Sentience

[mordant]

Originally Posted by Oakback

I believe I watched a documentary ( if memory serves me, Elon Musk had some input ) where a computer generated "person", that was controlled by AI, would have sophisticated conversations with different participants. The experiment was to see how many people where aware they were talking to a robot. Only about half realized the "person" on the screen was computer generated.

The image they were talking to was scary real.

Yes there is a web site you can go to and talk with a virtual computer-generated "friend" and you can build out the facial features and hair style and so forth, decide if its male or female, and probably by now can play with aspects of its personality. It learns about you through conversation and does a pretty good job of pulling off the whole thing I understand. I believe it's one of those free things with an optional paid premium version.

Musk himself is about to show a prototype of a humanoid robot. His initial motivation is to deal with a so-called "labor shortage" in his factories. Like most such "shortages" it is a function of paying employees too little and treating them too poorly. So he's explicitly building slave laborers -- doubtless, the fantasy of captains of industry everywhere. In this particular case I think he's bitten off way more than he can chew. Just getting a non-specialized humanoid robot to properly hold and use a wrench is a very difficult problem with current technology.

He's not successfully pulling off his long-promised autonomous / "fully self-driving" cars, either. He's found enough dim bulbs to pay him $10K (now $15K I believe) extra for the half-baked version available now, on he gamble that the "full" version will be working Any Day Now. No matter how hard they try, the software is still confused by novel situations. As a result, they can handle freeways and most highways pretty well but they still demand you pay attention to the road and be ready to take over driving at any moment, and many cityscapes are too much for it -- so what's the point.

The inside joke is that if one of these systems can not be confused when it sees an ostrich in the road, we'll know there's actual reasoning going on, instead of just really expensive pattern-matching. Until then I'm not interested in trusting my road safety to a computer, thankyouverymuch!

[Oakback]

Originally Posted by mordant

Yes there is a web site you can go to and talk with a virtual computer-generated "friend" and you can build out the facial features and hair style and so forth, decide if its male or female, and probably by now can play with aspects of its personality. It learns about you through conversation and does a pretty good job of pulling off the whole thing I understand. I believe it's one of those free things with an optional paid premium version.

Musk himself is about to show a prototype of a humanoid robot. His initial motivation is to deal with a so-called "labor shortage" in his factories. Like most such "shortages" it is a function of paying employees too little and treating them too poorly. So he's explicitly building slave laborers -- doubtless, the fantasy of captains of industry everywhere. In this particular case I think he's bitten off way more than he can chew. Just getting a non-specialized humanoid robot to properly hold and use a wrench is a very difficult problem with current technology.

He's not successfully pulling off his long-promised autonomous / "fully self-driving" cars, either. He's found enough dim bulbs to pay him $10K (now $15K I believe) extra for the half-baked version available now, on he gamble that the "full" version will be working Any Day Now. No matter how hard they try, the software is still confused by novel situations. As a result, they can handle freeways and most highways pretty well but they still demand you pay attention to the road and be ready to take over driving at any moment, and many cityscapes are too much for it -- so what's the point.

The inside joke is that if one of these systems can not be confused when it sees an ostrich in the road, we'll know there's actual reasoning going on, instead of just really expensive pattern-matching. Until then I'm not interested in trusting my road safety to a computer, thankyouverymuch!

It was actually Google AI I saw. This is an example:


https://m.youtube.com/watch?v=_tjo7g_JlA0

[mordant]

Originally Posted by Oakback

It was actually Google AI I saw. This is an example:


https://m.youtube.com/watch?v=_tjo7g_JlA0

There's something a bit stilted and "off" about this "conversation". Some of the responses appear to reflect that the AI has a low confidence level in its understanding of the question and so it says as little as possible to stay out of trouble. The "guy" AI keeps overusing phrases such as "that is different", too.

Still, it's a long way from the Eliza programs written in BASIC from 1975 that I cut my teeth on. Eliza was modeled after a therapist and worked from very simple pattern matching and noncommittal responses typical of Freudian therapy -- patient says X, therapist prompts, "tell me more about X", patient says more, therapist draws them out on some particular phrase or word in the response. That was, for the 1970s, pretty impressive , too. But it doesn't hold up today any better than this clip will hold up in 2050, I suspect.

There's also some distraction going on. The female AI has some barely visible hand gestures to make her seem more "human", for example, but it is just appearances.

What they are finding out is that most humans will readily accept such creations and relate to them like humans based on a few cues like that and the fact that the AI will give them focused attention that they seldom get in their everyday life.