How Brains Think: Evolving Intelligence, Then and Now [Hardcover]

As we plod along towards Alan Turing's dream of constructing intelligent machines, there are a few road-blocks we need to get around. Calvin mentions that any explanation of biological intelligence ought to have implications for artificial intelligence. He admits that, "the ad-hoc schemes of AI might also produce intelligent robots", but he clearly likes the idea that the most efficient path to intelligent man-made devices that can duplicate human mental abilities (what Calvin quaintly calls a "workalike") is to learn the essentials of how biological brains work and then apply those principles to the problem of making a workalike. One road-block is the fact that so many AI researchers ignore the task of reverse engineering the human brain or, at best, they assume that what was known about brains in the 1940's is enough. Unfortunately, I doubt that Calvin's hop-skip-and-jump over this issue will move any AI researchers away from their "ad-hoc schemes". Even AI researchers who like the idea of evolutionary processes pay little attention to the idea of adapting the physiological mechanisms of biological brains to evolutionary computing.

A second road-block is the distinction that is usually made between hardware and software. Turing was among the first to recognize how to use electronic devices to implement the power and beauty of this distinction, and most AI researchers remain devoted to hardware-software duality. Unfortunately, biological brains were not designed by an electrical engineer. It thus becomes a danger that biologists will mistakenly attempt to make sense of biological brains by looking at brain processes through the distorting lenses of hardware-software duality. I think that Calvin gets caught in this trap of dualism and it deflects him from paying close enough attention to the details of how biological brains really work.

Calvin's dualistic thinking starts with the harmless division of brain processes into two types, those that depend on "cerebral ruts" (hardware) and those that dance more freely through the brain and so are able to function like "software".....Calvin usually calls these "firing patterns". The dangerous step comes when Calvin suggests that the pattern of action potentials in any particular neocortical minicolumn can be replicated and spread through the cortex like a piece of software code and be "played" on the millions of other minicolumns in the same way you can play a million copies of a CD on a million CD players......the key difference being that while all CD players are designed to do basically the same task, the various cortical minicolumns can all have their own unique "ruts" and the copies of the firing patterns are not exact duplicates. This allows for a "cerebral symphony" rather than just a million-fold amplification of the same tune and a "survival of the fittest" process whereby those firing patterns that resonate best with the existing pool of "ruts" will dominate our consciousness and generate intelligent behavior.

Allegorically, this is an appealing model of cortical function, and the sort of evolutionary mechanism that AI researchers like to build into computers. Unfortunately, biological brains were not made by engineers. Where does Calvin go wrong?

For Calvin, "copying" is the essence of darwinism, but people such as Stuart Kauffman (see his, "The Origins of Order") and Freeman Dyson (see his, "Origin of Life") have long ago made the point that evolution does not REQUIRE copying. Life as it now exists involves molecular replication, but that is just a special trick that proved useful for living organisms. Life probably had an initial "metabolic" period without molecular replication. George Dyson (in his book, "Darwin Among the Machines") has pointed out that man-made machines (such as industrial robots) can evolve without replication.

Is there another way that evolutionary processes might exist in biological brains without the emphasis that Calvin places on copying? Gerald Edelman's theory of "Neural Darwinism" (see his book by that name) provides an alternative to Calvin's emphasis on copying. Anyone who reads Calvin's theory (chapter 7) should compare it to Edelman's theory. My guess is that Edelman provides a better framework for thinking about evolutionary processes in biological brains, but Calvin's theory is more accessible and "intuitive", so it may be better as an introduction to the importance of evolutionary processes in the brain.

Calvin does not dogmatically push the "darwin machine" mechanism described in chapter 7, and, in fact, seems to invite people to simply skip that chapter. He devotes considerable space (particularly chapter 3) to discussion of trying to find the level of detail required to find the essence of how biological brains produce consciousness and intelligent behavior. I suspect that Crick (in his "The Astonishing Hypothesis") and Edelman are closer to identifying this critical level, with Calvin just a bit too reluctant to delve into the details of how synapses work. Although Calvin does touch on the function of synapses briefly in chapter 7, he spends considerable space clumping the study of synaptic learning mechanisms in with quantum consciousness theories as examples of inappropriate attempts at reductionism. But even if Calvin has slightly missed the mark, he provides an accessible account of why it makes sense to continue trying to identify and elucidate evolutionary processes in the brain.

My biggest problem with the book was Calvin's idiosyncratic choice of terms. He seems to demand some potion of free will in the neuron's selection of input signals; he sees no value in random selection nor mutation. Intelligence he sees as "good guessing." Cerebral codes he sees as the winners of the intense competition over what will be copied in short term memory. He thinks Penrose's quantum field or"microtubles of neuron's cytoskeleton" is just another word for spirit -- the ghost in the machine, but his own stealth candidate is "dynamic patchwork quilt" of patterns. I enjoyed his metaphors but they need not conflict with Penrose's. What he has done with his cerebral codes is encrypt his own common reductionism of portraying man's mind as just a bag of neurons -- like his buddy Francis Crick (THE ASTONISHING HYPOTHESIS). There is nothing new in this book except new terminology of which we are already stuffed.