Opinions on the term "AI"?

[Skynet]

I was wondering what everyone thought about the meaning of the term "Artificial Intelligence". Most of the people I've talked with in the past seem to believe that the term isn't that helpful, since there's no clear definition of the term in a way that computer scientists and engineers can apply to their creations. After all, "artificial" has a very clear meaning, but what is "intelligence"? Is it efficient memory representation? Computational ability? Self-awareness? Ability to learn? Emotion? Or, on the lower end of things, is it just the ability to appear to have some sort of "human" characteristics?

Following from that, if there is no such thing as "AI" at this present moment, then what do we call that section of code which makes the bots in your FPS of choice a challenge? In the past, I've called similar modules "AI", but that was out of habit. I think there's a better name for it out there.

So, I ask CompSci.ca...what is AI to you? If, as I believe, we don't in fact have any examples of it in our modern computer programs, what should that sort of code be called?

[haskell]

AI is a huge field... So much so that people usually focus on a couple sub-fields of it.

Quote:

Typical problems to which AI methods are applied

* Pattern recognition
o Optical character recognition
o Handwriting recognition
o Speech recognition
o Face recognition
* Artificial Creativity



* Computer vision, Virtual reality and Image processing
* Diagnosis (artificial intelligence)
* Game theory and Strategic planning
* Game artificial intelligence and Computer game bot
* Natural language processing, Translation and Chatterbots
* Non-linear control and Robotics

Quote:

Other fields in which AI methods are implemented

* Artificial life
* Automated reasoning
* Automation
* Behavior-based robotics
* Biologically-inspired computing
* Cognitive robotics
* Colloquis
* Concept mining
* Cybernetics
* Data mining



* Developmental robotics
* Epigenetic robotics
* E-mail spam filtering
* Evolutionary robotics
* Hybrid intelligent system
* Intelligent agent
* Intelligent control
* Knowledge representation
* Semantic Web

That should help you get a grasp of how it applies, and some of the fields of AI.

And here is the definition you seek:

Quote:

The term Artificial Intelligence (AI) was first used by John McCarthy who considers it to mean "the science and engineering of making intelligent machines" [1]. Whilst this definition is concise itself, the definition of intelligence is vague enough to limit its practical use. e.g. it does not help answer the question "Is the chess program called Deep Blue an AI?". The terms Strong AI and "Weak AI" are used to narrow the definition

AI is an Interdisciplinary field of study that includes computer science, engineering, philosophy and psychology.

And yes, we do have modern examples of AI at work avilable at our fingertips. Any of these chatbots(ALICE, GOD, etc...) are all AI.

I hope this helps you get a tiny bit of a grasp of the gigantic world that is AI.

[Cervantes]

I'd say it's anything that learns unboundedly. You could write pong "AI" that learns how you play, but if that's the only thing it can ever learn about, it's not Artificial Intelligence.

That's not necessarily the definition I stand by, but one I wanted to through out there.

[Skynet]

haskell @ Tue Feb 13, 2007 7:32 pm wrote:

Wikipedia

I was interested in what people here thought, based on their own opinions and experience. It is my personal belief that "AI" is a hype word- that it's used to label technology that doesn't demonstrate any sort of "intelligence". If I'm doing image processing, I say "I'm doing image processing", not "I'm doing AI." If my robot needs to plan a path, I'd say "I'm writing path planners." Neither of those examples exhibits intelligence - image processing is primarily clever work with matrices and bell curves, and global path planners are usually done with a graph search. As a minor gripe with that particular article... behaviour-based/developmental/generic-psychology-term-here robotics are not "fields", they are "strategies." Just as a programmer chooses a suitable data type to hold information in a specific way, so too will a robotics designer pick a strategy for their system to use to solve the particular problem.

But, I digress. If you align yourself with the Wikipedia definition, that's cool. However, if you've got other insights, that's what I'm interested in. What do we call code that controls something in a way that appears human? As an example, I've helped with converting a RC car to drive autonomously. When it's driving, you'd think a human's controlling it. (Especially when one of said humans is chasing it around) However, the code is simple - some geometry wrapped in some conditions wrapped in a case structure wrapped in an endless loop. What should I call that code, if not "AI"?

Cervantes wrote:

...anything that learns unboundedly

That's a really good wording - I've never heard the concept stated like that before.

Quote:

ALICE chatbot

I couldn't leave this without some ALICE hate...
ALICE was a neat gimmick, but I don't think it was anything more. The so-called "AI" which runs it is a system which provides responses based on the interpretation of a prebuilt XML-format file. (In the grand tradition of inventing standards before a product, it's an 'Artificial Intelligence Markup Language') From the Wikipedia example...

code:

<category>    <pattern>WHAT IS YOUR NAME</pattern>    <template>My name is John.</template>  </category>

Oh damn, however will we do without...
Yes, AIML gets more complicated...but it's really nothing special.

[haskell]

Intelligence is hard to define. That the problem.

You asked AI. AI is a lot of things. It isn't what you see and hear about in media as AI. It is things like path finding, pattern recongnition, image processing and natural language parsing. They are qualities humans display, and are thus AI. We can decide how to get places, recognize patterns, process images around us(also known as sight), and we can understand natural languages. Basically AI is deriving the logic of human characteristics into a theory of computer science.

ALICE and other chatbots are much easier to make in PROLOG. Mix PROLOG and any general purpose language and you can make a powerful natural language parser, and make it capable or replying,

The reason I gave you the Wikipedia definition isn't because of the definition, but because of the fields it showed. The fields show what AI includes so people can extract their own definition from what is included within it.

[BenLi]

an even better question is whether true AI is possible, I mean you can program AI to learn, but the programmer knows exactly how it learns, and thus the result will always be predictable.

[Hikaru79]

BenLi @ Wed Feb 14, 2007 6:47 pm wrote:

an even better question is whether true AI is possible, I mean you can program AI to learn, but the programmer knows exactly how it learns, and thus the result will always be predictable.

Hm, not necessarily. There are many non-deterministic models of machine learning. For example, Go (the board game) is generally considered one of the holy grails of AI research; one of the most promising approaches right now (for small boards, at least), is called Monte Carlo evaluation (with UCT search) -- it is nondeterministic. It uses a huge amount of random positions to extrapolate a value for each node. Consider neural networks and genetic algorithms, which are also, as far as humans are concerned, non-deterministic. Even more conventional approaches, like the one Frank de Groot has taken with Moyo Go, which simply uses a very deterministic database of millions of positions, and "learns" to seperate a given position into subregions which are then matched against its ginormous pattern database, can hardly be called "predictable". In fact, Frank de Groot is a very very poor Go player himsef; his own software is capable of beating him hands-down, in a way that he is thoroughly unable to emulate.

So, I don't think it's safe to say that a programmer knows *exactly* how their own software learns. They programmed the method, but only in a very high-level theoretic sense. It has often been done that programmers have written programs to achieve a result, or develop a method, that they would not have been able to do themselves. (Scientists writing proof generators that can prove theorems they themselves cannot; Google's technicians wrote some translating software for Russian by having it "learn" from millions of cross-referenced translated works, but not a single member on the team spoke Russian!). I think that's when you know you are looking at "true" AI.

[BenLi]

wow, sounds impressive