News:

The default theme for this site has been updated. For further information, please take a look at the announcement regarding HAF changing its default theme.

Main Menu

Is Google beating the Turing Test?

Started by Dave, May 20, 2018, 01:03:27 AM

Previous topic - Next topic

Dave

Not yet methinks.

Fair does their demo was excellent, voice intonations, pauses, responses, catch-up when first option not available etc, but within a narrow frame. At the moment, outside of that restricted frame, I doubt that you can carry on a typical rambling conversation. Ask it a question "out of left fiekd", what colour hair it has, what does it think of [any music style you choose], does it think gun control is a good thing etc,  etc. Then expand these questions, whatever the answer (if there is one) by seeking qualifications. "Why?" Is the most open question ever, and can lead to serial "Why"s.

How many man hours will it take to train a computer to respond exactly as a human does in every kind of eventuality? OK, I grant that there are people who seem to be unable to express any real opinion or knowledge about anything beyond a narrow personal interest frame - one sometimes wonder if such people could pass the Turing Test!
Tomorrow is precious, don't ruin it by fouling up today.
Passed Monday 10th Dec 2018 age 74

Dave

Thinking further on that Google demo: it would be so easy to "fake" it.

Well, not exactly "fake" it but prepare the ground. Have several people phone the salon to get a "set" of their responses - typical questions asked, standard time slots, procedures and protocols etc . Then tailor your algorythms, variables and limiting paramerters around that set.

Google is already pretty good at interpreting less than specific search strings. "Donegan+New Orleans" was enough to find the song I was looking for. Actually, after that I realised "date battle New Orleans" would have been better but my mind was still hsnging onto the song becsuse it was playing in my head but of the battle date was unsure!
Tomorrow is precious, don't ruin it by fouling up today.
Passed Monday 10th Dec 2018 age 74

Davin

I think that they are close to being able to not seem like a robot in certain situations, but not to the level where they can be tested to seem like a human yet.
Always question all authorities because the authority you don't question is the most dangerous... except me, never question me.

Dave

Quote from: Davin on May 21, 2018, 03:11:17 PM
I think that they are close to being able to not seem like a robot in certain situations, but not to the level where they can be tested to seem like a human yet.

Yes, I agree. As I ssid they act quite convincingly within a, very limited, frame of "stimulus-response" actions. OK, so do humans to a degree but we have the capacity to store many more "s-r" sets in a very small and power efficient unit. Plus, to a degree, ignore ir modufy such if circumstances require.

In a prog on BBC they interviewed soneone about a super computer roughly equivalent to a mouse brain in capacity and processing ability. If I heard it right, I only heard part of it. I wanted to know more but forgot which prog and channel by the next day! Memory fail on my part! I got to think of the part instinct plays in such beasties - "inherent programming" that they are born with - and how they weave experience and learning into that. Perhaps that is part of the purpose of that digital mouse brain, to study pre-programmed firmware/os - "instinct" - and learning from subsequent input.
Tomorrow is precious, don't ruin it by fouling up today.
Passed Monday 10th Dec 2018 age 74

Tom62

The universe never did make sense; I suspect it was built on government contract.
Robert A. Heinlein

Dave

Quote from: Tom62 on May 21, 2018, 04:05:27 PM
Quote from: Dave on May 20, 2018, 01:34:26 AM
Thinking further on that Google demo: it would be so easy to "fake" it.

Pretty much all tech demos are fake as hell.

Yup, much as I suspected - and more! Pretty stoopid really, these things always get found out - but politicians have survived telling lies for centuries. People will always tend to believe that which panders to their desires.
Tomorrow is precious, don't ruin it by fouling up today.
Passed Monday 10th Dec 2018 age 74

Arturo

Just thinking about this. Tech Demos are not fake. I thought I posted in this thread but the Tech Demos are more like Demos. They only show you a small piece that is meant to wow you. So in the case of Duplex, they might only be showing the parts that look good. And if you listen to enough of them back to back, you will hear the pattern in the way it works. But if you are at work and duplex calls you, it's not likely to show itself as non-human.

