Please note: This was written many years ago, after reading the article Subcognition and the Limits of the Turing Test by Robert M. French. AI is not my field of research, so take this as a bit of speculation, nothing more. I am keeping the article here because of the number of links that are pointing to it, and for historical reasons (this was the first thing I ever published on the Internet, back in 1996 or so). 1. Introduction
Although there is no clear definition of AI (or even of intelligence), it can be described as the attempt to build machines that think and act like humans, that are able to learn and to use their knowledge to solve problems on their own.
A 'by-product' of the intensive studies of the human brain by AI researchers is a far better understanding of how it works. The human brain consists of 10 to 100 billion neurons, each of which is connected to between 10 and 10,000 others through synapses. The single brain cell is comparatively slow (compared to a microprocessor) and has a very simple function: calculating the weighted sum of its inputs and issuing an output if that sum exceeds a certain value. Through its highly parallel way of operation, however, the human brain achieves a performance that has not been reached by computers yet; and while we might see single computers with performance comparable to the human brain by 2020, it will take a lot more understanding of how the brain works to even just copy its function in a rudimentary way.
The Turing Test is currently the only accepted way to test a machine for exhibiting intelligent behavior. Such a thinking machine has yet to be built.
2. Why the current approach of AI is wrong
These following thoughts do not deal with technical problems of AI, nor am I going to try to prove that humans are the only intelligent species (exactly the opposite, see section 5). What I want to show is that the whole idea of AI needs to be changed in order to lead to more than just partial results.
2.1. Reference Points
Today's AI concentrates entirely on the brain. If you look at the human body, however, it is not clear where to draw the line between which parts of the nervous system belong to the brain, and which don't. A number of functions are performed by the spinal cord, for example, like withdrawing the hand quickly when touching something hot. It can be vitally important that this action is taken as fast as possible, in order to limit the damage. The only way of doing so is through reflexes, without the intervention of the brain. This is not an example of intelligent decisions outside the brain, but it provides an entry point to the following.
Whenever you talk to somebody, you use a huge amount of assumptions about the background of your counterpart. You usually start with assuming that other person is almost identical to yourself, and by small misunderstandings and questions of him you correct that picture you have. When you know somebody, you don't have a list of all his features in your mind, but you know the differences between you and that person (or the difference between him and a third party, be it a single person or a group). This explains why it often is so difficult to describe somebody to another person, since that other person's reference point is different from your own.
We build relations between all the things we know, and we build classes of things by putting objects with certain similar features into one such class. But the most basic difference we see is the difference between ourselves and the objects, the not-ourselves. Man's first reference point is himself, which is obvious when looking around oneself: Isn't that cat looking very nosy at you? Doesn't that monitor's face look at you? Don't that car's headlights look into a certain direction?
People's categories are based on people, first and foremost. This is not an intellectual decision, but a natural necessity. How could you ever find out anything without a first reference point that you could relate it to? This also is the reason why in children's minds everything 'lives': They live themselves, so why shouldn't other things? Why shouldn't that teddybear feel hunger, exactly like I do?
The point is clear: No knowledge can be accumulated without a reference point. AI doesn't obey this. Most 'intelligent' programs are equipped with knowledge, but none has ever had a clear picture of itself that it could relate everything else to. This is the first deficiency.
2.2. The Role of the Body
The most obvious difference between man and other animals is his mind, his ability to accumulate knowledge and pass it on to his descendants. Yet many of man's highly developed abilities can be completely switched off by the sheer terror of a single aching tooth. This also applies for other strong feelings like hatred, grief or pain in general. They can make people act against their better knowledge and their principles - these being higher developed parts of the mind. This leads to a conclusion that is obvious from looking at the ancestry of man: the vital functions rule over everything else. Man has not been built to wear digital watches (as Douglas Addams states), but he is the winner of a game that is as old as life itself: Evolution. If people were able to simply ignore hunger, they would starve to death; if they would have to control their lungs consciously, they would sooner or later suffocate. Vital functions must have priority over everything else. Considering this, it is not surprising that many of our expressions involve basic needs, like 'being hungry for love', 'being tired of something', 'being fed up', 'having a bleeding heart', 'saving someone's skin'. This is also an example for what was said in the last paragraph: The main reference point is oneself, this is of course also true for strong emotions that are on a less 'basic' level.
Additionally, most (if not all) emotions are accompanied by physical symptoms, such as the production of hormones, shivering, gnashing of teeth or goose-pimples.
The point I want to make here is that the human body plays an important role in all intellectual processes, since they are mere subordinates of its needs in order to stay alive. It is therefore short-sighted (another example!) to try to build artificial minds not only without any body, but also without even the concept of a body. How should an artificial mind ever be able to understand tiredness, excitment, happiness or fear without ever having felt it? And by feeling, I mean the physical symptoms, and the intellectual processes that accompany the fear of injury or death, for example. A body-less mind can never understand that, und thus will never be able to understand humans, let alone act like one. This is the second deficiency.
Two additional points here: AI, being a science, acts very unscientific here. What is being done (although not on purpose) is, that while denying the existence of an immaterial soul that is independent of the body, only the soul is taken into account, but not the effects of the body.
