Google A.I. a Twinkle in Larry Page's Eye
This week Google hosted a European conference called Zeitgeist '06. It was held in London, giving Google CEO Eric Schmidt and founder Larry Page a chance to talk with and answer the questions of Europe's technorati. You can see a Q and A segment here. It runs about half an hour long and provides some interesting insight into Google. Eric even mentions the Space Elevator! But the most remarkable statements came from Larry Page himself. He spoke to the future of search, a future which contains a Google AI."People always make the assumption that we're done with search. That's very far from the case. We're probably only 5 percent of the way there. We want to create the ultimate search engine that can understand anything ... some people could call that artificial intelligence."
Larry's remarks didn't end there. He hinted that such things were already afoot at Google. He refused to predict when Google would achieve their goal of an AI, but he did say that "a lot of our systems already use learning techniques".
Larry noted how powerful an AI powered search engine would be. "The ultimate search engine would understand everything in the world. It would understand everything that you asked it and give you back the exact right thing instantly," saying, "You could ask 'what should I ask Larry?' and it would tell you."
He finished his AI thoughts on a promising note. Explaining that he has learned that technology has a tendency to change faster than expected, and that an AI could be a reality in just a few years. Those are very strong words coming from the mouth of one of the founders of a company with the wealth and vision of Google. Words to mark in the years ahead.
Before I continue I must admit to being entirely too fond of Google. I use their services constantly throughout the day. For search, email, chat, translation, news and maps. Case in point, this blog is hosted on Google's Blogger, and all of our posts are first prepared in writely. For me, a day without Google is like a day without sunshine.
That being said, I think Google is a different kind of company. They are a collection of thousands of great minds. Google's goal is the collection, organization and dissemination of all of the information possessed by mankind. You don't get much more ambitious than that. And with few debatable exceptions, they have stood by their motto of "Don't be Evil." One of their latest projects is a digital library of Alexandria. They plan to scan and index every book. Google has consistently been at the forefront of Web 2.0. Their offerings have a tendency to work both smarter and harder than that of the competition.
If any currently existing company has the were-with-all to bring us a functional AI, it is the brain trust at Google. In fact, futurist George Dyson has theorized that Google may already be home to a budding AI. "For 30 years I have been wondering, what indication of its existence might we expect from a true AI? ... Anomalous accumulation or creation of wealth might be a sign, or an unquenchable thirst for raw information, storage space, and processing cycles". Sound familiar?
Regardless of whether or not Google is currently housing an Artificial Intelligence, I don't doubt that sometime soon they will be. Let us put aside our foolish knee jerk skynet fears and think for a moment of what the future will be like in the presence of that much knowledge and information. The development of an AI will parallel the discovery of fire, the invention of the wheel, or the creation of the scientific method. Something that will change the very nature of the world. That's a lot of responsibility for any one to bear. But I cant think of a group more up to the task than Google. We at Memepunks would like to officially reserve our spots on the waiting list for Gmind beta, but they already know that. [inspired by the Guardian and Kazinform]
"In attempting to construct such machines we should not be irreverently usurping His power of creating souls, any more than we are in the procreation of children. Rather we are, in either case, instruments of His will providing mansions for the souls that He creates." - Alan Turing
posted by MEMEPUNKS at 8:59 PM
16 Comments:
So, what division of Google do you work for?
Memepunks is in no way affiliated with Google. Although given the opportunity to work there, I would do so in a heart beat.
A very good question though. For the record any such ties are always fully disclosed in the posts here at Memepunks. We stand behind our Bloggeristic integrity.
Thanks for keeping us honest, and thanks for reading Memepunks!
Uh-oh, you said "Web 2.0". That is a copyrighted phrase. Please refrain from using it.
Use: Web 2.1
The hopelessly gullible author of this twaddle enthuses: "Regardless of whether or not Google is currently housing an Artificial Intelligence, I don't doubt that sometime soon they will be."
I do. I doubt AI will ever exist. I doubt it in the same sense that I doubt anyone will ever produce a perpetual motion machine. Even calling it "doubt" is such a grotesque understatement that it qualifies as parapraxia. Everything we've seen from AI shows that we're taking the completely wrong approach. We're trying to create a Philosopher's Stone. It's not working, it hasn't come anywhere close to working, and most significantly of all, there's no sign whatever that any known approach will ever come close to working. Cyc, recursive search trees, heuristics, neural nets...they all hit a brick wall fast and splatter. And no one
has any idea to get past that brick wall.
The brick wall involves something simple: understanding the meaning of
things. No one has any idea how to make a machine understand what things mean. Assigning numerical values doesn't work. Modus ponens logic doesn't work. Neural nets don't work. What works? Nothing anyone has ever tried or ever thought of.
The total failure of AI in google can be demonstrated from the following simple example: "Mary saw a puppy in the window and wanted it." Now ask google "Which did Mary want -- the puppy, or the window?"
