What Makes You Clever

The following sections are included:

• About the Author
• Dedication
• Preface
• Acknowledgements
• Contents

I think, and you think. But how we manage to think is almost a total mystery to science. In sharp contrast to almost every other topic that scientists have tackled, your cleverness remains an enigma despite decades of scientific enquiry. From this book you will find out why your easy skills — such as chatting — have proved so difficult for science to understand. Your cleverness is not easily explained, but by following the scientific probing described in the following chapters you will learn a lot about this accomplishment that makes you unique in the vast tree of life. You will also learn about the nature of scientific enquiry itself…

The main goals of this chapter are two-fold: to explore the problems that arise when the scientist attempts to extract the principles underlying intelligence by studying a working brain; and, at the same time, to illustrate the fundamentals of scientific enquiry. The problems that arise point to the importance of feature salience. They also lead to a better understanding of the gap between structure and meaning (in effect, Maxim 4). In particular, we will see the necessity to introduce levels of understanding…

The primary (but by no means the only²) sources of evidence for understanding the working brain are the various brain-scanning technologies. So far, we've simply assumed that detailed evidence — primarily the timings and strengths of neuron-network signals — is available and can be taken at face value…

“The condition for successful science… is that nature should have joints to carve it at: relatively simple subsystems which can be artificially isolated and which behave, in isolation, in something like the way that they behave in situ.” So said philosopher Jerry Fodor in The Modularity of Mind (p. 128)³…

Our earlier exercise in reverse engineering on the small brain module was perhaps over-optimistic. That was why we ended up perched on the edge of a brain-structured model, staring helplessly into empty space for a route to an unknown cognitive-level model. Wondering where to go as well as how to get there is not a promising basis for progress…

We've seen how our toaster models variously interrelate via the salient features of toasting, and that through such explicit interrelationships a more complete understanding of toasting develops. We've also seen that attempts to similarly interrelate “brain” models, and so understand a component of intelligence, have yielded no success. This is because we don't know what the salient features of intelligence are, or if we do, then we have no idea how these features, such as memory, are realised in terms of brain structures and behaviours…

AI is not dead, definitely not. Over the last half a century or more, the field has moved along in fits and starts — one might say it has moved along steadily, although others might prefer “sluggishly” as more apt. No one without a vested interest, or devoid of technical competence, is likely to be a great deal more positive…

In the previous chapter I attempted to cut across all the bickering about how well or how badly the AI project — the quest to simulate intelligence on a computer — has done by offering a capability to learn as the acid test of an AI system. I also noted the obvious problems with this idea. Nevertheless, there is a good case for the centrality, even definitional status, of this aspect of intelligence — non-adaptive intelligence is an oxymoron…

Knowledge in the form of facts can easily be stored in computers, and facts also offer the possibility of being mastered, but these are nothing like the same thing. Storage is easy; it is however mastery that is a likely key to intelligence. If we replace “master” with “store” in the second opening quotation, I'd happily go along with it. But as it stands, it's just plain wrong (given any vaguely reasonable meaning for “master”). So what are these facts, and what sort of mastering is needed within an intelligent system…

As far as the enquiring mind is concerned the mystery and wonder conjured up by the revelation that we now have machines that learn is nearly always enough to put paid to scepticism about intelligent machines. It's an eye-opener that clinches belief in AI as a current reality…

With the highs and lows of mountaineering machines firmly in mind, it is time to see exactly how the “new”, biologically-inspired learning technologies have been configured to deal with the problems, such as sub-optimal peaks. More generally, what can they contribute to our understanding of this core component of intelligence? What are these hot prospects for opening up new and more productive possibilities for AI…

There is a consensus among cognitive scientists that the human species jumped ahead of all the others over a period of time just a few tens of thousands of years ago — i.e., just “yesterday” in evolutionary timescales. This great leap forward was not in terms of our physical capabilities, but in terms of our mental abilities, our cognitive skills. The human species became intelligent. All other terrestrial life forms were left behind in the mire of a tooth-and-claw struggle for existence mitigated by some weak reflections of human cognitive skills…

Once science has a firm grasp of the mechanisms and principles underlying intelligence, improvements on the slow and error-prone human version will be forthcoming. Artificial Intelligence technology will soon leave our feeble ditherings behind. Science will create ultra-AI or super-AI. Will it? In answering this question, issues at the heart of what it might mean to understand intelligence are exposed…

Mirages are real. I've seen them clearly in the Sahara desert. At least, they are as real as other perceptions when viewed from a distance, but as you get closer they dissolve away. At the time, we were searching for the next oil drum that marked our route across the desert. It was then disconcerting to see an array of oil drums on the horizon, but with binoculars we soon sorted out the visual fictions from the single reality — and we had the heading we needed. If only we could find the “binoculars” to similarly strip out the true semantics from the plethora of possibilities that shimmer and dance around the periphery of our cognitive models, or worse, fail to put in an appearance altogether…

Recall the iCub robot from Chapter 1. It looks like a human child and even has some facial expressions. It speaks like a child, and it can do things that a human child can do. It can learn to recognise objects, and it can understand what it is told to do with these objects. In particular, we are shown that it can reach out and touch the objects it has just “learned”. Other versions, we are told, learn to crawl, to shoot arrows at a target, and so on…

Now is the time for truth, and reconciliation of the various points argued. A close look at Watson — the latest AI success — illustrates the differences between what has been achieved and what has been presented. This examination can be further used to give substance to many of the more nebulous claims about the diversionary pseudo-science engendered by AI projects in comparison to the real science that such projects embody…

The following sections are included:

• Glossary of Maxims
• Index