The second point is an observation: The reason why men and women often have such difficulties understanding one another, might be their different reference points, and also their different bodies, that affect the way they think. As shown above, these two points can easily lead to misunderstandings, when one assumes the counterpart to be too similar to oneself. This does not, however, mean that one of the sexes is superior, but that they are different, and that these differences have to be worked out and brought into people's minds.
2.3. Evolution
This little interlude also brought another point up: the desire to replicate. This being the most basic of all basic desires, it cannot be ignored. No experiment in the field of AI has yet specified which gender the mind would be. In a world, where there is no need for a partner to replicate (because there is a simple 'copy'-mechanism, for example), social structures would be completely different from man's. It is even the question, if different genders were necessary at all. Thus, an artificial mind would also need the 'ability' to die, as well as that to mate and replicate, otherwise the resulting being would be beyond any recognition by a human mind.
A mind that cannot die and that doesn't feel the need to replicate in a manner similar to humans, would be very different from man. Such an environment would have to be created artificially, but in a different sense: The conditions would have to be made more difficult than they needed to be, only to force the beings to act human-like. And only an evolutionary process would lead to a mind similar to a human one. This is the third deficiency.
3. The Art of Learning
Learning isn't a 'static' ability, but one which continuously changes: You must learn how to learn, and the way you learn changes. The new-born child can only learn by first-hand experience, and hardly generalize. But the older the child gets, the more he/she can learn without having experienced a corresponding situation. Indeed, most of what you know is what you were taught by others, read in books, etc. This probably is the main advantage of man over all other animals: that we can pass on knowledge, so that the next generation doesn't have to make the same mistakes again (it does anyway, but that's a different problem ...). Your knowledge includes the first-hand experiences of hundreds of thousands of people, whose knowledge and experiences were collected and put into a structured form, in order to make learning these facts easier. People nowadays aren't more intelligent than 1000 years ago, but we have more knowledge, and thus can achieve far higher goals. Like Newton said: "I am a dwarf, but I can see very far for I am standing on the shoulders of giants"
But what effects does this have on AI? The brain changes, and not only does its knowledge change, but also the way it accumulates knowledge and makes use of it. An artificial intelligence must be able to change its own programme.
4. Thinking Outside the Brain
The previous point contains another interesting thought: Whatever you do, whatever you think or say - it hasn't been thought up entirely by you, it always contains parts from other people. This is the key to developing beyond what a single generation can reach (see previous section). But it also leads us to a somewhat discomforting question: How much of that brain is actually mine? What percentage of what I think has been thought by others already? How different am I?
Imagine a group of five people working on the solution of a complicated problem. They're sitting around a table, discussing ideas and problems, taking notes, and thinking about whether or not that last proposal was good. In the end, they will come up with a solution that will be far better than one that any of them had worked out in his/her own! The result will even be better than if each of them was assigned a fifth of the problem, and solved it alone. This team is able to achieve more than five single persons can do independently.
But what is the difference between five isolated persons and a team of five? If the team develops an idea that the single persons don't, which of the members created it? It's not a person that created the idea, but the interaction process, the discussion. An act of thinking has been done by an immaterial process, not a single person.
5. The idea of a General Concept of Intelligence
All the points made above make one thing clear: In order to build an artificial intelligence, it must be built as human-like as possible. Without basic human 'ingredients', the resulting mind might not even be recognized as such. This boils down to the feeling that the goal is to build a mere copy of the human mind.
Why on earth, one might wonder, would anybody want to build a copy of the human mind? Isn't the original working fine? Isn't it superior to everything known? Isn't one's mind the most difficult thing to be examined by itself? What would be the use of such an artificial mind, that would need even more artificial means, only to stay human-like?
The only logical solution to this is to completely separate human from artificial intelligence, in order to build something entirely new.
This naturally leads to the idea of a higher principle of Intelligence, that human intelligence is only one manifestation of (in order to distinguish between the traditional human intelligence and this new idea of a more general concept, I want to spell the latter with a capital I: Intelligence). Another one would be artificial intelligence, and another one still the intelligence developed on a planet many lightyears from here. Again, I remind you of the points just made. Considering these, how should a mind that is the result of evolution on an entirely different planet be similar to ours in any way? There must be similarities, but on a higher level: on the level of Intelligence (note the capital I).
In that hierarchy, AI is on the same level as human intelligence, together with animal intelligence and any other kind of intelligence that one might encounter. The following figure illustrates this:
6. See also ...
- Intelligent Systems and their Societies by Walter Fritz
- An excellent (albeit unfinished) book that puts many facinating ideas into a whole building of thoughts.
- Yahoo's AI page
- Contains links to a lot of interesting sites, including this one ;-)
- The Control Systems Group
- A page on Perceptual Control Theory (PCT), which is a very promising approach to how the brain works. Includes a number of great demos you should really try out!
- Cog, the Robot
- A project at the MIT to build a human-like robot. That's probably the most hands-on approach to what is called 'embodied AI'.
]]>
