One of the main themes in science fiction today seems to be the Technological Singularity. In short, it’s a hypothetical event in the not-so-distant future where a more-than-human intelligence has been created. The logic goes thusly: If one were to build a more-than-human intelligence; this artificial intelligence (AI) would know how to improve itself better than the humans. It would proceed to do so. Due to the rapidity in which the AI can cause changes, the development of this AI and any related technologies would be exponential. Although the theory itself is sound, we still have many obstacles before such a thing could happen.
I.J. Good summarized it well when he, in 1965, wrote:
Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an ‘intelligence explosion,’ and the intelligence of man would be left far behind. Thus the first ultraintelligent machine is the last invention that man need ever make.
Later authors have revisited this subject quite often, one popular version being “The Singularity is Near: When Humans Transcend Biology” by Raymond Kurzweil. It’s a great read if you have the time, and can be read online on Google Books here. His other books are definitely recommended as well.
But what’s in the way of this happening today? It’s quite easy to look at Moore’s Law (“Since the invention of the integrated circuit in 1958, the number of transistors that can be placed inexpensively on an integrated circuit has increased exponentially, doubling approximately every two years“) and assume that we’re going to get there sooner rather than later. Ray Kurzweil has said, in his essay The Law of Accelerating Returns, that – based on Moore’s Law – one can assume the Technological Singularity will occur and be publically available “for $1,000 around the year 2023″. Other independent researchers have also reached the early years of the 2020:s as the most plausible date for this.
If we assume that Moore’s Law continues ad infinitum, I’m sure this is true. However, here’s the problem: With today’s understanding of what a computer is and does, a computer is very good at doing what it does when it knows how to do it. If I were to give my computer a list of all the books I own and told it to sort these books according to the author’s surname, the computer would only know how to do this if it already knew a sorting algorithm. A human being, even if he’d never sorted anything in his life, could figure out how to sort items. He could, for example, figure out one of these sorting algorithms (there are many more):
- Pick a book at random and place it in the bookshelf. Then pick another book at random and place it in the bookshelf before or after the previous book. Pick a book at random and place it before, between or after these two books. Repeat until finished.
- Place all the books in the bookshelf and start from the left. If the book you’re looking at has an author whose surname comes after the next book, move the book to the very end of the list and start over. This takes a long time, but is very efficient.
- Place all the books in the bookshelf and start from the left. Compare the first two books and rearrange them so that the first author, alphabetically, comes first of these two books. Then compare the next two (book 3 and 4) and do the same until you’ve done all books. Then start over, but skip the very first book and start with books 2 and 3. Then do books 4 and 5. When you’re done, start from the beginning again. Continue doing this until you’ve gone two full runs from beginning to end without swapping any books. Again, this takes a long time but is very simple to do since you don’t have to focus on more than a very small area at a time.
A human could figure these things out quite simply, but a computer would have to be told (firstly) what an alphabetical order is and (secondly) how to actually sort things. Until a computer is able to figure these things out on their own, we’re never going to be able to emulate a human more-than-human intelligence that we can begin to call an Artificial Consciousness. This means that the first question we need to tacke before we approach a proper AI (or AC) is: Can the properties of the mind be completely described on their own terms as an algorithm?
To answer this, we need to ask ourselves what the properties of the mind are. Is it feasible to fully understand the functional behavior of the brain and its structure and develop a mathematical model of it?
This question – this argument – could be approached from another perspective, however. Assume that we are able to fully map the human brain. We still don’t know the function of it or the way that things work together, but we do understand how to map every single neuron. Assume that we copy every single neuron from a human brain into a separate computer program that operates independently and is able to talk to other neurons in the same way that it does in the brain. We’d have a computer running a series of millions of different softwares that correctly emulates the human brain. Now; I understand that there wouldn’t be a brain stem for it to interact with, the sensory aspects of the simulated brain wouldn’t be getting information from the skin, eyes, nose and the other sensory organs. However, would we have the remaining functions of the brain? Would this hypothetical group of software be able to think? Would it, due to the high speeds of the computer, be able to think faster than the human brain? Would it be able to have independent thought?
Interesting questions, and I’d love to discuss them further with you, my dear readers. Please comment!
