understanding its individual symptoms.

Evelina Fedorenko, a McGovern Investigator and associate professor of brain and cognitive sciences, specializes in understanding how the brain processes speech and language. But recently, her group has teamed up with physician-researcher Ann Shinn at McLean Hospital to begin exploring why some people hear voices when no one is speaking.

About three out of four people with schizophrenia experience auditory hallucinations, which most commonly involve voices.

These hallucinations can be distressing, sometimes involving threatening language or commands to cause harm. Some people with mood disorders or post-traumatic stress disorder also hear them.

Scientist portrait
Tamar Regev was the 2022–2024 Poitras Center Postdoctoral
Fellow in Evelina Fedorenko’s lab. Photo: Steph Stevens

To investigate, Tamar Regev, a research scientist in the Fedorenko lab, asked people who experience auditory hallucinations to listen to different kinds of sounds inside an MRI scanner, then compared how their brains responded versus the brains of people without auditory hallucinations. Her study included participants with schizophrenia and bipolar disorder, both with and without a history of auditory hallucinations, as well as healthy controls.

Inside the scanner, participants listened to three kinds of audio: spoken language, gibberish, and gibberish so scrambled that it barely resembled speech. Regev analyzed how these sounds impacted activity in areas the brain uses to process auditory input at different levels: a part of the auditory cortex that is sensitive to all sounds; a higher-level region within the auditory cortex that usually responds to anything that sounds like speech, even if its content is unclear; and the brain’s language-processing network, which is called on to understand the content of speech, as well as written or signed communications.

Regev found that in people with hallucinations, the part of the brain that usually responds only to language responded to meaningless speech as well. “In this pathway from auditory to speech to language processing, the stimuli that should be filtered out somewhere on the way are now passing to higher stations,” she explains. While auditory hallucinations don’t require external sounds, Fedorenko and Regev propose that the brain’s language areas might be similarly activated by “internal noise” in auditory circuits.

Scrambled language

In people who experience auditory hallucinations, the brain’s language regions respond to sounds that aren’t language–including scrambled meaningless gibberish. Below is a sample gibberish clip used in Fedorenko’s study.

Early identification

McGovern scientists have also used brain imaging to investigate what happens in the brain before people develop clear symptoms of schizophrenia. The disorder is usually diagnosed in adolescence or young adulthood, when patients exhibit the first signs of psychosis—but its origins in the brain likely take root years before that.

“One of the things we’re super interested in is, can you identify people at risk early on, before they have a big problem,” says McGovern Investigator John Gabrieli, whose work is also supported by the Poitras Center and the Stelling Family Research Fund. That might give clinicians an opportunity to intervene and lessen or prevent the disorder’s most devastating effects, he says.

Gabrieli and his colleagues have studied the brains of children who, because they have a parent or sibling with schizophrenia, have an elevated risk of developing the disorder themselves. They found that a system called the default mode network (DMN), which is overactive in adults with schizophrenia, is already working overtime when children in this high-risk group are seven- to 12-years-old.

Gabrieli explains that the DMN is active when people are not actively engaged in an activity or thinking about the external world. “It turns on when you think about your family, your values, your hopes for the future, or important events of your life. It’s almost like a system of who /you are,” he says. Hallucinations and delusions experienced by people with schizophrenia may be associated with overactivity in this network.

MRI images of two brains, one showing an active DMN and the other showing a healthy DMN.
The default mode network (DMN) is a large-scale brain network that is active when a person is not focused on the outside world and the brain is at wakeful rest. The DMN is often over-engaged in adolescents with depression and anxiety, as well as teens at risk for these and other disorders like schizophrenia (left). DMN activation and connectivity can be “tuned” to a healthier state through the practice of mindfulness (right).

“They’re kind of living in their internal world of beliefs, as opposed to the reality that most of us occupy,” Gabrieli explains.

He and his colleagues think overactivity in the DMN might make people vulnerable to schizophrenia—and their data show this atypical activity can be detected many years before the core symptoms of schizophrenia appear. With further validation, children with hyperactivity of the DMN might be candidates for early intervention.

With new and better interventions, the ability to identify people who may be on a path toward schizophrenia will be even more impactful—underscoring the need for continued research on multiple fronts. A recent gift of $8 million to the Poitras Center from Patricia and James Poitras is helping accelerate this work in labs at the McGovern Institute and beyond.

