True AI PC GAME?

[bochen2]

Are there any computer games out there that can transcend the sum of its total programming and can patch and mod itself and reaching omega point and singularity to develop self consciousness?

 

We already have games out there like Crysis that approach photo realism in terms of sheer graphics

 

As far as true interpersonal AI and sentience goes we are still at the ENIAC / DOS stage. Why have we made so much progress on bronze and ZERO on brains? Characters in games are getting more and more prettier and their intelligence levels stays the same, making them effectively appear dumber and dumber...

 

Why can we have a natural language interpreter with software simulated neural network personalities capable of carrying a real conversation ? Why is every PC game so damn stupid and the characters so "dumb"? Why can't we create a game that is actually compelling and interactive/dynamic?

[alexander]

there are games that serve as research projects for AI, these are mainly written in Haskell and Lisp. But this all depends on your definition of a "True AI" game.

 

to answer your last question, because nobody would play it.... rather its a better investment of money to develop the latest graphics for a mediocre game plan, then it is to design top noch AI, and have people get bored because they get beaten all the time...

[Tormod]

IMHO artificial intelligence in games has little to do with graphics. Making a game look good doesn't make it's characters lifelike. You need an AI engine which mimics life and that is extraordinarily hard to do.

 

Here is an interesting page about this topic:

The Game AI Page: Artificial Life and Flocking

 

I can also recommend the book Creation by Steve Grand - the maker of several games based on artificial intelligence as the premise.

 

Amazon.com: Creation: Life and How to Make It: Steve Grand: Books http://www.amazon.com/Creation-Life-Make-Steve-Grand/dp/0674011139/

[CraigD]

Are there any computer games out there that can transcend the sum of its total programming and can patch and mod itself and reaching omega point and singularity to develop self consciousness?

To the best of my knowledge – which is not specialized in game programming – no. Self-modifying code is not unheard of – many people familiar with fairly commonly taught techniques as genetic programming have at least with “blurring the boundaries” by having the problem being selected for being the genome (usually termed a individual’s “vector”) changer itself – but in my experience the state of the art of such code is toy-like and little used in production code. Personally, I’ve found it useful mostly in writing automations of turn-based games where the games rules explicitly involve changing the game’s rules, such as the puzzle game “Where Are the Cows”. Self-modifying code that plays this game, along with a very simple self-modifying code example, can be found in the 2-year old hypography post “Self modifying code in an interpreted language”.

As far as true interpersonal AI and sentience goes we are still at the ENIAC / DOS stage.

I’d say this analogy, if anything, overstates the state of the art of game adversary programming. Even under early APIs like BIOS/DOS, computers were still general purpose computers. The use of the term “AI” to describe video game adversary programs is more of an advertising gimmick than of serious computer science.

Why have we made so much progress on bronze and ZERO on brains?

This question can be translated into more or less “why has the computer industry for the past half century been what is has been?”, a pretty deep and controversial question, but I’ll go with a short answer with two parts: “because the problem of AI is hard”; and “because that has been what the market has demanded/accepted”.

 

I believe I’m somewhat typical of programmers of my are group (born 1960), in that I expected Turing-test successful AI to be commonplace by at latest the mid 1990s. In retrospect, I’ve come to believe that futurists and technologist like me, and like famous one such as the late Arthur Clarke from who’s work so many of us formed our expectations, were badly ignorant of and failed to account for the importance and workings of social/commercial markets. For reasons too involved to broach in this post, there are strong long-term financial reasons for commercial enterprise not to develop AI.

Characters in games are getting more and more prettier and their intelligence levels stays the same, making them effectively appear dumber and dumber...

An interesting point – though it would be difficult to make an adversary program appear dumber than many of those in some of the arguably best video games ever made. I’ve recently acquired an old N64 and in the course of replaying an old classic, 007GoldenEye, find myself stuck on a level, a situation that draws attention to the adversaries’ stupidity. The 12-year-old N64 is a startling sophisticated machine (one of its processors actually has game cartridge-loadable microcode!), which even at a mere 94 MHz on its general-purpose CPU, could manage a lot more “smarts” in its adversary programs. Nonetheless, due to superbly designed levels and trigger conditions, GoldenEye remains an enjoyable game.

Why can we have a natural language interpreter with software simulated neural network personalities capable of carrying a real conversation ?

I assume bochen means “why can’t …”

 

This question appears to be a simple restatement of the Turing test question, another old, deep question. Suffice it to say that there’s far from a consensus that a real or simulate neural net is a good approach to this problem.

 

Of course, the generic answer to “why can’t a computer do X” is another question: “why don’t you quit your complaining and code it?” ;)

 

Seriously, I think the most promising communities for working on true AI programming are academics and hobbyists. Professional programming - business, scientific, game, and combinations – largely lacks the ability to delay return on investment long enough for such work. In addition, there’re those “strong long-term financial reasons” to avoid bringing true AI software to market I alluded to earlier.

