Turing Test

[Resha Caner]

Has anyone here ever tried an actual Turing Test?

 

[Resha Caner]

I'll take it that's a no for most. Do you think you would do well with such a test?

 

[Tinker Grey]

I'll take it that's a no for most. Do you think you would do well with such a test?

I'm not sure what you're asking, but I'll attempt three different answers.

1) Do I think I could convince just about anyone that the entity typing this response is human? Yes. Justification: Well, I've been a member of about a dozen message boards and no one has accused me of being a bot (while other entities have been accused of being a bot.)

2) Do I think I could convince someone that I am, in fact, a bot when, in fact, I am not: maybe. I think I could learn how to mimic machines-mimicing-humans just enough to fool that someone. To some degree, it would depend on the parameters of the test: duration, subject matter limitations, etc. Without training, maybe. With training, probably.

3) Could I discern a bot? Well, I dunno. It's quite possible I've already been fooled. And, by definition, I just wouldn't know. If I were part of a research team, I might just become un-foolable ... until the technology surpasses my training.

Have I participated in some variant of what I've described? Not knowingly.

 

[PloverWing]

Do you think you would do well with such a test?

Do you mean "Would I do well at writing software that would fool the interrogator in a Turing Test?", or "Would I be able to persuade the interrogator in a Turing Test that I was the human?"

 

[Resha Caner]

Do you mean "Would I do well at writing software that would fool the interrogator in a Turing Test?", or "Would I be able to persuade the interrogator in a Turing Test that I was the human?"

Sorry. Neither. Read on.

3) Could I discern a bot? Well, I dunno. It's quite possible I've already been fooled. And, by definition, I just wouldn't know. If I were part of a research team, I might just become un-foolable ... until the technology surpasses my training.

I had this 3rd option in mind.

Have I participated in some variant of what I've described? Not knowingly.

No. Many people do seem skeptical that they could actually distinguish a conscious being from a well-designed machine. I'm not surprised. I've long thought that when we identify a person, what we're really doing is looking for something similar to ourself - not for some abstract definition of consciousness.

 

[PloverWing]

At our current level of technology, I have a sense of the kinds of conversations that have given bots difficulties in past Turing Tests, so I suspect I'd have reasonable success telling the bots from the humans. 50 years from now -- hard to say.

 

[Tinker Grey]

No. Many people do seem skeptical that they could actually distinguish a conscious being from a well-designed machine. I'm not surprised. I've long thought that when we identify a person, what we're really doing is looking for something similar to ourself - not for some abstract definition of consciousness.

Didn't the original Turing test specify some sort of inability to see the entity with which one was conversing? I think that goes to what you're saying.

Right now, the audible awkwardness of Siri, and Alexa, etc., give it away. Their responses are quite good for the parameters of interaction. So, I'm pretty sure I could be fooled, certainly over a short duration, by a bot. The longer the conversation and more broad the context would probably expose the bot.

It would surprise me if, as @PloverWing suggested, that 50 years from now I could be fooled a robot sitting in front of me (if I weren't allowed to attempt to lift it!).

 

[2PhiloVoid]

Has anyone here ever tried an actual Turing Test?

No, but as word on the street tells it, Arnold Schwarzenegger will be conducting one ... in the future.

 

[Resha Caner]

Didn't the original Turing test specify some sort of inability to see the entity with which one was conversing? I think that goes to what you're saying.

I believe you are correct about the original test, which came up in a philosophy discussion at some point, though I can't recall the context. Someone noted the common tendency of people to say things like, "I think I heard a mouse, but I haven't seen it yet." In humans sight is considered the primary, confirming sense. All others only provide secondary support. That has all kinds of interesting implications regarding our perceptions.

So I wonder about the Turing test. Is it really a test of the AI, or is it a test of the designer's intelligence?

 

[PloverWing]

So I wonder about the Turing test. Is it really a test of the AI, or is it a test of the designer's intelligence?

The Turing test is a measure of how successful the designer has been at designing an artificial system with intelligent behavior. The idea is that if humans are intelligent, and the system behaves so much like humans that an interrogator can't tell them apart, then the system is also behaving intelligently. It's not a perfect test, but it's one good test to use.

Tinker Grey is correct that Turing's paper specifies a setup in which the interrogator can't see or hear either the human or the machine. The interaction is only through typed conversation. (For a modern version, imagine texting or exchanging CF posts or email.)

 

[Resha Caner]

The Turing test is a measure of how successful the designer has been at designing an artificial system with intelligent behavior. The idea is that if humans are intelligent, and the system behaves so much like humans that an interrogator can't tell them apart, then the system is also behaving intelligently. It's not a perfect test, but it's one good test to use.

Meh. Unless this were some kind of double blind test where they would also have to be able to detect that a person is a person … or distinguish an AI designed by Caesar from an AI designed by me, I would maintain all you're really testing for is the intelligence of the designer … not something I'm convinced can successfully be done.

 

[FireDragon76]

I'm already convinced that computer AI can easily be more intelligent than the average human being.

I am impressed by our Amazon Alexa and it's ability to parse natural language and respond intelligently- it's almost capable of holding a real conversation. IMO, the pundits keep raising the bar on what constitutes "strong AI", but in many ways we have already arrived, to a certain extent. Machine sentience, on the other hand, is a philosophical, not a scientific, issue.

 

[durangodawood]

I had this 3rd option in mind.

Thats me administering the Turing test.

The "AI" is the entity 'taking' the test and being evaluated in terms of 'doing well' etc.

 

[Resha Caner]

Thats me administering the Turing test.

The "AI" is the entity 'taking' the test and being evaluated in terms of 'doing well' etc.

Sorry. The thread went cold for awhile so I've lost my train of thought. What point are you trying to make?

 

[durangodawood]

Sorry. The thread went cold for awhile so I've lost my train of thought. What point are you trying to make?

The Turing Test is meant as a challenge for the proposed AI. So when you ask if I would "do well", it doesnt really make sense.

 

[Larniavc]

Has anyone here ever tried an actual Turing Test?

Great question about, that subject. I also have an interest in, that subject. I would like the talk about, that subject.

Will talking about, that subject mean that you can work from home in your spare time?

 

[Resha Caner]

The Turing Test is meant as a challenge for the proposed AI. So when you ask if I would "do well", it doesnt really make sense.

I think it's relevant. Does the AI pass the test if I fail to detect it is an AI but you succeed?

My position would be you're not just testing the AI. You're testing the intelligence of the tester and the creator as well. If the purpose is to uncover AI, how do you control for those factors?

 

[durangodawood]

I think it's relevant. Does the AI pass the test if I fail to detect it is an AI but you succeed?

My position would be you're not just testing the AI. You're testing the intelligence of the tester and the creator as well. If the purpose is to uncover AI, how do you control for those factors?

The idea is, I think, that the Turing test administrator just has to be anyone we recognize as having "normal" intelligence. That's where the bar should be set to call machine success "intelligent".

Its really about that designation, what we can call intelligent, and not too much more.

 

[Resha Caner]

Its really about that designation, what we can call intelligent, and not too much more.

I could agree with that, and such corresponds with my position that a discussion about "intelligence" is really a discussion about "things like us", not an independent abstract construct.

 

[durangodawood]

I could agree with that, and such corresponds with my position that a discussion about "intelligence" is really a discussion about "things like us", not an independent abstract construct.

Hmm. Maybe. I mean, I dont think the Turing test was proposed as the one and only way to assign intelligence.... but just one sure way.