Three anesthesia drugs all have the same effect in the brain, MIT researchers find

by Anne Trafton |

When patients undergo general anesthesia, doctors can choose among several drugs. Although each of these drugs acts on neurons in different ways, they all lead to the same result: a disruption of the brain’s balance between stability and excitability, according to a new MIT study.

This disruption causes neural activity to become increasingly unstable, until the brain loses consciousness, the researchers found. The discovery of this common mechanism could make it easier to develop new technologies for monitoring patients while they are undergoing anesthesia.

“What’s exciting about that is the possibility of a universal anesthesia-delivery system that can measure this one signal and tell how unconscious you are, regardless of which drugs they’re using in the operating room,” says Earl Miller, the Picower Professor of Neuroscience and a member of MIT’s Picower Institute for Learning and Memory.

Miller, Edward Hood Taplin Professor of Medical Engineering and Computational Neuroscience Emery Brown, and their colleagues are now working on an automated control system for delivery of anesthesia drugs, which would measure the brain’s stability using EEG and then automatically adjust the drug dose. This could help doctors ensure that patients stay unconscious throughout surgery without becoming too deeply unconscious, which can have negative side effects following the procedure.

Miller and Ila Fiete, a professor of brain and cognitive sciences, the director of the K. Lisa Yang Integrative Computational Neuroscience Center (ICoN), and a member of MIT’s McGovern Institute for Brain Research, are the senior authors of the new study, which appears today in Cell Reports. MIT graduate student Adam Eisen is the paper’s lead author.

Destabilizing the brain

Exactly how anesthesia drugs cause the brain to lose consciousness has been a longstanding question in neuroscience. In 2024, a study from Miller’s and Fiete’s labs suggested that for propofol, the answer is that anesthesia works by disrupting the balance between stability and excitability in the brain.

When someone is awake, their brain is able to maintain this delicate balance, responding to sensory information or other input and then returning to a stable baseline.

“The nervous system has to operate on a knife’s edge in this narrow range of excitability,” Miller says. “It has to be excitable enough so different parts can influence one another, but if it gets too excited it goes off into chaotic activity.”

In that 2024 study, the researchers found that propofol knocks the brain out of this state, known as “dynamic stability.” As doses of the drug increased, the brain took longer and longer to return to its baseline state after responding to new input. This effect became increasingly pronounced until consciousness was lost.

For that study, the researchers devised a computational model that analyzes neural activity recorded from the brain. This technique allowed them to determine how the brain responds to perturbations such as an auditory tone or other sensory input, and how long it takes to return to its baseline stability.

In their new study, the researchers used the same technique to measure how the brain responds to not only propofol but two additional anesthesia drugs — ketamine and dexmedetomidine. Animals were given one of the three drugs while their brain activity was analyzed, including their response to auditory tones.

This study showed that the same destabilization induced by propofol also appears during administration of the other two drugs. This “universal signature” appears even though the three drugs have different molecular mechanisms: propofol binds to GABA receptors, inhibiting neurons that have those receptors; dexmedetomidine blocks the release of norepinephrine; and ketamine blocks NMDA receptors, suppressing neurons with those receptors.

Each of these pathways, the researchers hypothesize, affect the brain’s balance of stability and excitability in different ways, and each leads to an overall destabilization of this balance.

“All three of these drugs appear to do the exact same thing,” Miller says. “In fact, you could look at the destabilization measure we use and you can’t tell which drug is being applied.”

The researchers now plan to further investigate how each of these drugs may give rise to the same patterns of brain destabilization.

“The molecular mechanisms of ketamine and dexmedetomidine are a bit more involved than propofol mechanisms,” Eisen says. “A future direction is to do a meaningful model of what the biophysical effects of those are and see how that could lead to destabilization.”

Monitoring anesthesia

Now that the researchers have shown that three different anesthesia drugs produce similar destabilization patters in the brain, they believe that measuring those patterns could offer a valuable way to monitor patients during anesthesia. While anesthesia is overall a very safe procedure, it does carry some risks, especially for very young children and for people over 65.

For adults suffering from dementia, anesthesia can make the condition worse, and it can also exacerbate neuropsychiatric disorders such as depression. These risks are higher if patients go into a deeper state of unconsciousness known as burst suppression.

