Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Sowelu]

Well, Milo certainly doesn't work, or they wouldn't have needed to show an increasingly-blatantly prerecorded demo faked up to look like it wasn't prerecorded.  Of course we have some of the AI tech.  I am always very interested in the yearly turing test competitions... 

What we don't have are good common-sense everyday-life bots that can act convincingly.  The Turing test entries still cop out half the time and refuse to answer your questions precisely, but they do so in a human way--but you can't dodge questions forever.  Although Milo didn't say very much that they couldn't have done, he just wouldn't be interesting for very long.  Milo is supposed to show interest in the user's life, but that requires a lot of learning...conversation and introductions start out very formal with very little assumed information, but as formality decreases, assumed information and references become way more common and that's hard to deal with.  And learning, itself, is so very tricky.  For example.  Milo asks "What did you do today?"  You say "Oh, I wrote a story."  What's Milo going to say?  If he doesn't even ask what it's about, then he isn't very interested.  If he asks what it's about and you say "Oh it's about a dragon" and he just says "That's nice", then he isn't very interested.  Really showing interest would mean asking "Is the dragon a good guy or a bad guy?" or something, and that takes a LOT of understanding and common sense to know enough to even ask!  I can't even guess on a timetable for having that, but I'd put it ten years from now at the very earliest that it would be seem remotely creative and understanding when you interact with it, and even then only on very very limited subjects--it would be an AI specifically designed to understand fantasy literature at a very basic level, for example, that might be able to grasp the concept of some of the Little Golden Books and not apply that knowledge outside in any way.

I haven't seen many AIs yet that pretend to have an ongoing life outside of interacting with people.  That sounds fun to work on.  Again, the chatbot entries usually just have to respond to like 25 statements and then the round ends.  Milo clearly has some kind of homework that he's supposed to be doing, enough to provide context from the last few days at least, which is nice.

We don't have good enough image recognition.  I keep close tabs on the state of the art, there.  The database of past experience is not anywhere near good enough.  We're very lucky to get 70% accuracy in clear and simple photographs, and that's for recognizing things like "the sky", "a road", "a tree".  Tech to recognize written numbers and letters with human-level accuracy is moderately recent (but has enormous commercial uses), and that has no creative elements whatsoever.  You'd need an enormous common-sense library to make it recognize all the symbolic representations that people draw--the way Western society draws a symbol of a fish is probably very different from other cultures, even--and it would be a huge stretch to take 'recognize a written numeral' and turn that into 'recognize something from this symbol library where each symbol can be even more different from the baseline'.  Five years before we can recognize a small library of drawn symbols like fishes and mammals and cars and people with 50% accuracy.  Thirty years and a major worldwide community 'common sense' symbol recognition contribution project before it's viable on a wide range of stuff.

I just don't see how this could stay interesting for very long if you try to have a real conversation.  It'd be like having a game set in a really deep and pretty forest, except there's invisible walls on either side of the path so you can't take a closer look.

Besides, if I had random AI kid to talk to, I would push that AI's creativity to the limit.  Within the first ten minutes I'd be trying to get him to give me ideas to continue writing a story, I'd be asking him personal beliefs in philosophy and trying to change them, I'd be trying to teach him how to play Tic-Tac-Toe (or some variant that he's not preprogrammed with, like...a 4x3 board), and trying to tutor him in math--and if he was preprogrammed, I'd keep raising the level to calculus and beyond.  And then, discovering that none of those things work, I'd get bored and leave.

[Sordid]

You want a case in point? Graphics.

Look at graphics 6 years ago. If they advanced that much, so did AI.

You're forgetting that the robot was an actual cutting edge research institute, Molyneux is a videogame developer. And also that robot acted based on nothing else than your tone of voice, while Molyneux claims to have all that plus voice recognition and language comprehension advanced enough to hold an actual conversation with the program. Advancement, yeah, sure. To that extent? I rather doubt it.
I'll tell you what Molyneux is, he's like those scam artists who go "oh we're developing a free energy machine and it's looking really hopeful, of course all we have so far is a non-working prototype, but all we need is some of your money to finish up the development and then you're going to be really really rich". And after a while they ask for even more money to develop a bigger and better v2 prototype, and the longer you go along with it the more money you sink into it and the more you keep thinking that the breakthrough's got to be right around the corner, any day now, and you're unwilling to let it go because you've invested so much into it. It's the exact same routine, only he's pulling it on the videogamer community. Well I for one won't be suckered into that. Yeah, sure, I'm going to try his new toy. But I'm not going to pay for it unless I'm really impressed, and given his track record I rather doubt that's going to happen.