Google can't answer. It's helpless. Tell google "The astronomer married a star," and google's servers will shoot sparks out of their front panels and scream "Landru, help me, help me, help me!"
Google's dismal failure is nowhere better illustrated than a page search by date. Google does not understand that just because a date is MENTIONED in a page, that is not the date on which the page was CREATED and that is not necessarily what the page is ABOUT. As a result, google consistently lies when you restrict page searches by date. Google returns garbage...yet a 2-year-old child can give the correct answer.
That's not AI. That's the blind grinding of a robot's gears.
There is no sign that google will ever be able to surpass this roadblock. It makes searches by date utterly meaningless in google, and in all other search engines. For instance, if I want to restrict my search only to new information put on the web in the last 6 months about some subject, I'm
f***ed, stuck, and outa luck. Because google doesn't know the difference between a new web page whose content dates to within the last 6 months, and an older web page that has been updated in some minor way so that the updated age of its links shows as a couple of months ago. As a result, grossly obsolete web pages chock full of obviously worthless info keep showing up with a high page rank in google.
There are many, many, many other basic problems with google that make it next to useless. For instance, try googling my name. You'll get a hash collision. I am not a Presbyterian minister who lives in Washington D.C. -- but google doesn't know that. Google doesn't grasp the elementary concept that there could be two people with the same first and last names living on earth. Google is stupid. It's not AI, it's dumb numerical algorithms, and it's not getting any
smarter. Every effort in human history to reduce meaning to numbers has failed: Jeremy Bentham's Hedonic calculus, Leibnitz's proposal that arguments can be settled by reducing propositions to logic trees and parsing 'em, Rational Choice Theory in economics...it all fails and falls apart the instant it touches the real world. Rational Choice Theory assures us that Martin Luther King's voting rights marches in the deep south could not have happened. Leibnitz's injuction "calculendum" tell us that fundamentalist Islamic jihadis and fundamentalist Christians can solve their differences by parsing logic trees. It's so stupid it's insulting. It's beyond foolish, into a realm of hallucinogenically demented goofiness. Every effort to reduce meaning to numbers has failed, and there is no sign that it can ever succeed any more than Pol Pot's Year Zero could ever have succeeded.
AI belongs to the same category of mythtech as flying cars, videophones, food pills, personal jet packs, and nuclear power too cheap to meter. Never gonna happen.
@anonymous,
you are shortsighted.
How can something as infinite stupid as man be able to create something as perfect as AI?
We are to stupid. Much to stupid.
Anonymous,
As the "hopelessly gullible author of this twaddle", please allow me to retort. First off let me say that your opinion is valid, valued and welcomed here at Memepunks. But I cant for a second say that I agree with it. Your initial response seems to be one of blanket cynical disbelief. Not even allowing for the possibility of the development of AI. And you do so with such vehemence that you compare a functional AI to a perpetual motion machine. It isn't possible, it never will be possible, end of story. But things are not as simple as all that. Harnessing the power of the atom seemed impossible until that warm July day in New Mexico, 1945. Going to the moon seemed impossible until the Eagle landed in '69. As poet Bryant H.McGill said "Yearning for the seemingly impossible is the path to human progress".
History is filled with example of science breaking boundaries and shifting paradigms. It's hubris to declare something impossible simply because we don't know yet how to accomplish it. And for your example, you sited the apparent ignorance of the Google search engine. Well, you have succeeded in making Larry Page's point. Google is admittedly not very smart. And that is precisely his problem with it. That is why he spoke of building an intelligent search engine, something that could understand the meaning of things. Which is admittedly a herculean task. But there are a great many minds and a great deal of money being put to it.
Take for instance the former Director of search Quality, and current Dirtector of Research at Google. His name is Dr. Peter Norvig. He has co-written what has become the industry standard textbook on Artificial Intelligence. Check out Artificial Intelligence:Modern Approach, it's a good place to start. It's also clearly written and easily understandable for the layman. As a side note, let me also mention that Dr.Norvig is a Fellow and Councilor of the American Association of Artifial Intelligence. These facts speak to me of a man who takes AI very seriously. And this man is the head of research at Google.
In closing you mention how reducing reality to numbers is a predetermined failure. Rational Choice Theory is applicable, when and where people are being rational. And you disagree with Gottfried Leibniz and his Calculendum. Leibniz was possibly the greatest mind of the 17 century... Lets not bring the children into this. For every cited shortfall of math vs. the real world, you have a Nash Equalibrium, Cumulative Prospect Theory, or Dempster-Schafer that does work. Do you propose we simply write off attempting to define meaning numerically? We would have to abolish a great number of scientific disciplines and discredit an army of brilliant minds. Now THAT is never gonna happen.
That being said, I would like to thank all of you; detractors supporters, contrarians, and visionaries, for joining us here. Thank you for sharing your opinions and thanks for reading Memepunks!
