User:Eric Norby/Norby Test
The Norby Test is a test for a machine's ability to demonstrate Darwinian capabilities, and perhaps one facet of advanced artificial intelligence. The basic test is to enter a computer against humans in their ability's to program. Should the computer win the competition, it would be deemed a better programmer.
The advanced version of the test is for a machine or a group of machines/computers to compete in making a better Artificial Intelligence (AI) program/machine than a human or group of human competitors.
Ultimately both humans and machines will design and judge the competition.
History
[edit]Eric Norby, while studying Electronics, Mechanics, and Computers at Washington state University around 1976-84, began dreaming of ways to get computers Amdahl 470 V8, TI-59's calculators to program themselves, thinking that would be the first steps in a computer's ability to "think" and "evolve".
Around 1982,83 he was introduced the idea of programming competitions by a fellow student/Teaching Assistant. Following that, a connection between computer competitions and the Turing Test was made. A side issue was also introduced when it was heard that "Airplanes were discovered, by Wilber and Orville, only after others had given up trying to make airplanes fly as the birds do. Airplanes are not birds." That rapidly led to the connection that the Turing Test was a test for a computer's human "like" intelligent, however, computers are not humans. Therefore something must be wrong with the concept of following the Turing Test.
Around 1985-86 the idea was first mentioned to coworkers at the Boeing company. It was received with distain, and lack of foresight. Back then, the Norby Test idea was just, to test a computer against humans, to see whom was a better programmer, and how the Turing Test idea was stalling the advancement of AI Science.
Airplanes don't fly better than birds by flapping their wings faster. They fly better than birds because they fly differently. Computers compute better because they think differently. To mimic a human is not going to get us better than a human. As in airplanes, it probably won't fly at all.
During the 1990's the idea became one where the test was to see whom could program a machine to be the better or best AI., which now could be called the Advanced version, or just the Norby Test.
During the 2000's the idea was bounced off some Internet Forum members and was again not fully understood.
Ramifications
[edit]The outcome of a computer winning programming competition, especially of programming an AI, is that human programming will be superseded by the programming computers, just as 'human computers' were replaced by 'electronic computers' in the 1950's 60's.
Computer AI's would have the ability to evolve independently from humans. Not that they will, at first. This would potentially lead to an increasing evolution speed for computer AI. One day they may even exceed us in warmth and friendliness.
The Norby Test tends to invalidate the Turing Test, as some sort of stage production. The Turing Test is similar to getting a bear to dance or a chimpanzee to paint. A superior computer would not prove much by devolving conversation just to mimic a human.
AI's need a goal to give direction; the Norby Test is one such goal.
Current Work
[edit]Watson has been demonstrated in competition on Jeopardy. This shows that AI's can in fact be better than humans in human like general competitions. Even the programmers of Watson are discussing the possibility of getting Watson or other AI to program AI's.
The concept of "better" is important part of the Norby test. The Turing Test just degrades what computers are really good at to be "as good as" humans. The Norby Test aims for supersonic airplane comparison to bird flight. Not that we still aren't learning about flight from the birds and fishes. As we won't stop learning about intelligence from humans, even with "super sonic" AI's.
Genetic Programming shows that computers can program themselves, currently limited to very specific ways. Perhaps, getting multiple computers to input and receive multiple specifics to/from a Watson like generalist, will lead research.