To help reduce those risks, Miller and Brown, who is also an anesthesiologist at MGH, are developing a prototype device that can measure patients’ EEG readings while under anesthesia and adjust their dose accordingly. Currently, doctors monitor patients’ heart rate, blood pressure, and other vital signs during surgery, but these don’t give as accurate a reading of how deeply the patient is unconscious.

“If you can limit people’s exposure to anesthesia, if you give just enough and no more, you can reduce risks across the board,” Miller says.

Working with researchers at Brown University, the MIT team is now planning to run a small clinical trial of their monitoring device with patients undergoing surgery.

The research was funded by the U.S. Office of Naval Research, the National Institute of Mental Health, the Simons Center for the Social Brain, the Freedom Together Foundation, the Picower Institute, the National Science Foundation Computer and Information Science and Engineering Directorate, the Simons Collaboration on the Global Brain, the McGovern Institute, and the National Institutes of Health.

How the brain handles the “cocktail party problem”

by Anne Trafton |

MIT neuroscientists have figured out how the brain is able to focus on a single voice among a cacophony of many voices, shedding light on a longstanding neuroscientific phenomenon known as the cocktail party problem.

This attentional focus becomes necessary when you’re in any crowded environment, such as a cocktail party, with many conversations going on at once. Somehow, your brain is able to follow the voice of the person you’re talking to, despite all the other voices that you’re hearing in the background.

Using a computational model of the auditory system, the MIT team found that amplifying the activity of the neural processing units that respond to features of a target voice, such as its pitch, allows that voice to be boosted to the forefront of attention.

“That simple motif is enough to cause much of the phenotype of human auditory attention to emerge, and the model ends up reproducing a very wide range of human attentional behaviors for sound,” says Josh McDermott, a professor of brain and cognitive sciences at MIT, a member of MIT’s McGovern Institute for Brain Research and Center for Brains, Minds, and Machines, and the senior author of the study.

The findings are consistent with previous studies showing that when people or animals focus on a specific auditory input, neurons in the auditory cortex that respond to features of the target stimulus amplify their activity. This is the first study to show that extra boost is enough to explain how the brain solves the cocktail party problem.

Ian Griffith, a graduate student in the Harvard Program in Speech and Hearing Biosciences and Technology, who is advised by McDermott, is the lead author of the paper. MIT graduate student R. Preston Hess is also an author of the paper, which appears today in Nature Human Behavior.

Modeling attention

Neuroscientists have been studying the phenomenon of selective attention for decades. Many studies in people and animals have shown that when focusing on a particular stimulus like the sound of someone’s voice, neurons that are tuned to features of that voice — for example, high pitch — amplify their activity.

When this amplification occurs, neurons’ firing rates are scaled upward, as though multiplied by a number greater than one. It has been proposed that these “multiplicative gains” allow the brain to focus its attention on certain stimuli. Neurons that aren’t tuned to the target feature exhibit a corresponding reduction in activity.

“The responses of neurons tuned to features that are in the target of attention get scaled up,” Griffith says. “Those effects have been known for a very long time, but what’s been unclear is whether that effect is sufficient to explain what happens when you’re trying to pay attention to a voice or selectively attend to one object.”

This question has remained unanswered because computational models of perception haven’t been able to perform attentional tasks such as picking one voice out of many. Such models can readily perform auditory tasks when there is an unambiguous target sound to identify, but they haven’t been able to perform those tasks when other stimuli are competing for their attention.

“None of our models has had the ability that humans have, to be cued to a particular object or a particular sound and then to base their response on that object or that sound. That’s been a real limitation,” McDermott says.

In this study, the MIT team wanted to see if they could train models to perform those types of tasks by enabling the model to produce neuronal activity boosts like those seen in the human brain.

To do that, they began with a neural network that they and other researchers have used to model audition, and then modified the model to allow each of its stages to implement multiplicative gains. Under this architecture, the activation of processing units within the model can be boosted up or down depending on the specific features they represent, such as pitch.

To train the model, on each trial the researchers first fed it a “cue”: an audio clip of the voice that they wanted the model to pay attention to. The unit activations produced by the cue then determined the multiplicative gains that were applied when the model heard a subsequent stimulus.

“Imagine the cue is an excerpt of a voice that has a low pitch. Then, the units in the model that represent low pitch would get multiplied by a large gain, whereas the units that represent high pitch would get attenuated,” Griffith says.