[Yanlin]

Or... You know... Stop underestimating science?

It all has to start somewhere. This will be about as much as a botch as Mirror's Edge.

But in 10 years from now, we can expect realistic parkour open world simulations which simulate things like injuries, fatigue and friction.

Same with AI.

[Karlito]

I have decided that the video of the woman playing the game is fake, and she's just acting through it. During the part with the water, she slips up, and her movements don't match with the movements on the screen. There's a apart where her left arm on the screen shoots out before her actual arm does.

This was exactly the impression I got.

[Sordid]

Well that was kinda my point, this isn't science. It's a videogame developer claiming to have something even scientists dream of. I just don't see that happening, any more than a garage enthusiast building a car in his back yard that can compete in a Formula 1 race.

Oh, and in ten years parkour will have ceased to be a fad, so I'm guessing that we won't see those. But yeah, I get your point. Or rather that's my point too. Yeah, in ten years we're going to have some awesome stuff. But Molyneux claims to have that kind of stuff now. That's why I'm skeptical of it, and by the sound of it you are too. It's not going to be all that he claims, is all I'm saying. Yeah, it's going to be innovative and maybe even fun... but not quite as breathtaking as this fake demo.

[cowofdoom78963]

Oh, and in ten years parkour will have ceased to be a fad

Pft, I was jumping off roofs way before it was cool.

[Enzo]

Look at graphics 6 years ago. If they advanced that much, so did AI.

If you'll forgive the analogy, that's like saying I'm way smarter now than I was in highschool because I can bench press twice as much.

Molyneux should have been in politics. Or something. He gets that crazy glint in his eye and it's hard to tell what exactly he's thinking about. I haven't ruled out world domination. But yeah, I facepalmed repeatedly when I say saw this.

[Sordid]

Oh, and in ten years parkour will have ceased to be a fad

Pft, I was jumping off roofs way before it was cool.

So was I, but it's not parkour if you land on your head.

[Neonivek]

It's a videogame developer claiming to have something even scientists dream of

Ohh my goodness... The technology is OLD!

Milo is a lot more simple then he looks. For example if you told a joke that made no sense and wasn't funny... Milo would laugh. If you yelled at him harshly with compliments he would be sad. Milo guesses your intent based on tone of voice.

After that he is basically a living grammar check.

What is left, to fill in the gaps left by the technology, is scripted events and scripted speach as well as likely limitations intentionally thrown in. (For example Milo doesn't allow continuous speach and will ignore you during some events)

or they wouldn't have needed to show an increasingly-blatantly prerecorded demo faked up to look like it wasn't prerecorded

You also forgot that they obviously bribed everyone who got to see a demo of this with big loads of cash! Look the "It was faked" doesn't seem plausable due to how needless it would have been. It would be like lip singing a song when you have real talent.

It is also harder to make her fake the video when an easier method would have been to heavily script the game so no matter what she said it would have caused those prompts. (Which Id more then consider reasonable and plausable)

But no apperantly I am supposed to believe that Lionhead is so stupid that not only could they not use technology years older but that they would rather 3d animate a video and rely on a woman who act out what was happening on the screen TWICE perfect and not even bother to edit the video... (Unless of course they edited it afterwords by editing on her movements in the water... but that just raises more questions) Then make the silly mistake of actually letting people try it out to find out how fake it actually was.

[Sordid]

It's a videogame developer claiming to have something even scientists dream of

Ohh my goodness... The technology is OLD!

Is it now? Can you point to even one example of it other than the robot video I posted? Some actual commercial application, maybe?

Milo is a lot more simple then he looks. For example if you told a joke that made no sense and wasn't funny... Milo would laugh. If you yelled at him harshly with compliments he would be sad. Milo guesses your intent based on tone of voice.

