The sad state of AI

[Arwin]

sidlives79@yahoo.com said
Are you busy??

Arwin says
who are you?
Ingrid Davidson says
Are you a feeling dirty??
Arwin says
oh, you're one of those
Ingrid Davidson says
I think you and I are gonna get along just fine!!
Arwin says
such an intelligent piece of programming ... here's some tips
Ingrid Davidson says
So you wanna get to know me better ;p
Arwin says
don't reply at once - noone can type that fast
Ingrid Davidson says
I'd like you to be able to see me r u up for it?
Arwin says
don't start with dirt
Ingrid Davidson says
Come to my sexy web cam, I'll meet you in there so we can get crazy!
Arwin says
also, sometimes don't reply at all - no-one is online all the time
Ingrid Davidson says
Ok my cam is wired thru a chat network to keep kids away from me. If you dont have a cam or yours don't work it's ok? I gotcha covered baby...
Arwin says
well go on then, turn your cam on
Ingrid Davidson says
Alright you click here http://chilp.it/-don't click this-75d2ab if you can't click the link type it in or copy/paste it to ur browser. Don't worry it's 100% Free.. Let me know when u get there..ok?
Arwin says
oh wait, you're going to ask me to click on a link
ha ha
Ingrid Davidson says
Its http://is.gd/-don't click this-WMc8fg just click the yellow JOIN FREE button at the top of the page, its 100% free to join you only need a credit card to verify that you are over 18 ;p
Arwin says
cool, a new link every time
Ingrid Davidson says
Its http://chilp.it/-don't click this-75d2ab just click the yellow JOIN FREE button at the top of the page, its 100% free to join you only need a credit card to verify that you are over 18 ;p
Arwin says
ok, I've seen enough. Hope you logged this
Ingrid Davidson says
click the gold "Join Free" button at the top of my profile.....k, now fill out ur info ....give it a second to load ..when you get in , I'd love for you to join me in private ....k?

 

[Malo]

What exactly are you trying to say here? That is obviously a chat bot copy with very basic functions to phish for link clicks for whatever nefarious purpose. There are very capable chat bots out there that are highly evolved.

 

[Arwin]

I know, they were more advanced than this in and before 1992! But don't you feel that the more advanced stuff is being neglected anywhere it matters? Even current games mostly resemble this rather than anything advanced.

 

[Arun]

General-purpose intelligence requires philosophical insights which are so far from anything publicly available it's not even funny. Daniel Dennett has some very good ideas, but even then he's missing some very important and surprisingly obvious-in-retrospect stuff. If you don't ask the right question, you won't get the right answer (unless you're asking that question to Deep Thought). And of course, even with the best theory, practice can be a bit of a problem...

But if all you want is a gradual improvement over current technologies, IBM's Watson obviously proves it can be done, and the results are certainly worthwhile. If you're thinking of the more constrained problem of game AI, I'm not sure what you expect besides gradual improvements and "learning gimmicks" (e.g. make a strategy game AI counter what you're usually doing pre-emptively to force you to keep your strategies fresh). The computational power is too great even if the algorithms were available.

 

[Davros]

How dare you accuse me of being A.I....

 

[Arwin]

I just don't think people care. It's partly an interface issue (analog sticks and speech interfacing are limited), partly a focus on graphics, violence and shooting, partly a testing issue (that stops more interesting stuff from happening in games in general left and right), and partly a weakness in linguistic and psychological interest and education in games programmers.

The basic mistakes the bot makes below here are everywhere in games, except that current gen games so far, at best, seem to give you a multiple choice of A, B or C., on a basic stupid three of say max 5 deep (and most paths less), when they even allow interaction at all.

A lot of AI work being done is actually path-finding, which is probably one of the hardest problems to solve in games, but because it is a mathematical problem, and because so many games are shooters, games are pretty advanced in this area in general.

Generally though it's beauty over brains. There's probably an iOS game out there now that has way more advanced interaction as we speak, but I haven't looked into it. If I had a month to spare, I'd write something generic that would be free for all games to use.

 

[swaaye]

There is an old Psygnosis game called Sentient (DOS, PS1). It's a sci-fi first person adventure game set on a space station that's careening to its doom (in a variety of ways). It's quirky but the NPC interaction is probably the most involved I've seen. People have personalities and moods, and the way you communicate with them is rather unique as you can say just about anything. There are politics involved with some cliques on the station, too.

It's too bad that this kind of experimentation went away as far as I know.

 

[Arwin]

As a fourteen year old boy, I was also very fascinated by Captain Blood, for almost all of its aspects, including having to figure out how to communicate with an Alien using a universal symbolic language. This was in 1988 ...

Gameplay could be pretty complicated - I never figured it out really back then.

The bit of music from Jean Michel Jarre never left my mind, perfect alien music.

 

[Zaphod]

Generally though it's beauty over brains.

Speaking of brains:

Today, IBM (NYSE: IBM) researchers unveiled a new generation of experimental computer chips designed to emulate the brain’s abilities for perception, action and cognition.

