Compressing prediction errors in the cerebral cortex
Chapter 8 endnote 18, from Lisa Feldman Barrett.
Some context is:
Your cortex learns concepts by separating similarities from differences, as you saw in chapter 6. It integrates information across vision, hearing, interoception, and the other sensory domains, compressing them into efficient summaries.
A recent paper by Barbara Finlay, the neuroscientist who debunked the “triune brain” as a myth in chapter 4, illustrates why mammalian brains are well equipped to learn concepts. If you could unfurl the cerebral cortex and stretch it out like a dinner napkin, you’d see that the upper layers (layers 2 and 3), which are important for processing prediction errors, are wired to integrate sights, sounds, and other sensory inputs and to compress them into efficient, multisensory summaries. In sensory regions, the upper layers contain a larger number of smaller neurons with fewer connections (for processing details). These prediction errors pass to regions with increasingly larger and better connected neurons until they reach the regions in the interoceptive and control networks in the front of the brain, which contain fewer neurons that are much larger and more densely interconnected. The result is that many details are condensed into multisensory summaries. As we learned in chapter 6, this wiring allows a brain to learn concepts and predict with them to form perceptions and guide actions. And as we learn in chapter 12, this wiring is enhanced in humans, providing us with greater computational power to integrate prediction errors from visual, auditory, interoceptive, and other sensory systems to develop an elaborate, detailed (and therefore granular) conceptual system.
Notes on the Notes
- Finlay, Barbara L., and Ryutaro Uchiyama. 2015. “Developmental Mechanisms Channeling Cortical Evolution.” Trends in Neurosciences 38 (2): 69–76.