Evolution of Intelligence in a Box
In our class discussions on biological evolution we have defined the term crane as something that builds without intent and the term skyhook as something that creates with intent. The question of drawing parallels between a computer developing artificial intelligence and the evolution of human intelligence depends on how those two terms are treated. A computer with intelligence programmed into it has been the creation of a skyhook, in this case the computer programmer who set out to create artificial intelligence. If you do not believe that skyhooks have any part in evolution then you could not believe in the evolutionary development of AI unless you take into account the way that the human understanding of computers and programming would have to evolve and the fact that to be truly intelligent the AI would have to keep developing and evolving to be considered so. An intelligence created by a skyhook that does not develop independently of the skyhook is not truly intelligent; the intelligence must evolve on its own to qualify for that designation.
The interesting question derived by considering how biological evolution relates to artificial intelligence is whether an intelligence created by a skyhook, that is something created to deliberately have intelligence, is really intelligent or merely an extension of the skyhooks intellect. All of the tests designed to determine the intelligence of an artificial system depend on a computer being able to trick a real person into thinking that the computer is also alive. This feat has not yet been accomplished but even if it is what does that prove? That the computer can mimic human behavior; this does not mean it has intelligence. A human could program a computer to mimic human behavior but it would be restricted to what the programmer knew. A computer could have a completely static program and still theoretically pass the Turing test. Why though is the test for artificial intelligence whether or not a machine can mimic its creator and not whether or not the computer can learn? The computer that I am writing on at this very moment is running a program to detect whether the words I am writing make sense in the English language. Recognition of words being actual English words is one of the common tests judges for the Loebner prize use. Microsoft Word would be able to pass that part of the test because our standard for intelligent conversation is based on human conversation. To me this implies a limited view of intelligence although it would be the only kind of intelligence that could be completely created by human beings.
There are programs in the world that can evolve in a limited manner by having algorithms that allow them to pick and choose from stacks of code so the actual code of an individual part of the program is different from the others. This is the technique used in the program SWARMEVOLVE 2.0, which is an artificial life simulation used by Lee Spector, John Klein, Chris Perry, and Mark Feinstein described in their paper Emergence of Collective Behavior in Evolving Populations of Flying Agents. This random sampling is supposed to mimic evolution but to my mind this is not a very good model of evolution. The program SWARMEVOLVE 2.0 could be considered to match evolution if you saw it as Dennett did, with the giant library of Mendel. The problem I have with the method it uses is that none of the code is new, the combinations of it can be new but it is all still old code that the programmers wrote and left open for access. The program can only evolve in certain ways and can only change certain parts of its algorithm. This is done to prevent the program from altering parts of its code necessary to function so the program doesn’t accidentally kill itself but it is these man made guide rails that prevent computers from truly evolving.
The evolution of human consciousness is not fully understood by scientists today but one thing that most people can agree on is that there was no creator who inserted all knowledge into the human brain one day. We have had to accumulate our knowledge; every human must learn everything that they know from birth. To create a machine that can mimic humanity there is surely some way to just insert all the appropriate knowledge into it and expect it to perform on cue, it will not however understand all of he information it has been given and why its actions make it seem human. If there was ever a deviation from what had been programmed into the computer it would not be able to respond. There must be learning for there to be intelligence. I believe that the way to create a true artificial intelligence would be to program a machine to learn and then teach it, allowing the machine to create its own algorithms to store the taught data. This machine would have intelligence from its creation but at the beginning it would not be able to pass a Turing test, but of course a human infant would have a difficult time doing that as well.
This would be an example of a skyhook allowing cranes to do there work on an object the skyhook has elevated. Intelligence can only be a product of evolution, or of cranes. If a skyhook has created intelligence I don’t think we can trust that because the intelligence created by the skyhook would just be an approximation of its own without the ability to adapt to situations that the skyhook did not anticipate. . To me intelligence is more than the ability to fool others into believing in that you are intelligent or human. It is an ability to respond to all situations and to learn from those situations. An artificial intelligence must generate its knowledge in a fashion mimicking biological evolution, with facts and ideas the things that will grow or evolve. Without the evolution of ideas and concepts in the mind humanity would not be intelligent because our minds would be stagnant and the same principles apply to AI as well.