Day 10 - Analogous Minds (Henry Kennedy, Josie Clowney, Barbara Webb)

 Hierarchies, claustrum, circuits, and bee dances

The day opened with Henry Kennedy on a humbling premise: we still do not really understand the cortex. At the center of the problem is hierarchy.

Cortical hierarchy is often described as a progression: information moves “up” through increasingly complex representations, building larger receptive fields and integrating more context. The classical visual system - from David Hubel and Torsten Wiesel’s receptive field work in primary visual cortex onward - remains the textbook example.

But Henry’s point was that the real anatomy makes this picture much messier. There are far more feedforward than feedback pathways, but both are everywhere. Information does not simply travel “up” and then “down” - much of it unfolds simultaneously, with recurrent and reverse hierarchies constantly interacting.

This naturally led into predictive coding. In that framework, feedforward signals carry prediction errors, while feedback signals carry predictions. It is an elegant idea, but Henry was cautious: it explains part of the story, but not all of it. Prediction itself remains slippery - prediction of the world? Prediction of our own actions? Prediction of sensory consequences?

And then came the claustrum.

A structure Henry called “mysterious” - anatomically close to cortex, developmentally related to it, but sitting underneath in the white matter. What makes it strange is its connectivity: almost every cortical area strongly connects to it. Inject a tracer almost anywhere in cortex, and you label claustrum. So what is it doing? An information mixer? A salience detector? A consciousness hub? A sleep regulator? No one seemed ready to commit.

Its position - physically and functionally sitting in the middle of everything - makes it unique. The claustrum feels like one of those places in neuroscience that everyone acknowledges but no one quite knows how to think about.

Later, Josi Clowney shifted the discussion to a different question: is there a trade-off between interpretability and generality?

Her argument was: the more specialized a circuit is, the easier it may be to understand. She showed a highly stereotyped male courtship circuit in the Drosophila melanogaster - a dedicated sensory-to-motor pathway linked to species-specific pheromones and stereotyped mating behaviors. A relatively narrow-purpose circuit. And therefore, relatively interpretable. This contrasted nicely with more general-purpose circuits like the mushroom body, where coding becomes distributed, sparse, and much harder to map onto specific functions. One striking point about mushroom body connectivity: unlike most of the fly brain, some of its connectivity patterns are effectively random at the individual neuron level. This sparked a deeper question: do we understand some circuits simply because evolution made them specialized, or because we have chosen the “easy” circuits to study first?

The day ended with Barbara Webb and one of the most elegant examples of insect cognition: the honey bee waggle dance. The dance itself is remarkable - a symbolic communication system encoding direction and distance to food, using angle relative to gravity and the sun. But the real question was: how do the follower bees decode it? In complete darkness.

The answer seems to involve direct antennal contact, mechanical signals, and internal reference frames for gravity and orientation. And underneath it all lies a familiar computational problem: vector coding. How do you represent direction, distance, and movement in neural circuits?

Barbara linked this to the central complex, a compact circuit used across insects for path integration and vector computation. What is striking is how conserved this solution appears to be. The same basic vector computations seem to reappear across species and behaviors - navigation, migration, and foraging. A reminder that some neural computations may be far more universal than the behaviors they support.

Understanding brains may depend as much on choosing the right level of description as on the data itself.