Re: AI
- From: Randy <joe@xxxxxxxxxxxxxxx>
- Date: Mon, 22 Aug 2005 07:08:52 GMT
An interesting post, but IMHO, it's irrationally exuberant in its lack of appreciation for how insanely hard AI has been, as demonstrated during the past 40 years of lots of very smart people making very slow progress in the development of intelligent systems.
Mike Laub wrote:
I think that we have already created artificial intelligence on the internet, we just need to give him/her a way of speaking. I would like to call this The first steps towards creating this artificial intelligence came from Google when they made the algorithm that promotes the best sites to the top of a search result.
IMHO, Google's algorithm has little in common with AI. If you look into the founders of Google and the origins of their ranking algorithm, you'll find that Brin and Page were Stanford grad students working in data mining, and that their big idea is based on trivial syntactic correlation between words, their locality in documents or web pages, and the number of incoming web links, and not the underpinnings of AI: causation, semantics, cognition.
Imagine building an intelligent framework does not recognize the existence or value of cause and effect, does not comprehend any subtleties of "meaning", and does employ any mechanisms of thought. In the absence of these fundamental underpinnings of "intelligence" (and many, many more), Google's inherent limitations reduce its potential to build intelligent system into recognizing static patterns of word forms that are "alike" and "popular".
For more on Google's ranking algorithm:
http://www-db.stanford.edu/~backrub/google.html
http://en.wikipedia.org/wiki/Page_rank
http://www.iprcom.com/papers/pagerank/
The thing that distinguishes human beings from animals is our ability to choose between two things, and Google does this when they put one website above another. If you agree with the book, Zen and the Art of Motorcycle maintenance, you understand that the quality is indefinable,
Actually, I think Pirsig *did* finally define quality, at least to degree that was workable and productive. Otherwise, the rest of the book's exploration of that concept would have led nowhere.
but it is the most important thing in life.
:-) An interesting assertion. Probably one that even Pirsig would disagree with, especially after losing his son at a young age.
For Google to see one page, and say that it is better than another, essentially has already created an artificial intelligence, we just need to find a better way for this algorithm with all of the data on the internet, to speak. I believe that we could understand the "collective soul" of the internet by using Google's page ranking system to make decisions. I think that discussion boards can be re-arranged in a way that, if they are combine an algorithm could promote advancements in AI.
I agree that Google's algorithm does possess the essential attribute of decision theory -- it employs a value function that identifies one alternative as being better than another. But it is unable even to deal in more than a single decision -- no consequences of decisions, no chain of consequence, no tactics or strategies. If that's AI, then it's purely reflexive. It might be enough smarts to direct a bacterium in its avoidance of pain and attraction to food, but little else. Even subsumption architectures based on finite automata have more utility than this (though not much).
But how might Google model memory? Learning? Planning? Perception?
If these are not representable in its architecture, Google will be enormously self-limiting and unscalable in its pursuit of achieving higher levels of intelligent behavior.
....
Google already has the technology to allow for synonym search. Google could include in it's ranking all of the websites that say, "George bush is an idiot", "George bush is a moron", and so forth.
No. Google does not model semantics at all. It's equally as likely to equate "Bush is a genius" with "Bush is a moron" unless the person posing the question injects the necessary semantics himself by posing leading questions, like:
match: "Bush is a " $A where $A is a FavorableAdjective
And an ontological lexicon has already been defined, such as:
FavorableAdjectives: nice, smart, clever, witty, genius
UnfavorableAdjectives: nasty, dumb, clueless, witless, moron
Truth Of course you wouldn't promote this website as saying, "Come to Google, we have the truth" you would say, "This is the collective soul of the internet." These are the decisions the internet would make if it was a person.
Of course, the Internet is so full of untruths that at best Google might hope to achieve accuracy, but never truth.
Transparency To help maintain a transparent process, you should list the top 10 pages that agree, with and disagree with, the idea.
"Top 10" by what criterion? Logically precise? Factually precise? Politically correct? Clever? Polarizing? Amusing? Popular?
An AI Game: I read "Agonistics: A Language Game" with great interest, because I have proposed a similar idea, however I have very little influence in the AI world, because I am just a recent graduate in electrical engineering and I work for the McDonalds Corporation designing the electrical part of their buildings, which very rarely exposes me to the world of AI. However, I think my proposal has merit. It, like "Agonistics: A Language Game" is designed to create a system where ideas can compete in a survival of the fittest-tournament. My proposal would also correlate the strength of an online character directly to the strength of an online idea, however I have some additional ideas. For instance there are a number of ways to tie different aspects of an idea to different aspects of an on-line character. For example, the number of people who vote on weather they agree or disagree could represent the strength of the online-character's attach. However the idea in my mind becomes very difficult to describe. I don't know if you have ever read "David's Sling" is very close. However many people would be able to participate at a time. I envision an idea at the top of a page with the online community brainstorming a list of reasons to agree or disagree with the idea. There would be no need to shorten these lists because the best ideas would go to the top.
Adversarial AI has been explored by a lot of folks, from robot wars to "the language complexity game", to adversarial games of all kinds, to numerous attempts to devise standard benchmarks for various AI facilities like planning, reasoning, parsing, translation, and more, much less the Turing Test itself.
Personally, I'm glad to see that interest in AI benchmarks seem to be on the rise, since it's always been hard to know whether one computer science theory or technique is superior to another until you build systems based on the proposed fundamentals and then have them compete head-to-head, or at least time their execution of a common suite of problems. Benchmarks have been the mainstay of perhaps the best example of success in CS -- computer architecture. In that case, the SPEC benchmarks are used extensively to evaluate novel CPU and memory system designs via simulation and modeling. I think similar benefits may befall AI's use of practical performance feedback.
That said, you're not alone in your belief that the web may be the origin of a "global mind" based on "web intelligence":
http://novaspivack.typepad.com/nova_spivacks_weblog/2004/06/minding_the_pla.html http://www.goertzel.org/papers/webart.html
However, what what I've seen, only a Net Kook (tm) would suggest that the web in its current state could create a self emergent hive mind.
If you have enough interest to explore AI further, I'd recommend picking up a copy of Russell and Norvig's "AI: A Modern Approach":
http://www.amazon.com/exec/obidos/tg/detail/-/0137903952/qid=1124691868/sr=8-1/ref=pd_bbs_1/103-6923077-9184631?v=glance&s=books&n=507846
It's a tome, but it's phenomenally comprehensive. Once you memorize this book, you'll be more than ready to begin your research on your PhD dissertation in AI. Seriously.
For lighter/cheaper introductions to AI:
http://www.aaai.org/AITopics/html/overview.html http://en.wikipedia.org/wiki/Artificial_intelligence
Randy
-- Randy Crawford http://www.ruf.rice.edu/~rand rand AT rice DOT edu .
- References:
- AI
- From: Mike Laub
- AI
Relevant Pages
|