Then, the model was given clips featuring a mix of voices, including the target voice, and asked to identify the second word said by the target voice. The model activations to this mixture were multiplied by the gains that resulted from the previous cue stimulus. This was expected to cause the target voice to be “amplified” within the model, but it was not clear whether this effect would be enough to yield human-like attentional behavior.

The researchers found that under a variety of conditions, the model performed very similarly to humans, and it tended to make errors similar to those that humans make. For example, like humans, it sometimes made mistakes when trying to focus on one of two male voices or one of two female voices, which are more likely to have similar pitches.

“We did experiments measuring how well people can select voices across a pretty wide range of conditions, and the model reproduces the pattern of behavior pretty well,” Griffith says.

Effects of location

Previous research has shown that in addition to pitch, spatial location is a key factor that helps people focus on a particular voice or sound. The MIT team found that the model also learned to use spatial location for attentional selection, performing better when the target voice was at a different location from distractor voices.

The researchers then used the model to discover new properties of human spatial attention. Using their computational model, the researchers were able to test all possible combinations of target locations and distractor locations, an undertaking that would be hugely time-consuming with human subjects.

“You can use the model as a way to screen large numbers of conditions to look for interesting patterns, and then once you find something interesting, you can go and do the experiment in humans,” McDermott says.

These experiments revealed that the model was much better at correctly selecting the target voice when the target and distractor were at different locations in the horizontal plane. When the sounds were instead separated in the vertical plane, this task became much more difficult. When the researchers ran a similar experiment with human subjects, they observed the same result.

“That was just one example where we were able to use the model as an engine for discovery, which I think is an exciting application for this kind of model,” McDermott says.

Another application the researchers are pursuing is using this kind of model to simulate listening through a cochlear implant. These studies, they hope, could lead to improvements in cochlear implants that could help people with such implants focus their attention more successfully in noisy environments.

The research was funded by the National Institutes of Health.

 

Neurons receive precisely tailored teaching signals as we learn

by Jennifer Michalowski |

Man seated on staircase, smiling at camera
McGovern Investigator Mark Harnett. Photo: Adam Glanzman

When we learn a new skill, the brain has to decide—cell by cell—what to change. New research from MIT suggests it can do that with surprising precision, sending targeted feedback to individual neurons so each one can adjust its activity in the right direction.

The finding echoes a key idea from modern artificial intelligence. Many AI systems learn by comparing their output to a target, computing an “error” signal, and using it to fine-tune connections within the network. A longstanding question has been whether the brain also uses that kind of individualized feedback. In a study published in the February 25 issue of the journal Nature, MIT researchers report evidence that it does.

A research team led by Mark Harnett, a McGovern Institute investigator and associate professor in the Department of Brain and Cognitive Sciences at MIT, discovered these instructive signals in mice by training animals to control the activity of specific neurons using a brain-computer interface (BCI). Their approach, the researchers say, can be used to further study the relationships between artificial neural networks and real brains, in ways that are expected to both improve understanding of biological learning and enable better brain-inspired artificial intelligence.

The changing brain

Our brains are constantly changing as we interact with the world, modifying their circuitry as we learn and adapt. “We know a lot from 50 years of studies that there are many ways to change the strength of connections between neurons,” Harnett says. “What the field really lacks is a way of understanding how those changes are orchestrated to actually produce efficient learning.”

Some actions—and the neural connections that enable them—are reinforced with the release of neuromodulators like dopamine or norepinephrine in the brain. But those signals are broadcast to large groups of neurons, without discriminating between cells’ individual contributions to a failure or a success. “Reinforcement learning via neuromodulators works, but it’s inefficient, because all the neurons and all the synapses basically get only one signal,” Harnett says.

Machine learning uses an alternative, and extremely powerful, way to learn from mistakes. Using a method called backpropagation, artificial neural networks compute an error signal and use it to adjust their individual connections. They do this over and over, learning from experience how to fine-tune their networks for success. “It works really well and it’s computationally very effective,” Harnett says.

It seemed likely that brains might use similar error signals for learning. But neuroscientists were skeptical that brains would have the precision to send tailored signals to individual neurons due to the constraints imposed by using living cells and circuits instead of software and equations. A major problem for testing this idea was how to find the signals that provide personalized instructions to neurons, which are called vectorized instructive signals. The challenge, explains Valerio Francioni, first author of the Nature paper and a former postdoctoral researcher in Harnett’s lab, is that scientists don’t know how individual neurons contribute to specific behaviors.

