Back and Forth: Understanding the Neuroscience of Social Interactions

Q&A with Read Montague, Ph.D.
Kayt Sukel
November 7, 2019

Read Montague, Ph.D.
Director, Human Neuroimaging Laboratory
Virginia Tech
Dana Grantee: 2005-2008

What is happening in the brain as you interact with another person? Given neuroscience’s past technical limitations, it’s a harder question to answer than one might think. Read Montague, Ph.D., director of the Human Neuroimaging Laboratory at Virginia Tech, has spent his career trying to understand the neural mechanisms underlying human interaction during social games, even pioneering a new neuroimaging technique called hyperscanning that allows scientists to simultaneously observe the brains of two people as they interact with each other during different social exchange games. Here, he discusses why social neuroscience is often considered the “red-headed stepchild” of the greater neuroscience realm, the challenges of acclimating people with autism spectrum disorder to the magnetic resonance imaging (MRI) scanner, and why social games are showing themselves to be an effective diagnostic tool for multiple psychopathologies.

What first prompted you to study how people interact with one another?

I’ve always been interested in interpersonal exchange and building models of ourselves and how we interact with other people. It’s a recursive problem, right? I have to be able to model you—and, at the same time, model you modeling me. Think of a job interview: if you sit across the desk from me during that interview, you need to have some internal model of me and what I’m looking for from you, so you don’t offend me or say something weird. Unless that’s your intent, of course. Then you have to have a model of my model of you so if, as we talk, it isn’t what you want it to be, you can send some signals over so I can change it. This process of modeling bounces back and forth between two people during any social exchange. And there’s a lot of neural machinery surrounding that.

How are these models of social exchange altered in autism spectrum disorder (ASD)?

We know very little about it, just that something is perturbed. And those that do have some good ideas about what’s perturbed don’t necessarily agree on what’s wrong. Uta Frith, at the University College London, would tell you there is something in the first-person modeling that’s off. In other words, autistic kids are good at modeling the things required to get responses out of other people but can’t model themselves properly in the exchange. Others might say the issue is modeling the person you are interacting with. But, of course, autism has the heterogenous problem: This is a condition where there are a lot of different ways to break the software. So, it may not be just one problem.

That said, it is a hard problem to investigate. We must be able to model to effectively communicate with one another. And scientists are now using very modern methods to model this recursive interaction in a simple setting.

One method you used is hyperscanning MRI. Can you tell me what it is and why it’s suited to capture the nuances of this social modeling?

Well, there’s nothing about MRI that is particularly suited to studying the brain. It’s the first non-invasive safe way to record human brain activity, but it has limitations. The signal is slow, the change in blood flow response in your brain is slow, and you have to lie in the belly of a 30,000-pound superconducting magnet–not exactly a natural setting–to get that signal. There are limitations to what it can offer. Early on, I wanted to be able to study two people at the very moment they were interacting. So, I created some software that could link up two scanners, with different people in each, to synchronize the activity together so we could see what was going up as people were doing social exchange games. That’s what I call hyperscanning.

When I started doing that, back in 2001, Nature magazine made fun of me. They asked why anyone would ever want to study two brains at a time. Why can’t you just study someone interacting with a computer? I remember thinking, “Why would you not want to do it?” If you want to know how the brain works in these kinds of interactions, it’s a better experiment to put them into a real exchange with someone else so you can see what’s going on.

What occurs during one of these hyperscans?

You have two people, one in each scanner, and they are basically playing social games. We use a menagerie of tasks, but they are all some form of a social exchange game. The thing that the people usually socially exchange is money.

If you are playing a game with me, I give you some money under a set of rules and then you send me some money back. One of these scenarios is what we call a trust game. Every round of the game, I’m given $20. I can keep that, and the round is over. Or I can send some fraction of that sum to you. So maybe I send you $10. On the way over to you, the game triples it to $30. Now you have control and you can decide to send back some of that to me. Maybe you split it and send me $15. Maybe you send me nothing. I took the risk initially by sending you that $10, I trusted you, and you send me nothing. You can see how the social exchange will change over time as we play multiple rounds.

It’s a very easy game to play. Everyone knows the rules. It’s like a trading game in a marketplace, with the complexities of trade stripped away. People know when the other person is being unfair–and they really react to it. You can see those reactions in the brain in a region called the cingulate cortex. In fact, we’ve now developed functional MRI signatures that are associated for different parts of the brain and the different ways we are modeling the game. We can then go up a level and then compare those signatures to what’s happening in healthy people as they play, as well as people with different psychopathologies, including ASD.

So, what’s happening in the brain when typical participants play this trust game versus people diagnosed with ASD?

You can see differences behaviorally in the way ASD kids play the game. They don’t sense the social signals as much. Let me give you an example. Suppose you send me $2 and you keep $18 from that original $20. The $2 triples to $6 and I keep all six. I send nothing back. On the next round, you send me a little bit more. You send me $4 and it triples to $12. I take it all and I’m taking all the money. I’m sending you a signal–and this kind of signals generate big brain responses as well as big behavioral responses in healthy people. What I did is not fair. It’s a social breach. And we see a response in the cingulate cortex as well.

