How Brains are Like Computers

I propose to collect here random notes on similarities between the brain and the computer.

We learn from metaphors. We understand by metaphors. Things are alike in unexpected ways. Computers arose by trying to automate certain activities in the brain that our consciousness had fairly direct access to. This is in the realm of puzzle solving and more generally the parts of mathematics that have helped us understand our world. Consciousness lacks access to large parts of the brain, mostly older emotional parts that are absolutely vital to even mathematical work. We find the new patterns which are the sole subject of mathematics thru brain mechanisms which are inaccessible to us. It strikes us out of the blue that some structure that we are considering is a vector space. These revelations do not yet occur to our automated theorem proving programs. I am of the firm opinion that conventional computers programed in conventional languages with yet uninvented algorithms can produce such insights. I am not convinced that such programs will be practical. I do suspect that special digital clocked hardware may provide such faculties. In short I expect both brain science and computer science to improve by cross fertilization.

Edelman & Tononi’s book scarcely lets a page pass without a disparaging comment on comparing a brain and a computer. The only benefit of such exclamations that I can see is to dissuade some class of readers the brain is a mere computer, but I have never met such a person.

New book that I have not read: “Probably Approximately Correct” by Leslie Valiant

The digital engineer is confronted with the question of what to put in hardware and what to leave up to ‘software’. The short answer is that if you do a lot of something you need hardware to do that thing. Otherwise the general purpose CPU can be applied to that thing by writing a program. Evolution had the same problem. It is a tradeoff! Curiously the von Neumann architecture is now the only successful general CPU plan.

The digital computer was invented in one fell swoop. Those who developed the first general computers did so because it was possible. They got funding to actually build because such computers had promise of early utility. Thinking is hard and there was this illusion that they could build something to do all sorts of thinking just as we did. In a sense computers are designed by humans to do a class of tasks that humans do much more slowly and this was possible because humans have as subject, partial access to their own thought process. The human brain evolved in 700 million years. This difference is vital.

A computer engineer (even pipe-organ engineers) knows several properties that a memory technology must have. There must be:

means of delivery of information to the memory,
means of retrieving a memory,
means of designating which memory to retrieve,
recognition of when it is important to capture a memory,
selection mechanisms to make physical m modifications to just the right parts of physical memory.

Each of these are substantial engineering tasks. These are not entirely orthogonal but no discussion of memory within the brain that I have seen comes close to covering all of these points. There does not seem to be recognition that these mechanisms must be explicated. There are plausible explanations for each of these within Kanerva’s schemes except perhaps for knowing when to capture the memory.

Decaying Bloom filters
See the JTAG connection.

Older note