This is intended to be read after Identifying the Differences Between Human vs. Computer Intelligence by those who are interested further

Jorgen Hookham
December 7, 2010


Why Computers Do Not

When Gary Kasparov was defeated by the chess-playing computer Deep Blue in 1997 a great milestone in computer science was achieved, but it was no triumph of artificial intelligence. The IBM team behind Deep Blue did not create intelligence at all; they simply assembled a combination of database, algorithm and computing power capable of computing it's way through a game of world class chess. By this I mean to say that intelligence is not vast processing power, an impressive database or a potent methodology: all of these things were assembled by the intelligences of the Deep Blue team, and Deep Blue the computer had no intelligence to speak of. In fact I take the stance that no 'Artificial Intelligence' (AI) to date has ever been intelligent, because they have all lacked two fundamental aspects of intelligence: creativity, intuition. “‘Artificial Intelligence’ is not a technology. It is a problem domain that is delineated by the criterion that arriving at solutions would require intelligence. We will recognize Intelligence when we see it; a Turing Test will not be required.” (Anderson, Artificial Intuition)

No definition of intelligence to date commands universal assent, so first let’s looks at the history of the definition of intelligence, then I will define my own best conceptualization of intelligence.

If we look back in to pre-history, we can imagine a cave man seeing another cave man use a tool for the first time: his brow rises, and he grunts in affirmation of the other caveman's smartness. There is no definition of intelligence here, but the essence of intelligence is in this moment.

Looking in to the 20th century, we can see a different definition of intelligence as necessity has dictated. In this time period, intelligence is seen as a degree of capability to produce, and to apply and recite catalogued knowledge. It is defined by a score on a standardized test, and described on a scale relative to the cultural or social habitat of a person as a number value: an intelligence quotient. The intelligence quotient is used in social studies to discover relationships between intelligence distribution and other variables, and to predict as well as rate the educational achievements of children. (Wikipedia: Intelligence Quotient)

Later in the 20th century Howard Gardner proposes the theory of multiple intelligences as a counter to the standing definition of intelligence, defined in psychometrics (IQ tests,) which he saw as an insufficient description of the wide variety of cognitive ability that humans display. An example of the important differentiation that this theory posits would be thus:

"A child who learns to multiply easily is not necessarily more intelligent than a child who has stronger skills in another kind of intelligence. The child who takes more time to master simple multiplication 1) may best learn to multiply through a different approach, 2) may excel in a field outside of mathematics, or 3) may even be looking at and understanding the multiplication process at a fundamentally deeper level. Such a fundamentally deeper understanding can result in what looks like slowness and can hide a mathematical intelligence potentially higher than that of a child who quickly memorizes the multiplication table despite a less detailed understanding of the process of multiplication." (Wikipedia: Theory of multiple intelligences, para. 1)

Another definition of intelligence from the same time period sees intelligence in somewhat of a different light: “(intelligence is a) mental activity directed toward purposive adaptation to, selection and shaping of, real-world environments relevant to one’s life.” (Sternberg, 1985, p. 45)

It is important to note the degree in which each of these definitions of intelligence relates itself to situational context. If we go back through the examples above with this in mind, we can see a chronological dissociation between the two. As far as an intelligence quotient is concerned, intelligence is inseparably linked to superficial context (a test score.) Later theories such as the ones posited by Gardner and Sternberg suggest a differential between face value and actual core intelligence. Psychometrics actually tries to describe such an underlying phenomenon of core intelligence with the 'general intelligence factor,' or g-factor. (Wikipedia: g Factor)

Human intelligence has a complex nature that is difficult to attribute to a single source. Perceptions and thoughts are reflected in the mind and the intellect acts as a sort of curator. It performs lower level operations such as categorizing and recalling thoughts, but it has higher order functions as well. They are the functions of creativity and intuition, which are imperative to the adaptive nature of human intelligence.



Creativity
A theory for helping to describe the creativity of intelligence is "neural darwinism," which suggests a sort of natural selection within the thought process. (Wikipedia: Neural Darwinism) As the mind receives new information, the intellect 'mates' the new data with other ideas and relationships to create offspring ideas. The offspring ideas that are 'strongest' are committed, and the weaker are discarded. New relationships and thought patterns are formed, causing whole parts of the neural network to evolve through this creative process. This is a fundamental aspect of intelligence that AIs do not possess.

