He elucidates the immune system’s scheme of producing a combinatorially large number of genetically distinct B lymphocytes cells with a corresponding distinct antibodies each waiting for an antigen to make its day. After one of these lymphocytes has found its nemesis it reproduces rapidly and also rapidly evolves to fine-tune its ability to zap the antigen. He counts this as selectionist evolution.

Now he goes on to the brain which is where I thought he was going when I bought the book.

By page 64 he has presented his argument that wiring the brain is accomplished by initial over-wiring followed by removal of about half of the wires.
Well if there are 10^{11} neurons and you eliminate half of them, there are about 2^{1011} ways to choose which ones to eliminate.
Now that is a big number but actually nearly not big enough to wire the brain.
I am sure that it serves some purposes but it cannot for each of the 10^{11} neurons, select which 10000 neurons its axon should contact.
It can certainly solve the left-eye - right-eye problem.
Neurons that carry visual information from the retina to the visual cortex form a map and this map must be continuous in some sort of topological sense.

I am guessing that there is traditional genetic morphological information that biases random connections.

I am enjoying Cziko’s historical perspective on ideas. He tells of ideas of Skinner, Piaget and Chomsky. He dismisses Chomsky’s innate language on page 133 for reasons that I don’t understand. I think that it is hard to separate language mechanisms from thought mechanisms. They may be isomorphic in some sense. I think that logic is based on patterns and that language may come along for the ride.

I am halfway thru the book and I have a quibble. Cziko speaks of random thoughts or ideas which are selected for or against. Any source of random things comes from some probability distribution. Even more fundamentally the things come in some format. If thoughts pop-forth as sentences in a native language, then the thoughts conform to some degree to that language. Chomsky generative grammars plus a few numbers, are a concrete generator of random utterances.

Mathematicians have thoughts in unnatural languages, again constrained by syntax rules. Good mathematicians have different rules, or patterns, that occur to them. These patterns evolve in an individual’s lifetime and are frequently neither expressible nor conscious. I think that selection of thoughts is a distinct from selection of patterns. The former are falsified by logic and variance with beliefs. Patterns are selected against by poor yield.

Finally at the end of page 289 Cziko gets to a point that has been bothering me regarding “random ideas”. There is no notion of random in spaces such as the “space of ideas”. For clearly specified finite sets there is a notion of equally likely. Even for the real line the notion of uniform probability (distribution) it is possible only for finite intervals. Variation is from some space of possibilities and that space itself is presumably a product of selection. This process comes in many stages and sometimes progress is to abandon a pattern. Einstein abandoned the notion of space given to us by successively more detailed patterns starting thru Euclid, Galileo, and Newton. Physical theories had constrained themselves to these before Einstein. Genetic programming does not generate programs by storing random numbers into memory and then branching to them—not even the initial programs. GP usually operates on floating point numbers with floating point instructions. Cziko finally made these points on page 290.