杏吧原创

Other animals are helping us understand how the human brain鈥檚 almost entirely flexible structure sets our intelligence apart, says John Duncan

THE human mind is among the most powerful forces on earth: the intelligence emerging from it allows us to cultivate vast cornfields or rice paddies and build sprawling cities; to launch spacecraft, paint pictures, compose music 鈥� or use reason to write this article.

But the full account of how human thought emerges from a biological brain, a network of billions of neurons communicating via tiny electrical impulses, still ranks among the great scientific mysteries. In 1951, Karl Lashley, one of the fathers of neuroscience, looked forward to a possible 鈥減hysiology of logic鈥�. In my new book, How Intelligence Happens, I tell the story of how this dream is approaching reality, thanks to the modern integration of research in psychology, artificial intelligence, brain imaging and neurophysiology.

To a large degree, a human brain resembles the brains of other vertebrates. The behaviour of many animals is founded on a basic building block called the 鈥渋nnate releasing mechanism鈥� 鈥� a fixed pattern of action prompted by a triggering sensory event.

Perhaps the best-known example of an IRM comes from the one of the founders of ethology, Niko Tinbergen, and his analysis of courting stickleback couples. At the mouth of his nest, the male sees a female enter his territory, her belly swollen with eggs and her posturing indicating she is sexually receptive. The sight releases the male鈥檚 response. He approaches and begins a characteristic zigzag dance. The dance induces the female to approach and on it goes. This pattern is repeated in many animals, with complex, apparently goal-directed behaviour built from the combined fragments, or subprograms, which are individual IRMs.

Human behaviour is also made up of complex programs. For much of our lives, the fragments of these programs are not IRMs: instead, I call them cognitive enclosures, an enormous number of focused, cognitive steps leading to the achievement of a goal. In any mental program, distinct cognitive enclosures follow one another in rapid succession. For example, in one second we search in our pocket for the car keys as we leave for work, in another we move into the car seat, and in a few more seconds, we check for traffic as we pull out into the road.

Breaking tasks into useful cognitive enclosures is as essential for this everyday sequence as it is for the more cerebral activities we consider 鈥渋ntelligent鈥�, such as constructing a proof to a mathematical problem or understanding the rules of a maze.

Ironically, from the first unwieldy tin boxes to the superfast laptops of 2010, computers fail abysmally at tasks, or subprograms, that humans find trivial. Computers are hopeless at distinguishing road signs, pedestrians or cars at a busy intersection, or at reading a paragraph of text and building an image of the world it describes. Where computers excel is in decomposing complex problems, dividing them up into separate, independently solvable subproblems 鈥� and then reassembling them into an organised, goal-directed structure. In humans, we can show some parts of this decomposition and assembly experimentally. But we are only beginning to grasp other parts of the puzzle, such as the rapid transition between cognitive enclosures.

The clues come from many places. For more than 100 years, we have known that damage to one part of the cerebral cortex 鈥� the frontal lobes, immediately behind the forehead 鈥� 鈥渓oosens鈥� the structure of complex thought and activity. Person A planes a plank of wood but 鈥渇orgets鈥� to stop and continues planing into the bench; person B writes an incoherent sequence of calculations which fail to solve a mathematical problem he has been given. Often such people do very well on simple cognitive tests but can no longer hold jobs or navigate family life. Individual fragments of thought and action have been preserved, but the coherence of the whole is lost.

With modern MRI or functional MRI techniques we can peer into the frontal lobe to learn a lot more about how this biological programming works. Some of the simple tasks used in intelligence tests (such as completing a series of letters, or choosing the odd-one-out from a group of images) are chosen because they predict success in many kinds of activity, from lab-based measurements of language or memory to lifetime achievement in education or at work. Looking at people taking these tests during an fMRI scan, we see activity in a specific brain network, with components in several separate regions of the frontal lobe. Since activity in this network is also seen as part of the brain鈥檚 response to any cognitive challenge, this suggests there is a core brain system which is key in building cognitive programs.

MRI images also tell us a lot about the coding in these regions. As we should expect of a system that can be programmed, the network responds to each new cognitive challenge by focusing on specific information needed to control current behaviour 鈥� that is, the specific content of the current cognitive fragment, or enclosure.

A neural network is composed of hundreds of millions of separate neurons, each incredibly small, yet often coding highly specific information in the electrical impulses it fires. Take the example of a toad noticing a worm. The toad responds with characteristic movements: it turns towards the worm, approaches, fixes its head in position and snaps its jaws.

Describing in detail how neurons make this happen, from the firing of a single neuron indicating the presence of the long, thin, wormlike object in one part of the visual field, right through to the integration of different kinds of information that will allow the toad to eat the worm, took three pages in my book 鈥� and that was an account for non-specialists.

At the same time, this also reveals how much is still missing, even in our explanation of this relatively simple case. We could not even begin to build an artificial neural network capable of mimicking the complexities of the toad鈥檚 behaviour.

When it comes to the massively more complex mental ability of humans, we must also flip between what we are pretty sure about and what is eluding us. On the plus side, we are confident that the IRM of the stickleback or the toad has been replaced in humans by an almost entirely flexible structure that can focus on virtually any kind of problem 鈥� from daily functioning to abstract reasoning.

We can begin to see how these enclosures are assembled in the brain, providing the essentially human element of our intelligence. It is a partial picture, like an early map of the world: some countries are clearly drawn, others no more than a sketch, and still others simply labelled 鈥渦nknown鈥�. But it is a picture that at last begins to realise Lashley鈥檚 dream.

There are also, I suspect, countries we don鈥檛 even know are missing. Our human brain has allowed us to understand the atom and probe the boundaries of the universe, but is it really configured for unlimited conceptualisation or understanding? Is just one species so very different from all the others? Or is it rather that, like all the others, we can imagine only so far as our own nervous system allows us?

鈥淎re our brains really configured for unlimited understanding?鈥