Why is every PC game so damn stupid and the characters so "dumb"? Why can't we create a game that is actually compelling and interactive/dynamic?

One reason, I think, is that, given the popularity of multiplayer games, there’s arguably not much need for an AI, when there’s a word full of human intelligence who, will pay to fill the role of an AI program.

 

Another – though I’m not certain about this, having only personal and anecdotal evidence to support it – is, I think, that the majority of gamers who prefer single player games don’t like the novelty and unpredictability that sophisticated adversary programming provides. Like all commercial software, to be successful, video games must be liked enough to be purchased in large quantities and/or high prices, ideally while being developed with the low cost/effort.

 

Last – again, guessing, as I’m not well acquainted with recent examples the type – I suspect that the place to look for the most sophisticated adversary programming is not in graphically-intense games like 1st/3rd person shooters, but rather in graphically trivial, turn-based RPGs. After all, the classical Turing test is turn-based.

 

A final point of accuracy re:

…reaching omega point and singularity to develop self consciousness?

Chardinian or Tiplarian omega points, Vingian singularities, and computer self consciousness are distinct concept. A self aware computer program might be run without resulting in an immediate or eventual in Vingian singularity, and neither are the omega point, an event by definition of universe-wide scope.

 

The three concept do have at least one attribute in common: in varying degrees, many experts thing them impossible, others inevitable.

[alexander]

There are plenty of games with advanced AIs, some of which are centered on creating most advanced AIs, some, using genetic algorithms to modify the thinking, by that having a self-modyfying and learning AI systems. I mean notably Galactiv Civilizations, bSerene, AI Wars, Dungeon Keeper (this uses behavioral cloning and actually learns from how you play, actually they claim the most advanced AI robots, bots that have over a meg and a half of info, dedicated to it's intelligence characteristics), Dark Reign, and others.

 

Some do things that none of the other games do, most are platforms for research and are ever-modifying resultants of thereof... None are fully what a true AI is though

[Galapagos]

Can anyone point to where I can read about the most advanced artificial neural network in existence? I know there are some that recognize faces, process language etc, but how close can they get to "doing what we do"?

 

ps- might be of interest to some reading, the Loebner Prize contest was held yesterday

 

Machines Edge Closer To Imitating Human Communication

“Today’s results actually show a more complex story than a straight pass or fail by one machine. Where the machines were identified correctly by the human interrogators as machines, the conversational abilities of each machine was scored at 80 and 90%.² This demonstrates how close machines are getting to reaching the milestone of communicating with us in a way in which we are comfortable. That eventual day will herald a new phase in our relationship with machines, bringing closer the time in which robots start to play an active role in our daily lives”

[alexander]

that once again depnds on what you call an AI network, i, for example, have as much fun time reading about research into ant behavior and getting a network of very simplistic bots to immitate that behavior, as i do about the Kiyomury robot that does martial arts, or the darpa challenge cars, all have some level of intelligence, infact the small ant bots have very little of it, yet they are both interesting behavioral studies, and they both have some ai, the ants have a simple one, but together they solve what looks like a very complex problem, using a very simple instruction set, while Kinomuri has an extensive network of computing devices and a fairly advanced technology to make it move and make decisions like it does.

[Theory5]

The only way I can think of to create a more intellegent AI model would be an inceasingly complex system of choices and data that the program writes itself. The more choices the program has to choose from the more it simulates realism. And a way the AI could learn is to write data and read it. For example, if the AI gets shot in an FPS from standing up to long it could write some data that would tell it what happened. That could be turned into a choice by a sub-process or command, and would be inserted into the AI's code. Or if it was a less important choice it could be written indexed and read later using a complex system of indexing so the program could find it almost instantly.

This is just an idea of how to make an AI simulate intellegence. Although this program would be big and probably slow, it could be networked with others.

[alexander]

The only way I can think of to create a more intellegent AI model would be an inceasingly complex system of choices and data that the program writes itself.

are you familiar with languages like Lisp and Heskell?

 

let me run you down the list, if you are not, and try to show you how we have these technologies, and they are used in most of today's AI research.

 

Lisp - one of the most commonly used languages for AI development, it's mathematical basis (lambda calculus), linked list foundation, eaily implementable tree DS, and code that is processed as an object and can be manipulated as such, make it one of the most ideal languages for AI developers. What's better then code that can easily run any genetic algorithm calcuations, and continuously modifies itself?

 

Prolog - interesting language, here everything is a relation, the program is ran by running data over the relations to achieve some sort of a result, thus the language is good for applications where a lot of reasoning occurs, such as AI :)

 

Haskell - not necessarily a really great language for AI design, they are working on creating an AI toolkit, with probability calculation libraries, fuzzy logic implementations, and polytypic unification, paired with insane math processing back end, once again, weilds some interesting ideas that make for good AI models, there's a game written in Haskell that scientists are using to study behavior, dunno what it's called, i remember you had to manage some dwarf colony or something, build, defend, attack, feed, etc, to keep your people in line, and the behavioral modeling was really interesting in it...