The Story of a Study of the Mind
Rebecca Saxe wants to know how our brains learn to be social.
More precisely, Saxe, an associate professor of cognitive neuroscience in MIT’s Brain and Cognitive Sciences Department, has built her career by trying to grasp how we make judgments about other people’s thoughts, a faculty dubbed Theory of Mind (ToM).
Our brains perform ToM cognition to decipher what lies behind a smile, a grimace, a catch in someone’s voice. As Saxe writes, ToM is “the mechanism people use to infer and reason about another person’s state of mind.” To track ToM, she has had to master the art of functional magnetic resonance imaging (fMRI). fMRI scanners are cumbersome and tricky to use well—and their results are challenging to interpret. But Saxe has become a virtuoso of the machine, beginning when, as a postdoc, she came to wrestle with a problem that she says still blows her mind.
“Early in development,” she explains, “human brains appear to be a collection of neurons very similar to those that form the brains of lots of other creatures.” But then “we gain capabilities that have no parallels in the animal kingdom.” How? What happens in maturing brains to create the life of the mind that distinguishes us as human beings?
That’s a huge question, the work of a lifetime. But a close look at just one of Saxe’s experiments hints at what it takes to distill meaning from the measurements—and offers a glimpse into the working life of a scientist at the top of her game.
Saxe, a 33-year-old Canadian, has been grappling with brains since the 1990s, beginning as an undergraduate at Oxford. She began to focus on the concept of ToM in graduate school at MIT, working in the lab of Nancy Kanwisher ’80, PhD ’86, where she identified the brain systems that are activated when someone is thinking about someone else’s state of mind. Cognitive neuroscientist Jon Simons of the University of Cambridge, who specializes in the study of memory, says Saxe has designed thoughtful experiments that yielded interesting and reliable data. “It’s tempting to think great science is these big breakthroughs,” he says. “[But] this is the way good science is done. It’s doing rigorous studies in great detail that creates results that last.”
Saxe took an important step in 2005, when she began to wonder whether she could study children as young as five to see how the ToM system develops over time. The roadblock? At that time, very few researchers in the world attempted to put kids in an fMRI machine, which requires people to sit still for 40 minutes or more. But one such researcher was Kevin Pelphrey, then at Duke University. Saxe wrote to him in October of 2005—”blind,” she says—proposing an experiment that would scan children’s brains as they processed stories centered on people’s thoughts. “I told him that I’d write the experiment and you can do the scans at Duke,” she recalls. Pelphrey, whom Saxe rates as one of the most generous of collaborators, said sure, come on down—but by February. So Saxe had just weeks to take her germ of an idea and turn it into a workable protocol.
“I and my boyfriend sat at the kitchen table,” she says, “and I wrote little bits of stories that had people’s thoughts in them.” She devised a dozen narratives, adding drawings to accompany them. Subjects would be exposed to three segments per narrative for 20 seconds each: a description of a physical setting, with a hand-drawn picture to illustrate it; a description of human characters; and a “mental” segment that offered information about what was going on in their minds. To ensure that the results wouldn’t be influenced by a subject’s reaction to the order of the segments, each kid in the study would hear the same stories, but with the segments arranged in different orders. After listening to each story, the children would be asked a yes-or-no question that required them to infer what the leading character might do next. Then, following a brief pause, another story would begin, and so on, until the subject had gone through the full protocol of 12 separate narratives, in scanning sessions that lasted 40 minutes or so.
Days after she completed the experimental design, Saxe went to Duke. The first subject, a girl about 10 years old, arrived around midmorning, and the team settled her into the scanner. In the control room Saxe could see the child’s two small feet poking out of the mouth of the machine. As the scanner began to sound its usual hum and click-click-click, Saxe was nervous. Well aware that any new experiment has multiple paths to failure, she knew she faced a more fundamental risk as well. Some research had suggested that kids younger than five can already make ToM inferences. So there was a good chance that she and her collaborators would find nothing at all—the ToM regions in these kids might look just like those in older people.
Worst of all, as she ran her first scan, Saxe had no way to check on whether she would find anything of interest. Though fMRI technology is popularly imagined (or feared) as a kind of probe that you can use to simply read the contents of people’s thoughts, the reality is far less dramatic, and much more demanding of the researcher.
Some of the problems are purely technical. fMRI machines measure brain activity only indirectly. When neurons fire, they suck in oxygen to prepare themselves to fire again. That evokes a demand for more oxygen—delivered by blood. When a region of the brain calls for more fuel, the local ratio of oxygenated to deoxygenated blood shifts by a fraction. Oxygenated blood has different magnetic properties from the deoxygenated version, and fMRI machines measure the resulting slight changes in local magnetic fields.