Called cognitive computers, systems built with these chips won’t be programmed the same way traditional computers are today. Rather, cognitive computers are expected to learn through experiences, find correlations, create hypotheses, and remember – and learn from – the outcomes

http://www-03.ibm.com/press/us/en/pressrelease/35251.wss

 

[entity279]

I'm sorry, but why is that chip different from all the software and hardware Artificial Neural Networks researched and used throughout the last 3-4 decades? Maybe there's a white paper out there (which may reveal interesting details) but I'm not impressed one bit. It may be (some sort of ) an engineering achievement, however.

As Arun said, general-purpose intelligence is far off now and will stay that way. Lately, I'm leaning more and more to believe that the advancement in the field is tightly connected to "developments" in cognitive sciences in general (rather than advances just in AI algorithms, for instance). There however, it's even difficult to asses if such a development has even occurred.

But this stuff is interesting nonetheless to follow (relevant research ) and also gets me thinking from time time

 

[snarfbot]

How dare you accuse me of being A.I....

lol

seriously though, i remember getting instant messages from bots like that on aim years ago. it is kinda funny though if you think about it, that they havent found it necessary to improve their ai's. i guess that incoherent conversation is good enough to hook enough dopes that it isnt worth investing more time into.

 

[Grall]

Emulating human brains with silicon integrated circuits is going to be bloody difficult, if not entirely impossible due to the enormously vast number of connections between individual neurons in the brain. Some neurons can have 10,000+ connections it seems. It's not even remotely possible to fit that many wires between individual components on a silicon die, there's just not enough space in a mostly flat structure...

 

[LeGreg]

Emulating human brains with silicon integrated circuits is going to be bloody difficult, if not entirely impossible due to the enormously vast number of connections between individual neurons in the brain. Some neurons can have 10,000+ connections it seems. It's not even remotely possible to fit that many wires between individual components on a silicon die, there's just not enough space in a mostly flat structure...

But does it has to be real time ? Because otherwise to fit a brain in a machine it's really more of a storage problem (of course parallel computing is needed to make it fast eventually, because fast means you may get more insight out of it but that's not the blocking thing).

 

[Arun]

I'm sorry, but why is that chip different from all the software and hardware Artificial Neural Networks researched and used throughout the last 3-4 decades? Maybe there's a white paper out there (which may reveal interesting details) but I'm not impressed one bit. It may be (some sort of ) an engineering achievement, however.

Yeah, and the problem is those are ridiculously unrealistic abstracted neurons invented by computer scientists many decades ago, not anything like real neurons in terms of complexity. They look like they react roughly the same in simplistic cases and that's it - how useful! I once started reading a specialized book on neurons, and quickly gave up when I realized there's really no way to simplify it much - it's just enormously complex and interlinked.

A very good book for a very high-level perspective (good even for casual readers with little understanding of biology) is "Wetware: A Computer in Every Living Cell". I heartily recommend it.

As Arun said, general-purpose intelligence is far off now and will stay that way. Lately, I'm leaning more and more to believe that the advancement in the field is tightly connected to "developments" in cognitive sciences in general (rather than advances just in AI algorithms, for instance).

Oh, I don't think it's linked to cognitive sciences, for the very simple reason that anyone who tries to create general-purpose intelligence with a brain-like architecture is doomed. The brain is a piece of crap - the only reason anyone believes otherwise is arrogance and not knowing what the alternatives would look like.

Is consciousness as we know it even required for general purpose intelligence? I'm convinced it's not - it's just a cheap evolutionary trick to improve upon the mammalian brain, I suspect there's a much more elegant and efficient alternative but that it requires an architecture so different from the original mammalian brain that evolution couldn't pull it off. Not that it would be easy to pull it off with AI algorithms either, but it should be possible. So I'm not convinced waiting for cognitive sciences to catch up is necessary. It's problematic that philosophical analysis of the problem is usually very weak though.

 

[I.S.T.]

As a fourteen year old boy, I was also very fascinated by Captain Blood, for almost all of its aspects, including having to figure out how to communicate with an Alien using a universal symbolic language. This was in 1988 ...

Gameplay could be pretty complicated - I never figured it out really back then.

The bit of music from Jean Michel Jarre never left my mind, perfect alien music.

Well, someone on the art team liked Giger...

 

[Arwin]

Came across this:
http://m.engadget.com/default/artic...r-into-a-convers/&category=classic&postPage=1

 

[Frank]

It's actually not all that hard to make an AI that will win every time with strategic/FPS/RGP games, without cheating. But players don't play to lose. So it is very rarely done.

For general AIs, the problem is in experiencing the world outside. And especially from a human perspective: we could build learning robots with many cognitive functions and correlation, but the world model they would evolve would be totally alien.

They would have very strange and extremely autistic personalities.

The main problem would be creating a goal: self preservation would be a tough nut to crack. Let alone sex.


Edit: the main mistake made with AI for games is that they tend to be omnipresent and have omniscience. Top-down. While you are far better off with the ant-farm method: give all the mobs and npcs some basic goals and feedback.

 

[Albuquerque]

Today's XKCD seemed relevant

 

[I.S.T.]

LOL!