Tone of voice in and of itself isn't enough to distinguish a joke, so no, I don't think it's quite that simple. To my knowledge there still isn't AI advanced enough to reliably understand even plain speech, and understanding jokes and other non-literal meanings would be orders of magnitude more complex.

You also forgot that they obviously bribed everyone who got to see a demo of this with big loads of cash! Look the "It was faked" doesn't seem plausable due to how needless it would have been. It would be like lip singing a song when you have real talent.

Hm, I can think of at least one way to fake a live demonstration like that, not to mention that a closed-door demo isn't exactly very impressive to anyone on the wrong side of the door.

[Sowelu]

Except that the water part WAS clearly faked.  There's no way to defend that whatsoever.  And if part of it was faked, there's no reason to think the rest wasn't faked as well.

The demo that has an actual conversation has not much to do with tone of voice, the impressive part isn't much of that at all--Molyneux was going on about how the human player could read Milo's expressions and responses more than the other way around, but it seemed like a real conversation.

Did you watch the movie, Neonivek?

[woose1]

Milo is a lot more simple then he looks. For example if you told a joke that made no sense and wasn't funny... Milo would laugh. If you yelled at him harshly with compliments he would be sad. Milo guesses your intent based on tone of voice.

Tone of voice in and of itself isn't enough to distinguish a joke, so no, I don't think it's quite that simple. To my knowledge there still isn't AI advanced enough to reliably understand even plain speech, and understanding jokes and other non-literal meanings would be orders of magnitude more complex.

Most likely it's a mixture of both, using the crappy emotion detectors we have now along with the crappy speech recognition software we have now to make a conclusion of what you're talking about.

While Milo is probably not as good as we saw in that video, it certainly is possible to recognize the meaning of certain sentences and words.

Here's an example:
You say to Milo, "Go to your room you little bastard!"
Milo hears, "Go to your room" in a loud voice, and/or "bastard" in a loud voice. 'He' also hears the entire speech fragment in a loud voice, and so determines that you're angry at him. If 'he' managed to pick up the 'Go to your room' part, he will go to his room.

You say to Milo, "...and the doctor says, 'That's not my wife, it's my truck!'".
Milo hears the entire speech fragment in a normal volume, with the end of it in a higher and harsher volume. 'He' also recognizes that the original question (A speech fragment with a normal volume ended with a lower volume) does not include enough words from his frequently asked questions, like "Where do you live", or "What do you like to eat", and so determines that he does not know the answer, or it is a question lining up to a joke. And so he says, "I don't know, what is it?", then laughs or responds accordingly depending on your answer.

And no, the water part wasn't faked. We have easily accessible technology to do that, and even if it was faked, it has nothing to do with the AI. So stop discussing it.

[Sowelu]

If you don't think it was faked, you haven't watched the whole movie, and you need to watch it again.  It's faked because it shows a reflection of the actress, except it's not synced.  The reflection moves her left hand into the water a full half-second before the actress does, and you can see that the actress noticably goes "Oh oops" and tries to catch up.

Look, ad hominem might be a logical fallacy, but that doesn't mean it isn't always justifiable.  The water part IS FAKED, very obviously so, because it is badly out of sync with the actress and the reflection often moves BEFORE the actress.  It's faked.  That puts the entire movie into question.  That movie has exactly as much relevance to the finished product as watching something from Pixar.

If you want to discuss only what Molyneux says, then fine, but the movie is not a demonstration of real tech whatsoever--except for the graphics.  The entire movie is pre-rendered.  What they demonstrated is non-interactive.  I'm sure this is what was explained behind closed doors.

[Vahan]

No, but the fact that it seems they had to fake the water bit doesn't really support the idea that they're any further along than I am, does it?

( Note: I only read AI design, I haven't even done anything to do with it, yet. )

[woose1]

Alright, let's say the entire damn thing was pre-rendered.
Ok....

Does that still make the fact that 'AI' like he's talking about not possible? If you want my true and honest opinion, here it is: Old Pete faked most of (If not all of) the video, it was a publicity stunt showing things Peter hadn't implemented into game form yet. While it's possible (All of it, including the water), it doesn't really exist in that form as of now.