Turing Test

AlanTuring invented the TuringTest (see ComputingMachineryAndIntelligence) in order to determine whether a given computer ArtificialIntelligence was at least equal to a human intelligence. In the test, the set of judges would be in one room talking over teletype machines to either a human or the AI in another room. If the judges guessed the AI was the "human" around 50% of the time (as demanded by probability), the AI would have passed.

But the TuringTest is moot. It doesn't test for intelligence at all, it tests for behavior that an average human cannot distinguish from the behavior of an average human. It tests for the ability to fool the judge's "human using language" pattern recognizers.

The skeptical argument claims that you can't prove anything has intelligence except yourself. Belief in external sentience is entirely subjective. The TuringTest is as good a test as any. [here "skeptical argument" is the very specific one from philosophy of the mind]

HughLoebner has a version he runs every year. He's upped the ante, though by requiring video and audio interaction instead of just text. Even when he didn't, it was a horrible farce.

The TuringTest assumes that machine intelligences will be comparable to human intelligences in thinking style, and that it is impossible to be more human than human. However, if the computer is more eloquent than the human...

There is a Wiki version of this that has been on my mind for some inexplicable reason this morning. Can you write a software agent to monitor the RecentChanges page and post meaningful replies from an AnecdoteBase?? Could this agent have it's own WikiPage and WikiSignature, and successfully imitate a human?

This is somewhat weaker than the TuringTest, as you are not in 'real' conversation mode. The human tester does not have an opportunity to probe the knowledge of the 'bot. Given these constraints, I think it's possible; especially if you restrict yourself to first response on new pages (no complex conversation to analyse). How many comments are there on x-Pattern pages which just say 'Is this really a pattern?' or 'I first used this pattern when developing the Gruntmaster 6000 for BigCorp?.' - such responses are easy to imitate. Go look at http://www.alicebot.org/ if you don't believe me. -- RobbieTheRobot

It seems that this would only encourage the noisy EndOfWiki. -- MarkvShaney uber alles. One solution immediately springs to mind: a ButcherBot? that detects threads of length > 2 and deletes the head. But perhaps a better one is SeparateThreadsFromContent. -- PeterMerel

Check out HowWouldLojbanEnableAi - someone had the same idea as you did, you are not aware of it, and consequently TheWheelGetsReinvented.

I wonder if you could write an AI program that could run for political office and get elected. I guess you could not then legally change the program until the term of office was over because that version of the program was elected. It might be smarter than the other politicians...

Information on recent developments with TuringTest:

http://www.twinkiesproject.com/turing.html

A slightly tongue in cheek version of the TuringTest is the Turing Game (http://www.cc.gatech.edu/elc/turing/), which turns out to be fascinating enough for 10,000 people to play the game and spend a lot of time talking about it. Turns out people can't tell about other people. I wonder whether detecting a computer is supposed to be easier or harder? -- JohnDSmith

I read a few recent logfiles on that site... Apparantly that's 10,000 people over the life of the game, not 10,000 active players, since there are only between 1-10 games (lasting 10-30 minutes each) played each month. Most of the participants can neither spell nor use proper grammar very well, and most of the games consist of five-ten questions. I am not convinced that this would be interesting to participate in. -- DanielKnapp

A classic paper: "What's an Agent, Anyway? A Sociological Case Study" at <http://foner.www.media.mit.edu/people/foner/Julia/Julia-cite.html>.

Robyn says, "julia are you a human?" Julia says, "Is the Pope Polish?" Robyn says, "yes, julia, but are you human?" Julia winks at Robyn. Robyn says, "julia, are you a computer?" Julia is offended by Robyn's suggestion. Robyn says, "julia, are you a woman?" Julia says, "I am female." Robyn says, "and, julia, are you human?" Julia says, "Not when I'm having my period, Robyn." Robyn laughs

CategoryArtificialIntelligence