Evolution develops forms gradually over time, in a step-by-step fashion (Figure 6.9), so brains had to evolve by a long series of modifications pruned by natural selection. This descent with modification is known to produce modular hierarchies (Gao, 2025), as regions of the brain fine-tune to a function by increasing their internal connections, and older structures aren’t discarded but become the foundation for later developments.
Motherboards are first designed then built, but nature had to in effect design as it built, so it produced one module, tested it for a few million years, then added another on top of it. The evolutionary advantage was less risk of functional failure, as adapted modules could operate while others evolved. Module hierarchies let evolution make brain design decisions one at a time, giving common ancestors that led to the next step.
A transistor can’t grow into a motherboard, but a cell can grow into a brain because evolution found the way, and embryos follow that path as they grow. In growth and evolution, the brain hierarchy started with one cell, the neuron, so we begin there.
A nerve is a cell whose body receives electrical input from dendrites and projects electric pulses down an axon to other nerves (Figure 6.10). When the dendrite strength passes an input threshold, the nerve fires, to trigger other nerves. For example in Figure 6.11, inputs B and D are strong enough to fire neuron A, but B and C aren’t, as they don’t reach its threshold of four. Nerves are then devices that selectively pass on electrical impulses.
Embryonic nerves grow from the brain to form the retina, so light entering the eye touches the brain directly. It is estimated that each eye inputs about 8.75 Megabits a second, and the brain total is over 20 Mbps, so a baby’s first impression of the world is probably information overload:
““The baby, assailed by eyes, ears, nose, skin, and entrails at once, feels it all as one great blooming, buzzing confusion” (James, 1890).
The baby’s brain has to transform the data from millions of optic nerves into objects, like a cup, to interact with the world, but how? Computers do it by identifying features like shape, so videos can be compressed by keeping key features and discarding the rest, and the brain the does the same. Essentially, it handles information overload by reducing the incoming sense data to key features, like borders and shapes, that let us recognize objects.
.
The first step of vision then is to differentiate black from white. Photo-electric cells do this by firing when there is light, so in information terms, black is 0 and white is 1, as set by its designer. But brains had no designer so evolution, as it always does, took both options, as one type of retinal cell responds to light and another to light (Figure 6.12). Instead of defining data absolutely, the retina responds to both light and dark, which lets it identify the next key feature, of borders.
Retinal cells then interact, to excite or inhibit each other, to enhance the borders that define object shapes. The next step in object recognition is to make one side of the border the object, and the other its background, but which? The visual data is ambiguous, so vision assumes as best it can which side is figure and which is ground. For example, in Figure 6.13, making white the background just gives black blobs, but making black the background lets you read MAIL BOX.
Vision then isn’t a simple input-output process because it must make assumptions that could be wrong. Visual processing begins when the eye detects light, and this data is then subject to layer upon layer of processing modules that detect relevant key features. For example, some visual cortex nerves fire for different line angles (Figure 6.14) and others detect other features.
Computer processing tends to be a simple one-way flow, but brain nerves have bottom-up, lateral, and top-down links, so processing can be top-down and sideways as well as bottom-up. Bottom-up paths analyze data as computers do, but lateral paths enhance results, and top-down paths prime, interrogate, and check the result for consistency or errors (Dehaene, 2014) p139. Higher modules of the hierarchy can then re-run the process with different assumptions to get an entirely different result.
For example, is Figure 6.15 a young lady or an old one? If you see a young lady, can you see an old one? To do this you must rerun your visual processing, to make the ear of the young lady the eye of an old one. The visual system makes a best guess, but you can ask for a redo because nerves in the visual hierarchy project down as well as up. Note that the change in what you see is sudden, and you can see one or the other but not both at once. Lower processing may be out of sight and out of mind, but it can still be redone by top-down control. This is necessary because all perception is just a hypothesis of an ambiguous world.
Subconscious processing is often assumed to be primitive but the spinning ballerina illusion (Figure 6.16) suggests otherwise. Click on the link to see a ballerina spinning but the rotation is ambiguous, so you might see her spin clockwise or anti-clockwise. Try to see her spin the other way. If you can’t, pause the video and if you see an extended leg at the front, imagine it at the back, or vice-versa. Restart the video and if she spins the other way, you just reprogrammed some very complex unconscious visual processing. This ability lets higher conscious modules reconfigure lower ones to see the world in a new way.
The cortex is then a nested hierarchy of nerve modules that evolved to process sense data in various ways. But while the optic nerve has about a million neurons, the auditory nerve only has about 50,000, so the visual hierarchy base is much broader. As shown in Figure 6.19, for the same processing capacity, a broad input base has less processing depth, while a narrow base has more. The narrow base of sound allows deeper processing than vision, so language began with speech because sound allows the deep analysis that language needs. Given the trade-off between processing breadth and depth, the hemisphere better adapted to deep processing specialized in language, and the other specialized in the broad processing of spatial analysis.
Sensory hierarchies let lower modules process their data to identify key features that pass on to higher ones, but how the results combine into the unified experience we call consciousness is unclear. What is clear is that this unified view can be checked and rerun, as illusions show, so consciousness lets us revise our reality assumptions to see it in a new way. For example, this book challenges the assumption that objects exist by pointing out the factual inconsistencies that follow, so we see quantum-generated events not self-existing particles. But if nested module hierarchies run the brain, how does it maintain control?