“If I was recording your brain activity while you were learning to play piano,” Francioni explains, “I would learn that there is a correlation between the changes happening in your brain and you learning piano. But if you asked me to make you a better piano player by manipulating your brain activity, I would not be able to do that, because we don’t know how the activity of individual neurons map to that ultimate performance.”

Without knowing which neurons need to become more active and which ones should be reined in, it is impossible to look for signals directing those changes.

Brain-computer interface

To get around this problem, Harnett’s team developed a brain-computer interface task to directly link neural activity and reward outcome – akin to linking the keys of the piano directly to the activity of single neurons. To succeed at the task, certain neurons needed to increase their activity, whereas others were required to decrease their activity.

They set up a BCI to directly link activity in those neurons—just eight to ten of the millions of neurons in a mouse’s brain—to a visual readout, providing sensory feedback to the mice about their performance. Success was accompanied by delivery of a sugary reward.

“Now if you ask me, ‘How does the mouse get more rewards? Which neuron do you have to activate and which neuron do you have to inhibit?’ I know exactly what the answer to that question is,” says Francioni, whose work was supported by a Y. Eva Tan Fellowship from the Yang Tan Collective at MIT.

The scientists didn’t know the exact function of the particular neurons they linked to the BCI, but the cells were active enough that mice received occasional rewards whenever the signals happened to be right. Within a week, mice learned to switch on the right neurons while leaving the other set of neurons inactive, earning themselves more rewards.

Francioni monitored the target neurons daily during this learning process using a powerful microscope to visualize fluorescent indicators of neural activity. He zeroed in on the neurons’ branching dendrites, where the appropriate feedback signals have long been suspected to arrive. At the same time, he tracked activity in the parent cell bodies of those neurons. The team used these data to examine the relationship between signals received at a neuron’s dendrites and its activity, as well as how these changed when mice were rewarded for activating the right neurons or when they failed at their task.

Vectorized neural signals

They concluded that the two groups of neurons whose activity controlled the BCI in opposite ways, also received opposing error signals at their dendrites as the mice learned. Some were told to ramp up their activity during the task, while others were instructed to dial it down. What’s more, when the team manipulated the dendrites to inhibit these instructive signals, mice failed to learn the task. “This is the first biological evidence that vectorized [neuron-specific] signal-based instructive learning is taking place in the cortex,” Harnett says.

The discovery of vectorized signals in the brain—and the team’s ability to find them—should promote more back and forth between neuroscientists and machine learning researchers, says postdoctoral researcher Vincent Tang. “It provides further incentive for the machine learning community to keep developing models and proposing new hypotheses along this direction,” he says. “Then we can come back and test them.”

The researchers say they are just as excited about applying their approach to future experiments as they are about their current discovery.

“Machine learning offers a robust, mathematically tractable way to really study learning. The fact that we can now translate at least some of this directly into the brain is very powerful,” Francioni says.

Harnett says the approach opens new opportunities to investigate possible parallels between the brain and machine learning. “Now we can go after figuring out, how does cortex learn? How do other brain regions learn? How similar or how different is it to this particular algorithm? Can we figure out how to build better, more brain-inspired models from what we learn from the biology?” he says. “This feels like a really big new beginning.”

Unpacking social intelligence

by Jennifer Michalowski |

Experience is a powerful teacher—and not every experience has to be our own to help us understand the world. What happens to others is instructive, too. That’s true for humans as well as for other social animals. New research from scientists at the McGovern Institute shows what happens in the brains of monkeys as they integrate their observations of others with knowledge gleaned from their own experience.

“The study shows how you use observation to update your assumptions about the world,” explains McGovern Institute Investigator Mehrdad Jazayeri, who led the research. His team’s findings, published in the January 7 issue of the journal Nature, also help explain why we tend to weigh information gleaned from observation and direct experience differently when we make decisions. Jazayeri is also a professor of brain and cognitive sciences at MIT and an investigator at the Howard Hughes Medical Institute.

“As humans, we do a large part of our learning through observing other people’s experiences and what they go through and what decisions they make,” says Setayesh Radkani, a graduate student in Jazayeri’s lab. For example, she says, if you get sick after eating out, you might wonder if the food at the restaurant was to blame. As you consider whether it’s safe to return, you’ll likely take into account whether the friends you’d dined with got sick too. Your experiences as well as those of your friends will inform your understanding of what happened.

