ON INTELLIGENCE

Hawkins, the PalmPilot’s inventor, is keen to build truly intelligent machines based on his ideas of how the brain really works.

The brain is no computer, this guru of handheld devices makes clear. The AI folks have got it wrong, even with their neural networks and feedback devices. The brain is not a super-fast PC computing all the moves so as to beat you at chess. It’s slow, but oh-so-clever. It can fill in the quote when you supply it with a few words, read your crummy handwriting, hum the tune given a couple of notes. All this is possible, Hawkins says, because of the hierarchical structure of the neocortex covering the brain. The cortex is composed of six layers of cells that connect up and down in columns, sideways to cells in other columns, and also connect to other parts of the brain. Incoming signals, say from the eye, send a constantly changing barrage of signals to cells in the visual cortex that, through a succession of relays up and down, get transformed into invariant patterns (a face, for example). Memory, in Hawkins’s theory, is a neocortical function based on extracting invariant features of spatial (or temporal) sequences of patterns and employing a process of “autoassociation” in which pattern one invokes an associated pattern two, etc. Hawkins provides many a homely example to comfort the reader traversing the neuroanatomical details, culminating in what he calls the memory-prediction concept of intelligence, in which we use memory to make analogies that allow us to anticipate what happens next, even to devise creative moves. The strength here lies in the solid work of neuroscientists under-girding Hawkins’s ideas. Its weakness is his failure to consider the influence of the brain’s emotional/motivational circuitry on learning and memory.

Ever the optimist, Hawkins considers building intelligent machines eminently doable. Given his track record, maybe he’ll succeed. If not, the exercise may provide further insight into how the brain really, really works.