What if the brain is just a tangled, specialized mess?
 The brain contains around \(10^{11}\) neurons and \(10^{14}\) synapses.^{1}
 If the exact connectivity of these elements is critical to the brain’s function, then there may not be underlying principles
 We’d be totally unable to identify units of functionality, or create effective abstractions
 “Hardcoding” is a programming term to describe handling cases individually, without abstraction or generality
Fortunately, this cannot be true.
 The human genome only contains 3.2 billion base pairs.^{2}
 Each pair is 2 bits, so the genome is around 6.4 gigabits (0.8 gigabytes).
 Some Fermi estimation (in base 2):
 Suppose no more than \( \frac{1}{4} \) of the genome talks about the brain. It’s probably much less.
 So we have \( 1.6 \cdot 10^9 \approx 2^{30} \) bits to work with.
 There are \(86 \cdot 10^9 \approx 2^{36}\) neurons, so specifying a connection between two neurons requires 72 bits.
 Of course, most pairs of neurons are too distant to connect to each other. Let’s say that some sort of neighborhoodencoding scheme saves half of the data requirement (i.e. 36 bits per connection, where each neuron can choose from \(2^{18}\) neighbors). This seems like a good ballpark since neurons tend to have something like \( 2^{15} \) synapses.^{1}
 Then the braincoding portion of the genome can only specify \( \frac{2^{30}}{36} \approx 2^{25} \) synapses.
 This is only enough for 1024 neurons at \(2^{15}\) synapses each.
 So the synapsegeneration algorithm must be profoundly general and the brain is not hardcoded in any significant way.*Be skeptical!
 This fits our experience: babies have to learn to walk, even though walking is a prime example of something that would make sense to hardcode.

Chudler, E. Brain Facts and Figures. ↩︎

Chial, H. (2008) DNA sequencing technologies key to the Human Genome Project. Nature Education 1(1):219 ↩︎