Dumb AI to smart self-aware programs

[edepot]

It seems artificial intelligence routines are not up to par with technology. They are still following predefined scripts, and those scripts are usually simple state machines with no real dynamic thought. As programs (and games in particular) concentrate on getting the most out of the graphics, the artificial intelligence is put on the back burner, and speed of execution is more important than intelligence.

However, no matter how much AI in games and programs advance they are still not reaching the ultimate goal of their existence: Self-awareness. Many movies and science fiction have already delved into these (like skynet and HAL and others), no one has actually delved into the requirements or general direction needed to advance into this realm of self-aware programs, at least until now:

http://www.edepot.com/forums/viewtopic.php?f=12&p=2556

Now we are able to finally meet the next frontier of computers. Self-awareness and the ability to do what we humans have been doing all along.

 

[neliz]

a whole lot ado about nothing? You're talking about AI and how it should be different from "dumb" programs but yet later on you limit it by confining it to the same programming rules. And, nice catch BTW on throwing away the evolution theory and trying to focus this on a creationist. Maybe that's the reason the article is limited?

How does one go about creating procedures that state-check (power) without it becoming paranoid? as you said, the A.I. should satisfy it's primary needs first before embarking on other routes. how close in paranoia intertwined with self-awareness and self-preservation?

IF you force a routine to state-check against a primary need and give it the consequences (power-down) of the lack of the primary need and the speed at which the power-down would occur, how long would it take for the AI to spend 98% of it's time checking his power input?

One cannot try to translate instincts into routines since instincts are un-learned responses, an A.I. doesn't begin to lactate when a baby cries, it needs to be programmed to do so, no A.I. could ever "instinctively" write a routine that responds to a new factor since it's parameters are not known, even A.I. needs to be maintained.

 

[Arun]

Errr... If you believe in what seems to be Intelligent Design then, by definition, you believe that 'automatic competition' cannot generate results as complex as human beings. If you believe that then, by definition, you believe that you can't create an intelligent being except through programming hardcoded routines and then letting them organize themselves slightly differently to react to events.

And so it blows my mind that you believe your approach is fundamentally different from "They are still following predefined scripts, and those scripts are usually simple state machines with no real dynamic thought." - it's not. The only thing you're really proposing is a vague method for the AI to be more reactive to the human's actions. It's just an incremental change to the current approach which, you know, is also reactive to the human's actions by selecting which strategies to execute and when to change strategy. What do you think a top-notch RTS AI does?

Much of the complexity lies not in that, but rather in *how* to form these new strategies. Just saying "hi computer, design new strategies based on a strategy to design new strategies!" won't get you far. I'm sorry to be so blunt, but there is no way that anyone who believes in Intelligent Design will ever solve that problem. That's because it's fundamentally a competitive and evolutionary problem (competition between ideas, possibilities) and if you believe competition is not a sufficient answer to life's complexity, then you're off to one hell of a bad start...

 

[willardjuice]

Needs more Strong AI...

 

[Davros]

Have you ever been to a msgboard that deals with a.i programming
Its just a bunch of people argueing wether program x,y,z qualifies as a.i

 

[trinibwoy]

We need to work more on real intelligence before worrying about artificial ones

 

[BoardBonobo]

You also seem to imply that in every Human should by default be identical copies of their progenitors. Since we don't program the genes of our offspring to be different to us, which according to your system, we would have to.

Evolution takes care of that little problem for us by recombining genes and adding that little element of chance in there, mutations (good or bad), misreads in the tRNA etc.

But then again, if you believe in creationism\ID you don't really go for random chance do you?

 

[neliz]

So, after a week, the MC Hawkings Allstars 7, Honk if you love the Creationist 0.

 

[MrFloopy]

So, after a week, the MC Hawkings Allstars 7, Honk if you love the Creationist 0.

Not wanting to get into that debate, but isn't A.I. by definition I.D.?

 

[Arun]

Not wanting to get into that debate, but isn't A.I. by definition I.D.?

Most AI researchers would look at it that way (or would at least agree to describe what they're doing that way after a long debate). I don't think it's necessarily the case though.

 

[BoardBonobo]

Not wanting to get into that debate, but isn't A.I. by definition I.D.?

I guess you could describe the majority of the symbolic AI in that way. Second or third generation non-symbolic systems are themselves free of their designers hands, SANE systems for example. I prefer to look at the creation of these systems as a form of abiogenesis!

 

[Arwin]

Not wanting to get into that debate, but isn't A.I. by definition I.D.?

It doesn't seem like that ... ID uses two principles: irreducable complexity, and specified complexity.

The first is something which perhaps only certain neural net applications match, but not really. The second, well, most A.I. systems really aren't that complex.

Of course, all A.I. systems do have sort of an intelligent designer (hopefully, though I often question that too ), but then definitely not the all-omnipotent kind that I.D. refers to.

 

[Simon F]

Dumb AI to smart self-aware programs

.....
However, no matter how much AI in games and programs advance they are still not reaching the ultimate goal of their existence: Self-awareness.

What makes you think that a self-aware program would want to play against a human player and not get completely bored and go and sulk in the basement like a "Sirius Cybernetics Corp" lift?

