People who live with animals often feel these creatures have some perception of our intentions and feelings—for example, that a pet senses when we’re sad and tries to cheer us up. But finding out whether this is true or unwarranted anthropomorphising requires a mind-reading ability that has so far remained outside the grasp of psychology and neuroscience. In research with humans, investigators can simply ask study participants about their perceptions of others’ beliefs and motivations. With nonhuman animals, how does one tease out the ability to infer another creature’s point of view, known as “theory of mind,” from simply responding to cues that the animal has learned to associate with a particular outcome?
Michael Platt, who has studied cognition in monkeys for decades, and his team, then at Duke University, recently harnessed a combination of brain monitoring and machine learning in a bid to get closer to answering this question. They devised a computer game in which two monkeys faced each other and used joysticks to drive circles on a computer screen, choosing either to go straight or to swerve. As in a real-life game of chicken, if both monkeys went straight, both lost—meaning that they didn’t receive a reward in the form of a sip of juice. If a monkey swerved while its opponent steered straight, it got a small juice reward, and if it steered straight while the other monkey veered, it won, receiving a larger portion of juice. If both monkeys veered, they received the small sip of juice plus a bonus reward that varied from trial to trial.
The monkeys learned to play the game with sophistication, adapting their strategies to those of their opponents.
The rhesus macaques (Macaca mulatta) in the experiment were restrained in their seats and played the game over and over, with a few variations. Sometimes, instead of facing off against a live opponent that they could see on the other side of a computer screen, they played alone against the computer, or against a computer with a decoy monkey who was present but wasn’t playing. The sizes of the rewards were varied, as was the clarity of the visual signal the monkeys saw on the screen indicating how their opponent was steering.
The monkeys learned to play the game with sophistication, Platt says, adapting their strategies to those of their opponents in a way that is consistent with a principle in game theory called the Nash equilibrium, in which individuals settle on the best strategies available to them in light of their opponents’ strategy. Using machine learning, the researchers found that they could successfully predict the monkeys’ actions only when their model incorporated details of their opponents’ circumstances and behavior, such as the rewards available to the opponents and how they steered their on-screen circles, indicating that the monkeys were incorporating these cues into their decision making. While he’s observed complex behaviors from primates in the wild, Platt says, “I’m always surprised when we are able to elicit behavior in a task like this one, which is super complicated.” The animals’ behavior in the wild clearly doesn’t include playing video games, but Platt, who’s now at the University of Pennsylvania, says that the game likely draws on the same cognitive resources that monkeys use to make strategic social decisions such as when to offer or accept grooming.
One factor that bore on the monkeys’ play was their preexisting social hierarchy, notes Wei Song Ong, a postdoc at UPenn who worked in Platt’s lab and is the study’s first author. The animals knew one another, she explains, as they’re housed together in individual enclosures that enable them to see and hear one another, and they establish hierarchies that researchers can determine by putting them in the same room, equally distant from a piece of fruit: the subordinate monkey will allow the dominant monkey to grab the food.
In the game, “the action of the same monkey is different [depending on] whether or not he’s playing a dominant monkey or a subordinate monkey,” Ong explains. “But the subordinate monkeys all play very similarly, and the dominant monkeys play very similarly, which basically means that if you’re a mid-ranking monkey, you change your strategy depending on who you’re playing with.” Moreover, the subordinate monkeys were more attuned to their opponent’s strategies. “The subordinate is always trying to learn about the dominant, and the dominant mostly just does whatever they want,” showing a lesser propensity to learn about their opponents, Ong notes.
As the monkeys played, electrodes monitored the activity of single neurons in two areas of their brains thought to correspond with brain regions used during social decision making in humans: the superior temporal sulcus (STS), which is thought to be analogous to humans’ temporoparietal junction, an area involved in thinking about the beliefs or thoughts of another person; and the anterior cingulate gyrus (ACCg), an area that’s active when an animal sees a conspecific receive a reward and is associated with empathy in humans. The results revealed that the animals drew on both the ACCg and the STS during the game, although more so on the STS, suggesting that “cooperation is a strategic act, not an empathetic act,” Ong says.
By monitoring the firing of single neurons, says Ziv Williams, a neuroscientist at Massachusetts General Hospital and Harvard Medical School who was a peer reviewer on the new study, Platt’s group was able to show “that there are certain dedicated neurons that really are very good at representing these predictions. And that’s something that has never really been shown before.”
“This is really a nice study,” says Liesbeth Sterck, who studies social behavior and cognition in primates at Utrecht University in the Netherlands and was not involved in the work. There’s a gap, she says, between the type of research she does—observing primates’ behaviors in more natural settings—and neuro-
science, which has focused on very simple actions in the lab. “This is the kind of paper that’s really bridging” that gap, she says, by homing in on the neuroscience behind more complex behavior.
Daniel Povinelli, a biologist at the University of Louisiana at Lafayette who studies behavior in chimps, humans, and other animals, argues that the fact that the monkeys seem to make those predictions doesn’t show whether the animals have any understanding of one another’s mental states. “I can keep very complicated track of what another organism [does] in certain [kinds of] situations, and I can make predictions, therefore, about what they’re going to do, without ever reasoning about their internal private motivations for doing so,” he says.
Platt concedes that the study didn’t “precisely test” for theory of mind. But by showing that the monkeys take into account factors such as one another’s past behavior and incentives, he and his colleagues have established that the animals have the “building blocks of theory of mind,” he says.
Whatever the role of theory of mind—or lack thereof—in enabling the monkeys to make those predictions, Platt argues that it isn’t an all-or-nothing proposition. Autism is sometimes associated with impaired theory of mind, but Platt argues that even among neurotypical humans, “there’s a lot of variation in the degree to which any of the components that together make up theory of mind are actually utilized and contribute to behavior. So very few people go through the fullest steps of reasoning, in terms of intentionality with regard to other people. . . . Certainly, my wife thinks that I don’t engage in a high degree of depth of theory of mind in my own behavior.”