The Prehistory of the Mind

At the end of chapter 3 Mithen has set up the mystery of how we come to have what seems to be general intelligence. He has invoked successively the ideas of

Jerry Fodor’s two tier architecture
- specialized input modules (the senses)
- generalized intelligence
Howard Gardner’s innate multiple intelligences
Cosmides & Tooby’s explicitly evolved “Swiss army knife” of faculties, each to cope with particular practical problems of the last few hundred thousand years
Domains of intuitive knowledge observed in children by developmental psychologists
Patricia Greenfield’s suggestion that infants seem to have general intelligence before they develop the specialized intelligence at the age of about 3
Annette Karmilof-Smith’s proposal that these domains of specialized intelligence are somewhat cultural in nature

To this point little light is shed on our ability to do mathematics and philosophy. Then come the ideas of Margaret Boden and Arthur Koestler about cross-overs between domains—meta concepts that span faculties. This is yet mysterious.

My Take

Before proceeding with Chapter 4 I want to say what I think is going on. First I note that Mithen speaks as if computers can only do general purpose things. He distinguishes little between the computer and the program. This bias comes perhaps from reading about 20th century computer AI efforts. I can’t see that this bias damages his arguments however—so long as he does not draw conclusions about what computers cannot do.

I think that the various special faculties share a great deal of mechanism. Just as evolution freely reuses proteins so may it reuse modules. Even if two faculties are physically disjoint in the brain they are likely to share the design of associative mechanisms perhaps as proposed by Kanerva. There are adaptive advantages to keeping the information stored in these modules separate. Each horse may innately know that what it eats can cause sickness, while still being able to learn which particular foods are bad.

It seems plausible, perhaps even likely, that shared design of associative circuitry is liable to leak information across faculties and that the advantages of isolation actually require extra investment. The cross-overs may be merely nature’s management of a tradeoff between the advantages of isolation and integration.