 

[neliz]

Depends on what your applying the term A.I. to.

everybody shouts A.I. nowadays but [get(location) move x,y] is not A.I.

in my book A.I. is still something that is programmed (at a base) but has the ability to add or modify it's own code to adjust to a situation and NOT wait till routines are added by a programmer.

Easiest would be to create a fighting game A.I. that would not simply scan inputs, but create responses based on previous matches. It would show easily if this is working because no one will come up with "the game is broken, I only need to use this attack to defeat x character"

Using probability mechanics to set responses to opponent actions. I only wonder how to "even" the playing field when it comes to the various actions. I guess you could let an A.I. try to learn "combo's" for itself in such a game but it would take mere seconds to figure out every combo available.

Now you have something that is faster than a human, can state-check without losing focus, something that has a response to everything you do (be it through default actions or probability -> Opponent jumps? 32% chance he combo's 68% chance he'll kick -> wait for block, then counter.) on his main task and generally beat everyone in the world after a couple of hours of "training."

now..if I could just write the procedure "Training" i'd be rich.

 

[Arwin]

You'd think so but personally I think the reason why we generally have weak A.I. is simply because it's low priority and therefore low-budget in most games.

 

[BoardBonobo]

Depends on what your applying the term A.I. to.

everybody shouts A.I. nowadays but [get(location) move x,y] is not A.I.

in my book A.I. is still something that is programmed (at a base) but has the ability to add or modify it's own code to adjust to a situation and NOT wait till routines are added by a programmer.

Easiest would be to create a fighting game A.I. that would not simply scan inputs, but create responses based on previous matches. It would show easily if this is working because no one will come up with "the game is broken, I only need to use this attack to defeat x character"

Using probability mechanics to set responses to opponent actions. I only wonder how to "even" the playing field when it comes to the various actions. I guess you could let an A.I. try to learn "combo's" for itself in such a game but it would take mere seconds to figure out every combo available.

Now you have something that is faster than a human, can state-check without losing focus, something that has a response to everything you do (be it through default actions or probability -> Opponent jumps? 32% chance he combo's 68% chance he'll kick -> wait for block, then counter.) on his main task and generally beat everyone in the world after a couple of hours of "training."

now..if I could just write the procedure "Training" i'd be rich.

But we have all that! We had all that when I was doing my degree in AI (Yes such a thing existed - a degree in pure AI!).

ANNs learn by training, most black box systems learn by training. It just happens that the number of epochs required to do most training would mean that, in a game, the player would probably have died of boredom before they get to play properly.

Nearly all AIs are faster than human, no AI loses focus! Depending on how badly you wanted the player beaten, and how much memory you are willing to throw at the problem space, you could easily create undefeatable AI (yeah ok so Go! would be a problem...).

AI in games is simple because its two decades out of date. Simple path finders and search tress, A* alogortihms et al. They're just so yesterdecade. It just boils down to money and the fact that gamers don't really want smart AI, they just want the illusion of smart AI. who wants to get beaten everytime by a game? That would suck!!

 

[Arwin]

Yeah, that's why I'm saying it's a budget thing. I also started on a degree in A.I., but in 1992 it was the second year that was at all possible and I was thoroughly disappointed with the sorry state of education at university so after two years I decided to go for an English degree instead, figuring that being able to read the latest developments in English on the net at twice the speed of my peers with double the understanding would be better for my career than staying there and killing myself out of sheer misery (not in the least because out of 186 students, 6 were girls).

 

[MrFloopy]

Second or third generation non-symbolic systems are themselves free of their designers hands, SANE systems for example.

Still not seeing the difference. I don't think I.D. proponents claim that the designer still controls their behaviour? Even then second or third generation systems were still the result of a design.

 

[BoardBonobo]

Still not seeing the difference. I don't think I.D. proponents claim that the designer still controls their behaviour?

ID proponents claim that god designed every living thing, they recognise no evolutionary process at all. Some also say god designed evolution. But that kind of makes god a bit useless, so lazy it might as well not exist.

Even then second or third generation systems were still the result of a design.

In the same way that a separate species arises through a process of evolution which in itself is a kind of directed design, not randomly chaotic as some would have you believe. By the second or third generation of most self organising systems they bare very little resemblance to their progenitors. The only difference being, of course, that computer programs, no matter how sophisticated, exist outside the influence of Natural Evolution and have none of the external pressures you would expect.

The next step of course would be to give a simple black box evolutionary network an external form that would be subject to Natural Selection. Admittedly you would either have to bring in reproduction of some kind (parthenogenesis or regular xy transfer) or at least the ability for the AI to rebuild a better model and transfer itself to the new model.

That's the kind of simulation I'd like to build but though the AI's ability to learn is with us the technology for building an external body that is affected by natural processes isn't. Well not to my knowledge anyhow.

Geez... I've just rambled!

 

[MrFloopy]

But that kind of makes god a bit useless

Or just efficient

Geez... I've just rambled!

That's why I was hesitant on getting into this thread! There is an entire satircal tangent that I'd love to go down with regard E vs C and AI but it's fraught with danger from both sides, and I can't be stuffed!