But in individuals with ASD, there isn’t that response, behaviorally or in the brain. And if you track the response in cingulate cortex, you can distinguish ASD and non-ASD kids at a pretty sensible rate. The idea, broadly, is that you can use these social games to probe across boundaries of traditionally defined psychopathologies in the context of trying to develop biological signatures of them.

How did you acclimate study participants diagnosed with ASD to play these games in the MRI?

That was a lot of work. We developed relationships with clinics in Tuscaloosa, Alabama and at the Texas Medical Center. We took movies of the people the participants would be interacting with and let them see them. They practiced being confined in the MRI using plastic trash cans onsite. This is all before they came here to be scanned. We also got them used to the sound in the scanner before they got here. These particular kids were high functioning, with an average IQ of about 103, and they understood the rules of the game. Not all them worked out, some panicked in the scanner. But we got better at learning tricks so that they could be fairly comfortable while we scanned them.

How might investigating these social exchange paradigms maybe help diagnose ASD?

This particular trust game is very sensitive as a behavioral probe. As we exchange money, one person is making a proposal, the other is making a response. This happens across thousands of pairs. We can plot out those proposals and responses and run statistical learning methods and natural categories just drop out.

When you start putting people with different psychopathologies in the responder role, you see different conditions cluster based on how they respond to a proposal. People with depression are in one cluster. Those with ASD are in another. Kids with attention deficit hyperactivity disorder end up in yet another cluster. And medicated versions of those kids are in another one. It’s a good probe of traditionally defined disease boundaries just based on how people behave.

We can also use the game to generate biological signatures. We started with MRIs, but it’s not nearly granular enough. We now use the game in people who are undergoing surgery to have a deep brain stimulator implanted or for epilepsy, and we can measure neurotransmitter fluctuations at rates like 10 milliseconds per estimate—much faster than what you can see in an MRI.

How could this game be used in the clinic for diagnostic purposes?

It could be just one more arrow in your quiver. A patient comes in and plays a little game that takes about four minutes. You can’t cheat at it. When you fill out typical questionnaires, you are kind of performing. You are performing for your image of yourself and performing for the clinician, an authority figure, at the other end. It’s very hard. But this game is more implicit, less cognitive in an overt sense. And it can give us a quick shot about whether a person meets that cluster for a particular condition. As more labs use these kinds of games, and collect data on them, it could become part of a diagnostic inventory that doesn’t require you to make the same kind of assumptions that these questionnaires do.

Social neuroscience is sometimes criticized for being too messy to effectively study. What would you say to that?

I think it’s because it’s just been so difficult to effectively study. My background is in math and physics. I moved into social things because people like me didn’t do that. And, of course, my friends made fun of me. I think they still do. They asked me, “Why would you do that?” But social interactions are the most interesting things that people do, period. It’s the thing that breaks first in disease states. It infiltrates every single aspect of what we call cognition. Social neuroscience holds some of the most interesting questions in neuroscience.

It’s taken us a long time to take social neuroscience seriously because it is such a complicated thing to investigate. It’s a lot easier to shine a laser into an eye and study photoreceptors and then pretend the brain is putting together the visual world in some simple way. In the past, we didn’t have the tools to really look at social interactions. That’s what inspired me to make a tool that could, hyperscanning, and start using it.

I think we’ll see more of these tools in the future. With access to fast and cheap computing, we can start to put wearables on people and follow them as they go out into the wild. There’s the possibility to get some interesting data from people interacting with others in context out in the world. We could even pump a low-dimensional experiment into a person’s phone. So, I think there’s a lot of data coming that is going to upgrade how we think about social neuroscience.

How do you plan to follow this body of work?

We hope to take our MRI results to help inspire a whole range of more granular experiments where we can use epilepsy wires, the wires being monitoring for depth recordings in humans as part of their plan for neurosurgery, or we use the acute setting of the operating room when surgeons are putting a deep brain stimulation electrode in, to go down and record in that place for some amount of time. We’ve developed a new approach, built on the back of the stuff I started with my Dana Foundation funding that was then picked up by Wellcome Trust, to record dopamine, norepinephrine, and serotonin at sub-second time scales. We’ve never been able to record these neuromodulators at those kind of timescales before. I consider it a breakthrough. The disease burden alone of any of those three neuromodulatory systems is immense. So being able to study this at this level could have implications for people who are diagnosed with depression, psychosis, drug addiction, or Parkinson’s disease. It’s exciting.


Kenneth T. Kishida, Josepheen De Asis-Cruz, Diane Treadwell-Deering, Brittany Liebenow, Michael S. Beauchamp, P. Read Montague. Diminished single-stimulus response in vmPFC to favorite people in children diagnosed with Autism Spectrum Disorder. Biological Psychology, 2019; DOI: 10.1016/j.biopsycho.2019.04.009