On Intelligence (Paperback)

But isn't artifical intelligence (AI) a good metaphor for human intelligence? No, says Hawkins. In AI a computer is taught to solve problems beloning to a specific domain based on a large set of data and rules. In comparison to human intelligence AI systems are very limited. They are only good for the one thing they were designed for. Teaching an AI based system to perform a task like catching a ball is hard because it would require vast amounts of data and complicated algorithms to capture the complex features of the environment. A human would have little difficulty in solving such everyday problems much easier and quicker.

Ok, but aren't neural networks then a good approximation of human intelligence? Although they are indeed an improvement to AI and have made possible some very practical tools they are still very different to human intelligence. Not only are human brains structurally much more complicated, there are clear functional differences too. For instance, in a neural network information flows only one direction while in the human brain there is a constant flow of information in two directions.

Well, isn't the brain then like a parallel computer in which billions of cells are concurrently computing? Is parallel computing what makes human so fast in solving complex problems like catching a ball? No, says the author. He explains that a human being can perform significant tasks within much less time than a second. Neurons are so slow that in that fraction of a second they can only traverse a chain of 100 neurons long. Computers can do nothing useful in so few steps. How can a human accomplish it?

All right, human intelligence is different from what our computers do. What then is it? I'll try to summarize Hawkin's theory.

The neocortex constantly receives sequences of patterns of information, which it stores by creating so-called invariant representations (memories independent of details). These representations allow you to handle variations in the world automatically. For instance, you can still recognize your friends face although she is wearing a new hairstyle.

All memories are stored in the synaptic connections between neurons. Although there is a vast amount of information stored in the neocortex only a few things are atively remembered at one time. This is so because a system, called `autoassociative memory' takes care that only the particular part of the memory is activated which is relevant to the current situation (the patterns that are currently flowing in the brain). On the basis of these activated memory patterns predictions are made -without us being aware of it- about what will happen next. The incoming patterns are compared to and combined with the patterns provided by memory result in your perception of a situation. So, what you perceive is not only based on what your eyes, ears, etc tell you. In fact, theses senses give you fuzzy and partial information. Only when combined with the activated patterns from your memory, you get a consistent perception.

The hierarchical structure of the neocortex plays an important role in perception and learning. Low regions in the structure of the neocortex make low-level predictions (about concreet information like color, time, tone, etc) about what they expect to encounter next, while higher-level regions make higher-level predictions (about more abstract things. Understanding something means that the neocortex' prediction fits with the new sensory input. Whenever neocortex patterns and sensory patterns conflict, there is confusion and your attention is drawn to this error. The error is then sent up to higher neocortex regions to check if the situation can be understood on a higher level. In other words: are there patterns to be found somewhere else in the neocortex, which do fit to the current sensory input?

Learning roughly takes place as follows. During repetitive learning memories of the world first form in higher regions of the cortex but as your learn they are reformed in lower parts of the cortical hierarchy. So, well-learned patterns are represented low in the cortex while new information is sent to higher parts. Slowly but surely the neocortex builds in itself a representation of the world it encounters. Hawkins: "The real world's nested structure is mirrored by the nested structure of your cortex."

This model explains well the efficiency and great speed of the human brain while dealing with complex tasks of a familiar kind. The downside is that we are not seeing and hearing precisely what is happening. When someone is talking we by definition don't fully listen to what he says. Instead, we constantly predict what he will say next and as long as there seems to be a fit between prediction and incoming sensory information our attention remains rather low. Only when he will say something, which is actively conflicting with our prediction, we will pay attention.

The author takes his model one step further by saying that even the motor system is prediction driven. In other words, the human neocortex directs behavior to satisfy its predictions. Hawkins says that doing something is literally the start of how we do it. Remembering, predicting, perceiving and doing are all very intertwined.

I think this is a fascinating and stimulating book. Many questions about intelligence may remain unanswered but I believe this book to be a step forward in our quest to understand intelligence. The author predicts we can soon build intelligence in computersystems by using the principles of the neocortex. He is optimistic about what will happen once we succeed in this. He (reasonably convincing) argues these systems will be useful for humanity and not a threat.

Coert Visser, www.m-cc.nl