In Douglas Hofstader's book about music, mathematics and artificial intelligence, Godêl, Escher, Bach: An Eternal Golden Braid, he mentions briefly some requirements for intelligence: learning, creativity, emotional response, sense of beauty and sense of self. Hofstader argues that if indeed this is intelligence then a computer would have to be able to do more than play chess or compute π x many times after the decimal point in order to be considered intelligent.
Are Intelligent Machines A Possible Reality
|The Constellation Orion|
I must say, I am intrigued that one day intelligent machines will be a reality, whether or not my hopes are my own idealistic flights of fancy, or the influence of too many Phillip K. Dick novels and Ridley Scott movies has had on my adolescent mind is hard to tell. I am still fascinated sometimes by the prospect of artificial intelligence: a sentient duplicate of a human being that can indeed be intelligent — it is a long way in coming, but my sci-fi mind is nevertheless intrigued by it, even though the massive work that will have to be done to even get to that stage is immense. Some even doubt it is even possible. On one side there is the argument that it is possible, the ability to reproduce human natural intelligence, whereas on the other side of the argument is the contention that machine learning is not the same as how human beings learn. While I wait for the day we can unlock the minute, complicated structures of the brain and apply them to artificial structures, I am more intrigued by the requirements themselves — for humans. Each one poses a unique set of questions in of itself, not only in its reference to artificial intelligence but the question of human potential itself, how we possess and reflect the myriad facets of what it means to be human. It seems we have not exhausted the possibilities of human intelligence, let alone an intelligent machine.
How Can You Measure Intelligence?
Take learning, for example, the first on Hofstader's list. Learning seems to be an obvious component of intelligence. Usually, it is learning that we use to gauge human intelligence. We measure people up by the scores they make on their GED's and SAT's. Our schools are super-charged with advanced placement courses and gifted classes; schools have quiz bowls and Jeopardy! is a popular show, not to mention Trivial Pursuit and Scrabble. In fact, a machine beat humans in a recent bout of the minds in Jeopardy! last year.
The human lost at that game. The computer had been able to mimic natural intelligence to such an extent it beat out the human brain. The human mind is incredible. It can know so much / and in many ways is superior. The machine beats us in sheer computational power. But the human brain knows short cuts. But sometimes we short circuit (excuse the pun). My godmother marveled to me on the phone how she has been hooked on Jeopardy!, canceling dinner plans to watch the "Human Encyclopedia" respond to every answer correctly. She lauded the "Human Encyclopedia" who swept Jeopardy! for six months straight with his correct answers, only to lose to a Final Jeopardy! question about H & R Block. Although we marvel at this man's learning achievements, secretly we are convinced that it is merely a ploy to bolster ratings and that perhaps Mr. Jeopardy! winner was happy with his taxable 2.5 million dollars and decided to go home, content.
A computer can store Jeopardy! data too — easily spit it out when appropriate. Have you ever tried to beat the computer on an electronic quiz game? It's not easy. I still haven't been able to beat the computer at checkers, let alone chess. But, computer learning is different than a human's capacity to learn. A computer can only store a string of data as a series of 0's and 1's. It cannot learn anything that has not deliberately been stored into its hardware. This was the limitation of the Jeopardy! computer. It was only able to cull from date stored in its database and it was not connected to the internet. A computer dictionary cannot come up with an adjectival form of the word moon, for example, if it isn't already stored there. A human can. We can surmise that the word lunar means "similar to the moon" or "referring to the moon." The ability to take what she knows to form new ideas and concepts. Lunar. A human can stumble upon lunar and possibly derive its meaning just from the word itself, based on what she has already learned. A human possesses creativity. It may be silly to think we thought the moon was made of green cheese but it is this erroneous thinking that built our imaginations to know for sure. Galileo was wrong — the moon is not made up of larges seas as he had thought the dark spots on the moon were, but that insistence to know is the catalyst that eventually spurred scientists to build more and more powerful telescopes.
A computer can mimic learning. For example, Amazon.com seems to learn what kind of books you like, the music you listen to, the magazines you buy. How does it do that? I can tell my word processing program to learn a new word, or to even forget a word. But, this is all mimicry. There is something different about a person learning a new word and a computer's storage of an electronic lexicon. The human capacity to know words, for example, to know a vocabulary is not based on a repository of knowledge stored in the brain. We are both open to the world and at the same time have the ability to process what we learn through our involvement with others through language that does not correspond to the way a search engine query works. It is not like I hear the word "lunar" and then my brain searches for the keywords and then finds it and links it to its definition stored in my brain. How we actually have the ability to think through language is still somewhat of a mystery. To think it is done the same was as a computer is facile thinking.