Major neuroscience update: Zeroing in on ways to measure and characterize cognition

A new research paper discusses an absolutely amazing aspect of cognition in the brain. In essence, when engaged in significant cognitive effort, like listening to a story, the brain compresses a huge amount of information into a small set of electrical signals. 

One author described the stunning degree of data compression like this: “If human language was similarly efficient, I’d be able to tell you the details of every Wikipedia article just by speaking a dozen or so words.” That has to be either a gross exaggeration, or the high degree of compression is incomprehensible to me. I do not see how this can be remotely possible.

If this research holds up on replication, it is mind-blowing.  PsyPost reports:

A new neuroimaging study reveals that when we engage in more complex cognitive tasks, our brain activity becomes not only richer in detail but also more streamlined. The findings suggest that the brain adjusts its patterns of activity to match the demands of the task, allowing for more efficient processing during mentally challenging activities.

The study, published in the Proceedings of the National Academy of Sciences, was driven by a desire to understand how the brain manages different cognitive demands. Previous research by the same team had revealed the brain’s remarkable ability to reconstruct missing data from minimal measurements, raising questions about why the brain can generate such detailed and efficient activity patterns with limited input.

“Several years ago, my co-author and graduate student at the time, Lucy Owen, and I came out with a precursor to this study, where we found something very surprising,” explained study author Jeremy Manning, an associate professor of psychological and brain sciences at Dartmouth College and director of the Contextual Dynamics Lab.

“At the time, we were working with neurosurgical patients who had electrodes implanted in their brains to monitor for seizure activity. A challenge with working with those recordings is that our brains contain roughly a hundred billion neurons, but we can only safely implant around a few hundred wires into someone’s brain. So there is a massive undersampling problem: for every measurement we take, we miss roughly a billion others! We wanted to understand how much of that ‘missing’ data we could reliably and accurately reconstruct using statistical ‘hacks.'”

“We were very surprised to find that just a few hundred measurements from an essentially random sampling of locations throughout someone’s brain could give us enough information to fill in an accurate guess about activity patterns throughout their entire brain, at millimeter-scale resolutions (roughly on par with the best fMRI available today), but at millisecond-scale sampling rates (roughly 1000 times faster than fMRI),” Manning said. “If human language was similarly efficient, I’d be able to tell you the details of every Wikipedia article just by speaking a dozen or so words.”

To assess the informativeness and compressibility of brain activity, the researchers used advanced computational techniques. They measured informativeness by analyzing how much specific information about the task was reflected in participants’ brain activity. Compressibility, on the other hand, was evaluated by examining how efficiently the brain’s activity patterns could be represented using fewer components or data points. A highly compressible brain pattern is one in which fewer pieces of information are needed to reconstruct the full activity.

“In the world of machine learning, the ability to reconstitute a detailed pattern from its parts is called ‘compression,'” Manning told PsyPost. “Highly compressible patterns can be accurately rebuilt from just a tiny sliver, like reconstructing the complete text of a novel from just a single word. Another related property is called ‘informativeness.’ This refers to how ‘expressive’ a sequence of patterns is– akin to the length of a novel.”

The researchers uncovered two key findings. First, brain activity was more informative and compressible when participants engaged in the more demanding task of listening to a coherent story compared to the scrambled story or resting conditions. This suggests that during higher-level cognitive tasks, the brain produces detailed, information-rich activity that is also organized efficiently. In simpler tasks, or during rest, the brain’s activity is less organized and contains less specific information.

Second, the study found that these brain patterns became more informative and compressible over time as participants continued to listen to the coherent story. As the narrative unfolded, the brain seemed to adapt by refining and optimizing its activity patterns. This pattern was less pronounced in the scrambled conditions, where the lack of a coherent structure in the story likely led to less mental engagement and, consequently, less organization in the brain’s activity.

“Going into this study, we would have guessed that ‘compression’ and ‘informativeness’ would have changed in opposite directions,” Manning said. “That would be analogous to either being able to reconstruct short novels from just a few words (perhaps under certain cognitive circumstances — representing high compressibility but low informativeness), or being able to reconstruct longer novels from more words (perhaps under different circumstances — representing low compressibility and high informativeness). Finding that compression and informativeness change in the same direction helped us to understand that these two aspects of how our brains respond can vary independently from each other.” 

“We looked at data from a little over 100 participants, using one set of experimental conditions, and using one method for measuring brain activity,” Manning noted. “Although it is tempting to generalize to ‘all humans and circumstances,’ the true test of these findings, as with any study, will be in how well they replicate and generalize.”  

“We are deeply curious about understanding fundamental questions about how our brains work, and what makes us ‘us.’ This line of work is a tiny part of a much broader literature aimed at uncovering the neural basis of thought,” Manning said. “My website is www.context-lab.com. It has links to all of my lab’s publications, data, and software, along with some open courses that could be of interest to people who want to learn more about this stuff.”

In their research paper (behind a paywall), the authors describe the significance of their research like this:

How our brains respond to ongoing experiences depends on what we are doing and thinking about, among other factors. We examined two fundamental aspects of brain activity under different cognitive circumstances: informativeness and compressibility. Informativeness refers to how specific the brain activity we measure at a given moment is to whatever was being done in that particular moment. Compressibility is a measure of how redundant the activity patterns are. We found that when people were engaged in higher-level cognitive tasks, their brain activity was both more informative and more compressible than when they were engaged in lower-level tasks. Our findings suggest that our brains flexibly reconfigure themselves to optimize different aspects of how they function according to ongoing cognitive demands. 

So, this paper is saying that during high-level cognition (high cognitive load), the brain dynamically, i.e., cognitive load-sensing, produces detailed, information-rich activity that is organized and compressed with astounding efficiently. The effect was more pronounced in higher-order brain networks associated with complex functions like decision-making and memory. 

As participants continued engaging in a complex task like listening to a coherent story, brain patterns became more informative and compressible over time. That suggests the brain adapts and optimizes data process while engaging in a significantly cognitive loaded task. In essence, the brain's data compression ability seems to become more efficient and effective during complex, engaging cognitive tasks, allowing for rich information processing while maintaining compressible, organized activity patterns.

This research challenges the researchers' initial hypothesis that informativeness and compressibility would trade off against each other. Instead, they both change in the same direction during complex cognitive tasks. That is counterintuitive, at least to me. That alone ought to prompt real quick testing in other labs this to see if these results replicate and get either verified or debunked.

Germaine mental status: Mind blown