Researchers from Cedars-Sinai’s Center for Neural Science and Medicine and the department of neurosurgery have uncovered how signals from a group of neurons in the brain’s frontal lobe simultaneously give humans the flexibility to learn new tasks—and the focus to develop highly specific skills. Their research provides a fundamental understanding of performance monitoring, an executive function used to manage daily life.
The study’s key finding is that the brain uses the same group of neurons for performance feedback in many different situations—whether a person is attempting a new task for the first time or working to perfect a specific skill. The team hopes that their study results could help to inform the future development of therapeutics. Overactive performance monitoring can manifest as obsessive-compulsive disorder, causing a person to check obsessively for errors that don’t exist. At the other extreme is schizophrenia, where performance monitoring can be underactive to a degree that a person doesn’t perceive errors or the inappropriateness of their words or actions.
“We believe the mechanistic knowledge we have gained will be critical to perfecting treatments for these devastating psychiatric disorders,” said Ueli Rutishauser, PhD, professor of neurosurgery, neurology, and biomedical sciences, director of the Center for Neural Science and Medicine, the Board of Governors chair in neurosciences, and senior author of the team’s study, which is published in Science, and titled, “The geometry of domain-general performance monitoring in the human medial frontal cortex.”
Humans can rapidly learn to perform novel tasks given only abstract rules, even if the task requirements differ drastically, the authors wrote. “To achieve this, cognitive control must coordinate processes across a diverse array of perceptual, motor, and memory domains and at different levels of abstraction over sensorimotor representations.”
A key part of this cognitive control is “performance monitoring,” a kind of internal signal, or self-generated feedback, that lets a person know they have made a mistake. So performance monitoring allows us to evaluate whether we made a mistake, are in a conflict situation, or are in an easy, or difficult task environment, the authors noted. “In real-life situations, we often need to rapidly learn to perform novel tasks with minimal instruction and time for practice, which requires domain-general performance monitoring that functions ‘out of the box.’ At the same time, we need to resolve unforeseen errors, difficulty, and other performance disturbances with task-specific measures ad hoc.”
One example is the person who realizes they drove past an intersection where they should have turned. Another example is the person who says something in conversation and recognizes as soon as the words are out of their mouth that what they just said was inappropriate.
Zhongzheng Fu, PhD, a postdoctoral researcher in the Rutishauser Laboratory at Cedars-Sinai and first author of the study, further explained, “That ‘Oh, shoot’ moment, that ‘Oops!’ moment, is performance monitoring kicking in.” These signals help to improve performance on future attempts by passing information to areas of the brain that regulate emotions, memory, planning, and problem-solving. Performance monitoring also helps the brain adjust its focus by signaling how much conflict or difficulty was encountered during the task. “So an ‘Oops!’ moment might prompt someone to pay closer attention the next time they chat with a friend or plan to stop at the store on the way home from work,” said Fu.
The medial frontal cortex (MFC) serves a central role in evaluating our own performance and decisions. However, as the authors noted, little is known about how neural representation can support both domain-specific and domain-general adaptations. “It is unknown how performance monitoring can be both flexible, to support different tasks, and specialized, to perform each task well.”
For their reported study, and to see performance monitoring in action, the researchers recorded single neurons in the human MFC while subjects performed two tasks that involved different types of cognitive conflict. “We designed our study to decipher how the brain can generalize and specialize at the same time, both of which are critical for helping us pursue a goal,” Rutishauser added.
To see performance monitoring in action, investigators recorded the activity of individual neurons in the medial frontal cortex of study participants. The participants were epilepsy patients who, as part of their treatment, had electrodes implanted in their brains to help locate the focus of their seizures. Specifically, these patients had electrodes implanted in the MFC. “Part of the magic of the human brain is that it is so flexible,” said Rutishauser.
The team asked participants to perform two commonly used cognitive tests. In the Stroop task, which pits reading against color naming, participants viewed the written name of a color, such as “red,” printed in ink of a different color, such as green, and were asked to name the ink color rather than the written word. “This creates conflict in the brain,” Rutishauser said. “You have decades of training in reading, but now your goal is to suppress that habit of reading and say the color of the ink that the word is written in instead.”
In the other task, the Multi-Source Interference Task (MSIT), which involves recognizing numerals, participants saw three numerical digits on screen, two the same and the other unique—for example, 1-2-2. The subject’s task was to press the button associated with the unique number—in this case, “1”—resisting their tendency to press “2” because that number appears twice. “These two tasks serve as a strong test of how self-monitoring is engaged in different scenarios involving different cognitive domains,” Fu added.
As the subjects performed these tasks, the investigators noted two different types of neurons at work. “Error” neurons fired strongly after an error was made, while “conflict” neurons fired in response to the difficulty of the task the subject had just performed.
“When we observed the activity of neurons in this brain area, it surprised us that most of them only become active after a decision or an action was completed. This indicates that this brain area plays a role in evaluating decisions after the fact, rather than making them.”
There are two types of performance monitoring, domain-general and domain-specific. Domain-general performance monitoring tells us something went wrong and can detect errors in any type of task—whether someone is driving a car, navigating a social situation or playing Wordle for the first time. This allows them to perform new tasks with little instruction, something machines cannot do. “Machines can be trained to do one thing really well,” Fu said. “You can build a robot to flip hamburgers, but it can’t adapt those skills to frying dumplings. Humans, thanks to domain-general performance monitoring, can.”
Domain-specific performance monitoring tells the person who made the error what went wrong, detecting specific mistakes—that they missed a turn, said something inappropriate, or chose the wrong letter in a puzzle. This is one way that people perfect individual skills.
Surprisingly, the researchers found through their study that neurons signaling domain-general and domain-specific information were intermingled in the medial frontal cortex. “Neurons encoding conflict probability, conflict, and error in one or both tasks were intermixed, forming a representational geometry that simultaneously allowed task specialization and generalization,” they wrote.
“We used to think there were portions of the brain dedicated to only domain-general performance monitoring and others to only domain-specific,” Rutishauser said. “Our study now shows that’s not the case. We’ve learned that the very same group of neurons can do both domain-general and domain-specific performance monitoring. When you’re listening to these neurons, you can read out both types of information simultaneously.”
The authors further noted, “… behavioral analyses showed that the reaction times and likelihood of making an error varied systematically with trial history. At the single-neuron level, cells encoded different types of conflict, CP [conflict probability], and/or errors in one or both tasks. Individual neurons thus demonstrated heterogeneous task specificity when encoding these performance-monitoring variables, precluding a simplistic interpretation of domain generality at the single-neuron level.”
To understand how these signals are interpreted by other areas of the brain, it helps to think of the neurons as musicians in an orchestra, Rutishauser said. “If they all play at random, the listeners—in this case the regions of the brain receiving the signals—just hear a garbled set of notes. But if they play an arranged composition, it’s possible to clearly hear the various melodies and harmonies even with so many instruments—or performance monitoring neurons—playing all at once.”
The authors further pointed out, “This “geometry of the population activity could allow downstream brain regions to read out both domain-general and domain-specific signals from the same group of neurons and to initiate corresponding physiological and/or behavioral adaptations. These findings reveal how representations of evaluative signals can be both abstract and task-specific and suggest a neuronal mechanism for estimating control demand.”