The research team wanted to know how this works: When we make decisions that draw on both direct experience and observation, how does the brain combine the two kinds of evidence? Are the two kinds of information handled differently?

Social experiment

It is hard to tease out the factors that influence social learning. “When you’re trying to compare experiential learning versus observational learning, there are a ton of things that can be different,” Radkani says. For example, people may draw different conclusions about someone else’s experiences than their own, because they know less about that person’s motivations and beliefs. Factors like social status, individual differences, and emotional states can further complicate these situations and be hard to control for, even in a lab.

To create a carefully controlled scenario in which they could focus on how observation changes our understanding of the world, Radkani and postdoctoral fellow Michael Yoo devised a computer game that would allow two players to learn from one another through their experiences. They taught this game to both humans and monkeys.

Their approach, Jazayeri says, goes far beyond the kinds of tasks that are typically studied in a neuroscience lab. “I think it might be one of the most sophisticated tasks monkeys have been trained to perform in a lab,” he says.

Both monkeys and humans played the game in pairs. The object was to collect enough tokens to earn a reward. Players could choose to enter either of two virtual arenas to play—but in one of the two arenas, tokens had no value. In that arena, no matter how many tokens a player collected, they could not win. Players were not told which arena was which, and the winnable and unwinnable arenas sometimes swapped without warning.

Only one individual played at a time, but regardless of who was playing, both individuals watched all of the games. So as either player collected tokens and either did or did not receive a reward, both the player and the observer got the same information. They could use that information to decide which arena to choose in their next round.

Experience outweighs observation

Humans and monkeys have sophisticated social intelligence and both clearly took their partners’ experiences into account as they played the game. But the researchers found that the outcomes of a player’s own games had a stronger influence on each individual’s choice of arena than the outcomes of their partner’s games. “They seem to learn less efficiently from observation, suggesting they tend to devalue the observational evidence,” Radkani says. That distinction was reflected in the patterns of neural activity that the team detected in the brains of the monkeys.

Postdoctoral fellow Ruidong Chen and research assistant Neelima Valluru recorded signals from a part of the brain’s frontal lobe called the anterior cingulate cortex (ACC) as the monkeys played the game. The ACC is known to be involved in social processing. It also integrates information gained through multiple experiences, and seems to use this to update an animal’s beliefs about the world. Prior to the Jazayeri lab’s experiments, this integrative function had only been linked to animals’ direct experiences—not their observations of others.

Consistent with earlier studies, neurons in the ACC changed their activity patterns both when the monkeys played the game and when they watched their partner take a turn. But these signals were complex and variable, making it hard to discern the underlying logic. To tackle this challenge, Chen recorded neural activity from large groups of neurons in both animals across dozens of experiments. “We also had to devise new analysis methods to crack the code and tease out the logic of the computation,” Chen says.

One of the researchers’ central questions was how information about self and other makes its way to the ACC. The team reasoned that there were two possibilities: either the ACC receives a single input on each trial specifying who is acting, or it receives separate input streams for self and other. To test these alternatives, they built artificial neural network models organized both ways and analyzed how well each model matched their neural data. The results suggested that the ACC receives two distinct inputs, one reflecting evidence acquired through direct experience and one reflecting evidence acquired through observation.

The team also found a tantalizing clue about why the brain tends to trust firsthand experiences more than observations. Their analysis showed that the integration process in the ACC was biased toward direct experience. As a result, both humans and monkeys cared more about their own experiences than the experiences of their partner.

Jazayeri says the study paves the way to deeper investigations of how the brain drives social behavior. Now that his team has examined one of the most fundamental features of social learning, they plan to add additional nuance to their studies, potentially exploring how different abilities or the social relationships between animals influence learning.

“Under the broad umbrella of social cognition, this is like step zero,” he says. “But it’s a really important step, because it begins to provide a basis for understanding how the brain represents and uses social information in shaping the mind.”

This research was supported in part by the Yang Tan Collective at MIT.

Sven Dorkenwald

Synapse-resolution connectomics

The synaptic connectivity of neurons, their connectome, is fundamental to how networks of neurons function. Sven Dorkenwald develops computational and collaborative tools to map, analyze, and interpret synapse-resolution connectomes. His work has led to large connectomic reconstructions of the fruit fly brain and parts of mammalian brains. He uses these connectomes to investigate how neuronal circuits are organized and how their structure supports complex computations.