But that’s no mean feat. The brain shifts slightly in the skull with every breath and heartbeat, and people—especially kids!—don’t lie perfectly still for an hour. “We have a wiggling, pulsing, wobbling machine consuming vast quantities of oxygen for many reasons, on top of which we are trying to measure a tiny signal,” says Saxe. Then there’s the timing problem: neural activation takes fractions of a second, but changes in blood oxygen levels occur over six seconds or more—and the levels don’t always return to the same base number, adding to the mess. Hence, says Saxe, “we do motion correction. We do spatial smoothing to get rid of some of the noise. We filter temporally anything that happens over 10 minutes or longer”—the time scale on which artifacts like the scanner heating up would appear in the data. It takes hours of analysis to locate a signal—all just to begin to figure out what any measurement might mean.
By late morning, the first child completed her slate of stories, and then the team repeated the sequence with the second and last subject for the day, a nine-year-old boy. When he climbed down from the scanner, most of the group made plans to head out for dinner.
Saxe stayed behind. Now began the stoop-labor phase of the experiment: transforming the raw numbers from the scan into a form that could be analyzed. Hours passed. Saxe’s colleagues ate and returned. She remained at the computer. By late evening, she’d completed the first-pass processing. Motion artifacts—gone; machine noise—under control; frame after frame matched prompt to neural action.
What she saw allowed Saxe to exhale: she had data. The team had demonstrated that it is possible to get usable findings from children most researchers felt were too fidgety to image. More important, as they gathered scans from subjects as young as six, Saxe and Pelphrey found something new. “The older kids looked pretty much like adults,” Saxe says. Their ToM regions lit up when they heard a segment that forced them to think about what someone else was thinking but not during the parts of the stories that merely described somebody doing something. But the little kids’ scans were different. “Their brain regions responded to anything about people”—not only anecdotes about people’s mental states but anything with people present.
There are several ways to explain that result, but Saxe favors what she calls a “crazy idea”: that “the brain starts out with a general-purpose faculty for thinking about social interaction, and then specializes.” Her team began to discern that as we grow up, our brain regions “drop some jobs to get expert in others.”
This claim is still tentative, but subsequent work in Saxe’s lab and elsewhere has strengthened it. Now, Saxe says, “we think we have a pattern of development. Now we want to know how and why this development happens and what can make it change.” To find out, her lab is running studies on different paths to ToM specialization. One is looking at children on the autism spectrum; another is examining blind children, who engage others’ thoughts through hearing rather than sight; and a third is investigating deaf children, some of whom are exposed to language relatively late if their parents don’t know how to sign. None has yet yielded definitive results.
Saxe is no longer a footloose postdoc. She heads a lab now, and she’s still not used to everything it takes to do science on an increasingly large scale. More than a dozen researchers are involved in just one of the new studies, she says. There’s a note of wonder—disbelief, really—as she adds: “Back in the day, it was just me drawing on my table. Now we need one person at 100 percent time just to track that project.”
Still, as she says that, she grins. She and her colleagues published the results of the Duke study in 2009, but there’s something that didn’t come across on the page—an aspect of science that can’t be captured in any affectless tally of the data. That night in North Carolina, “that first moment when it worked?” she says. “That was thrilling.”
A sample story from the children’s ToM experiment
Out behind the big red barn at the edge of the walnut grove is the most magnificent pond in the neighborhood. It is wide and deep, and shaded by an old oak tree. There are all sorts of things in that pond: fish and old shoes and lost toys and tricycles, and many other surprises.
Old Mr. McFeeglebee is a gray wrinkled old farmer, who wears gray wrinkled old overalls, and gray wrinkled old boots. He has lived on this land his whole life, longer even than most of the trees. Little Georgie is Mr. McFeeglebee’s nephew from town.
Mr. McFeeglebee doesn’t want any little boys to fish in the pond. But little Georgie pretends not to notice. He likes fishing so much, and besides, he knows he can run faster than anybody in town. Georgie decides to run away really fast if Mr. McFeeglebee sees him fishing.
What do you think? Does little Georgie fish in the pond? [pause] Good job! Time for the next story!
Geoffrey Hinton tells us why he’s now scared of the tech he helped build
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
ChatGPT is going to change education, not destroy it
The narrative around cheating students doesn’t tell the whole story. Meet the teachers who think generative AI could actually make learning better.
Meet the people who use Notion to plan their whole lives
The workplace tool’s appeal extends far beyond organizing work projects. Many users find it’s just as useful for managing their free time.
Learning to code isn’t enough
Historically, learn-to-code efforts have provided opportunities for the few, but new efforts are aiming to be inclusive.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.