When people do something mentally engaging, their brains do not just get “more active.” The patterns of activity across many regions can change in structured ways. A study in PNAS proposes a simple way to describe this shift using two properties: informativeness and compressibility. Surprisingly, the authors report that during coherent story listening, brain activity patterns were both more informative and more compressible than during temporally scrambled versions of the same story or rest.
A counterintuitive pattern: more detail and more redundancy
The study analyzes functional brain imaging data where participants listened to an intact story, listened to temporally scrambled versions of the same story (including paragraph- and word-scrambled conditions), or rested. They define “informativeness” as the maximum across-participant timepoint decoding accuracy (how well a model can predict when in the session a brain snapshot was recorded). They define “compressibility” as the number of features/components needed (after dimensionality reduction) to reach a fixed decoding-accuracy threshold.
Across conditions, the intact story listening condition had the highest peak decoding accuracy and also required fewer features to reach a reference decoding threshold. In their framing, this means the brain patterns were both information-rich and robust to noise. In “High-level cognition is supported by information-rich but compressible brain activity patterns”, the authors also note the paper was published online on 23 August 2024 (issue date 27 August 2024).
What “informativeness” and “compressibility” mean in this paper
These words can be misleading if we import everyday meanings. In this study, informativeness does not mean “the brain contains more facts.” It means the maximum accuracy for across-participant timepoint decoding. In plain words, it is how well a model can predict when in the story the brain snapshot was recorded. Compressibility does not mean “less complex thoughts.” It means the same decoding threshold can be reached with fewer features/components after dimensionality reduction.
One helpful intuition is that a system can be both detailed and robust if it uses structured redundancy. The authors frame this as a tension between information-rich patterns and robustness to noise or corruption.
Why a coherent story may produce more shared structure
The authors interpret the intact story as a “cognitively richer” condition than temporally scrambled story audio or rest. A coherent story tends to constrain what listeners are tracking, because it provides context. Over time, that shared context can make different people process the same moments in more similar ways, which helps across-participant decoding.
This fits with related narrative-comprehension work that emphasizes stimulus-locked network dynamics. For example, a Nature Communications paper on the default mode network during narrative comprehension introduced inter-subject functional correlation to isolate stimulus-dependent inter-regional correlations between brains exposed to the same stimulus: dynamic reconfiguration of the default mode network during narrative comprehension.
It also connects to a later Nature Communications study that found the best decoding during intact story listening when using high-order dynamic correlation features, while scrambled-story decoding worked best with lower-order or non-correlational features: high-order dynamic correlations during story listening.
If you want a broader angle on how modeling choices affect what we can decode from brain data, you can compare with this related Gromeus article: AI extracts mental images from human brain activity, and this updated article on how AI detected brain sex differences in white matter.
What this does not show (and how to read similar headlines)
This type of work is not “mind-reading.” The decoding target here is a timepoint label under tightly controlled experimental conditions, not the free-form content of a person’s thoughts. Also, these metrics depend on modeling choices: the chosen decoder, preprocessing, dimensionality reduction method, and the threshold used to define “compressibility.”
The safe takeaway is narrower: under some tasks, the brain’s recorded activity patterns can become both easier to decode and easier to compress, at least in the sense used by this paper. That is an interesting descriptive claim about brain dynamics, not a claim about intelligence, moral value, or clinical diagnosis.
Sources and related information
PNAS – High-level cognition is supported by information-rich but compressible brain activity patterns – 2024
This is the primary source for the claim that brain activity patterns during intact story listening were reported as both more informative and more compressible than during temporally scrambled story audio or rest, based on a decoding-and-compression framework: information-rich but compressible brain activity patterns.
Nature Communications – Dynamic reconfiguration of the default mode network during narrative comprehension – 2016
This paper provides peer-reviewed context that coherent narratives can elicit stimulus-locked network dynamics shared across people, using inter-subject functional correlation: dynamic reconfiguration of the default mode network during narrative comprehension.
Nature Communications – High-level cognition during story listening is reflected in high-order dynamic correlations in neural activity patterns – 2021
This related paper supports the broader idea that story listening can be captured by higher-order patterns in neural activity dynamics, and it links decoding performance to the type of neural features used: high-order dynamic correlations during story listening.