However the real reason I am posting now is because if Google Duplex is beating the turing test, then isn't every video game character since they started having narratives already beaten the Turing Test? The ability to demonstrate human like intelligence indistinguishable from actual humans is a pretty broad standard when you get to this idea that video games have been doing that for a couple decades by now.
It's Okay To Say You're Welcome
     Just let people be themselves.
     Arturo The1  リ壱

Davin

Humans have fidelity. Humans find fidelity natural. When you talk a person and they say the same thing the same way several times outside of an understandable task (like working at a cash register), they seem off. Fidelity can be tricked to a point, but so far, either the amount of responses is limited because many canned responses have been prepared, or the responses start to sound funny because the responses are generated.

Part of any good Turing Test would involve some ability to learn and some ability to predict based off of what was learned. So far we haven't been able to produce an AI that can do that yet.

I mean, if you talk to a robot in a situation like booking a reservation, then there isn't much opportunity to tell if the thing is a robot or a person. Though the voice thing makes it sound more impressive, I would have been more impressed if the voice had the same vocal fidelity as the person taking the call.
Always question all authorities because the authority you don't question is the most dangerous... except me, never question me.

Dave

Quote from: Davin on June 04, 2018, 04:02:24 PM
Humans have fidelity. Humans find fidelity natural. When you talk a person and they say the same thing the same way several times outside of an understandable task (like working at a cash register), they seem off. Fidelity can be tricked to a point, but so far, either the amount of responses is limited because many canned responses have been prepared, or the responses start to sound funny because the responses are generated.

Part of any good Turing Test would involve some ability to learn and some ability to predict based off of what was learned. So far we haven't been able to produce an AI that can do that yet.

I mean, if you talk to a robot in a situation like booking a reservation, then there isn't much opportunity to tell if the thing is a robot or a person. Though the voice thing makes it sound more impressive, I would have been more impressed if the voice had the same vocal fidelity as the person taking the call.

Much as I feel. If you are having a conversstion with a "black box" system, one where you know the input to the "box"  (your voice, responses and questions) and the output of the "box" (the Turing device's, or human's, voice etc) but have no idea what is actually in the "box" you need to be able to ask the very human, but often irrelevant, questions that humans do. Take an awful lot of memory and processing to cope with the infinite (?) number of possible questions and responses. Or falling back on get-outs, opt-outs and cop-outs that, eventually, give the true human a feeling there is a machine in the box.

I think it is the audio version of even the vwry best computer graphics, many tiny clues give it away to the discerning viewer or audience.
Tomorrow is precious, don't ruin it by fouling up today.
Passed Monday 10th Dec 2018 age 74

Davin

Quote from: Dave on June 04, 2018, 05:03:29 PM
Quote from: Davin on June 04, 2018, 04:02:24 PM
Humans have fidelity. Humans find fidelity natural. When you talk a person and they say the same thing the same way several times outside of an understandable task (like working at a cash register), they seem off. Fidelity can be tricked to a point, but so far, either the amount of responses is limited because many canned responses have been prepared, or the responses start to sound funny because the responses are generated.

Part of any good Turing Test would involve some ability to learn and some ability to predict based off of what was learned. So far we haven't been able to produce an AI that can do that yet.

I mean, if you talk to a robot in a situation like booking a reservation, then there isn't much opportunity to tell if the thing is a robot or a person. Though the voice thing makes it sound more impressive, I would have been more impressed if the voice had the same vocal fidelity as the person taking the call.

Much as I feel. If you are having a conversstion with a "black box" system, one where you know the input to the "box"  (your voice, responses and questions) and the output of the "box" (the Turing device's, or human's, voice etc) but have no idea what is actually in the "box" you need to be able to ask the very human, but often irrelevant, questions that humans do. Take an awful lot of memory and processing to cope with the infinite (?) number of possible questions and responses. Or falling back on get-outs, opt-outs and cop-outs that, eventually, give the true human a feeling there is a machine in the box.

I think it is the audio version of even the vwry best computer graphics, many tiny clues give it away to the discerning viewer or audience.
It's the movements in CGI that will give them away for some time. Clothing helps hide most of the problems areas, but the face is always going to be a problem. Something they should do, that I think will take it a big leap beyond the current level, is to treat skin like cloth draped over bones and muscles. As it is currently done, they just use motion keys along with "bones" on essentially a shell. I'm sure that since I had the idea, that there have got to be people working on it right now.

For the voice, and AI passing the Turing test, right now as little as asking the same question a few times reveals the machine. We still have a lot left to go, but I'm still excited by the progress.
Always question all authorities because the authority you don't question is the most dangerous... except me, never question me.