Intuition
Most of human thought processes are not logical, they are intuitive. This is why computers have such a hard time with so many things that are so simple for our human minds. Computers have a very hard time in what AI researcher Monica Anderson calls "bizarre domains." (A New Direction in AI Research) In these domains logical certainty is an impossibility and the only course of action is to guess, and intuition is the art of guessing wisely based on patterns and past experience. A good example of such a bizarre domain is language. We can make computer programs capable of reading text and recognizing voice commands, but we cannot so far create an intelligence capable of understanding the meaning behind any real world form of human linguistic communication. The meaning of single words is one thing (type 'yellow duck' in to a google image search, for example,) but AIs can't interpret the meaning that emerges from larger strings of words. This illustrates the irreductability of language, because the meaning of a sentence is greater than the sum of it's words. Language is also deeply complex. The meanings of words are not organized in to a neat hierarchy, they are richly connected, just like neurons in the brain. Inside these deeply complex systems is where the meaning of a piece of language originates. Communication relies on a word having similar synapses in the receiving system to those of the sending system to convey meaning. This means that to understand language a computer will likely need a world model of it's own. An intelligence is necessary to navigate the waters of these systems. More than that, it is intelligence combined with experience that has brought us as humans to this common ground, and even if a machine were to become able to comprehend our language it would not be intelligent until it could keep up with and contribute to the evolution of language under said machine's own volition.

Closing Thoughts
I consider humans to be a form of technology, and the progress of modern biotechnology seems to reinforce this idea. We have created the first synthetic life (in the wired article Rachel Swaby actually says, “Man-made DNA has ‘booted up’ a cell for the first time.”) (Swaby, para.1), and we have sequenced the human genome. It seems likely that, given enough time, we will one day be able to create a race much like our own to take on the tasks that we often envision artificially intelligent robots taking on. At that point it would be hard to refute the authenticity of an AI’s intelligence, or their consciousness. Perhaps it’s better that AIs are not intelligent and self-aware in a human way, so that we can keep them doing our dirty work without feeling too bad about it.

Works Cited
A New Direction in AI Research. Vimeo, 2009. Web. 7 Dec. 2010.

Wikipedia: Intelligence. Wikipedia. Web. 7 Dec. 2010.

Wikipedia: Theory of Multiple Intelligences. Wikipedia. Web. 7 Dec. 2010.

Wikipedia: Intelligence Quotient. Wikipedia. Web. 7 Dec. 2010.

Wikipedia: g Factor (psychometrics). Wikipedia. Web. 7 Dec. 2010.

Wikipedia: Neural Darwinism. Wikipedia. Web. 7 Dec. 2010.

Neisser et al. “Intelligence: Knows and Unknows.” Wikipedia. University of Illinois, 1996. Web. Dec. 7 2010

Anderson, Monica. Artificial Intuition. Syntience. Web. Dec. 7 2010.

Sternberg. Beyond IQ: A Triarchic Theory of Intelligence. Cambridge University Press, 1985.

Swaby, Rachel. Scientists Create First Self-Replicating Synthetic Life. Wired, 2010. Web.
Dec. 7 2010.

September through December 2010 was my first semester of education since I left high school in 2007. In one of my courses, Contemporary Issues, I wrote a paper (a very short one) entitled "Why Computers Do Not," which was meant as a counter to the hopelessly superior article by Marvin Minsky of MIT "Why People Think Computers Can't."

When it came down to writing out my argument, I consistently ran in to the problem of being able to think through how in fact a computer program could be written to breach any limitation I tried to put on computer intelligence. It turned in to a very painful exercise as the logic of my arguments, which attempted to put a definitive limitation on computer intelligence, came back upon me. It occurs to me now that logic is itself very mechanical, and that no argument to prove my point could ever arise from such a domain.

Since that first semester, I've been reading, in parts, Godel, Escher, Bach: an Eternal Golden Braid by Douglas R. Hofstadter. I have just finished reading one part in particular which has given me a new perspective on the point that I was trying to make in my paper one semester ago. In it, Hofstadter has written about a computer's ability to be made entirely unobservant, whereas for a human this is impossible, and also the ability of human intelligence to remove it's self from a system. An example of the latter, from Godel, Escher Bach:

"For example, a human being who is reading a book may grow sleepy. Instead of continuing to read until the book is finished, he is just as likely to put the book aside and turn of the light. He has "stepped out of the system" and yet it seems the most natural thing in the world to us."

This example made me think that the reason I had so much trouble distinguishing human-specific intelligence traits from computer AI was that these distinguishing traits seemed so natural to me, even to the point where stating them in my paper would have seemed ridiculous. For example, another excerpt from Godel, Escher, Bach:

"...a car will never pick up the idea, no matter how much or how well it is driven, that it is supposed to avoid other cars and obstacles on the road; and it will never learn even the most frequently traveled routes of its owner."

I found these two examples to be, for the moment, helpful distinctions.