I humbly suggest that a structural (rather than numeric) mathematics will be the basis for the "missing link" in AI. Indeed, such a mathematics has been in development, quietly, for over a decade now ... I should know, because I'm one of the team developing it. ;-)
The short form is that AI deals with the notion of a mind's representation of things, and representations of real things must reflect the nature of those real things ... and the nature of things is structural, rather than numeric. Thus, I believe (and I'm speaking for myself, not my group!) that only a framework that offers structural representation as a basic feature (i.e. not tacked on) is appropriate for the development of AI in particular, and of a theory of mind, in general.
If you're interested, Google for: ets5 unb. Don't let the size of the paper scare you, it's not that bad if you take your time....
And yes, I'm a big Google fan. =D
Thought you all might appreciate this article that I snagged from the excellent 'minding the planet' blog, since it sounds pretty similar to what Larry is talking about, but would be open-source..
http://cthings.crispynews.com/article/show/4767
To respond to the rather long diatribe about why AI won't happen:
I'll begin with your ending, and move onto AI a bit later.
"AI belongs to the same category of mythtech as flying cars, videophones, food pills, personal jet packs, and nuclear power too cheap to meter. Never gonna happen."
Flying cars aren't going to happen not because they are technologically infeasible, but because they are a dumb idea. We have a hard enough time maneuvering a vehicle in 2-dimensions: planes are safer because you don't deal with unqualified drivers. This advantage would vanish faster than a cupcake on Oprah if you have every moron (drunk, old or distracted) flying about. Additonally, where in the hell do you need to go that requires flight? We don't have lighter-than-air building materials yet, so everything is anchored to the ground, and we usually include stairs. This is cute rhetoric, but a crap argument.
Videophones have, for all intents and purposes, arrived. We generally call them webcams. People don't use them primarily because they, like flying cars, are something of a crap idea. The chunk of meat you inhabit is generally completely inconsequential, it's your ideas/information that matter. That requires speech or (better) text. That these are less bandwidth-grabby is just an added bonus. Again, no technical limitation, just a dumb idea.
Food pills: multi-vitamins, an enzyme solution and a bag of sugar will keep you alive and in decent health pretty much indefinitely. People don't do this because eating is fun, and IVs are not. Taking a pill is much less satisfying/social than actually eating, so we continue to do as our ancestors. Again, dumb idea.
Personal jetpacks: hydrogen peroxide does an amazingly good job as a jetpack fuel, we don't use jetpacks because, once again, they are a dumb idea. Personal injury lawyers would have a field day with them, otherwise, they are nigh-useless for practical activity. If they're fast, you need protective gear, which bulks it up, if they're slow, you might as well run or walk instead.
Nuclear power too cheap to meter: the idea that energy would ever be free is kinda dumb, admittedly. It is, after all, a fixed resource, as far as we can tell. We aren't using much of it, in universal terms. That said, nuclear power does have the potential to be ludicrously cheap, it is storage/regulations that make it expensive.
AI has an advantage that none of the others does, though: it's software. The marvelous thing about software is that working on it is, for all intents and purposes, free. More importantly, there are an awful lot of really smart people that are willing to work on software, for free. (at least initially--it may later become a product) Additionally, it is much more easily distributed. If you're designing a car, and you want my help, you send me the blueprints, and maybe some concept drawings or whatever. That's great. It is not anywhere near as pure as source code, or a mathematical function. Those are just ideas with a little bit of language. If it's in some variant of LISP, as most AI-attempts are, it's about as pure a rendering as an idea can get. (very little syntax/conventions to get in the way of expression)
I suspect that AI will happen. We're getting pretty good text-parsers, and better text generators (I'm willing to bet my kidneys that a LISP program can write better poetry than you can). People think that it's cool, so they'll hack on it for the foreseeable future. What we do with it is the real question.
Yes, it'll happen, yes, it'll probably be Google, no it probably does not already exist, yes it'll probably exist within 20 years.
I think A.I systems will be launched soon, but no system is perfect.
I mean it will not be as intelligent as google pictures it.
it is my understanding that Sydney GooglePlex is looking for LISP experts !!
huh ?? it maybe ancient lingo --but what the heck, LISP was built in the MIT labs for AI ...correct ??
AI will and may very well already exist. But will we know of it. I don't think so, unless if it is stupid and shows itself.
Their is no problem in the creation of an AI we can't solve. The cheer power and storage needed for it are just a material thing fixed with enough money. And the problem of making reality understandable is blollocks. Do we understand reality? Ofcours we don't. All we can do is crunch numbers and perceptions. If anyone can give me an idea anyone has ever had that was not learned or based upon something in nature I'll eat my shoe. In the end humans have developped some realy nice things but nothing new. We can't make something new because it is impossible fysically.
So an AI that matches humans doesn't have to be creative. All it should be able to do is apply patterns and rythms and ofcours make new ones out of the results of the existing.
It doesn't have to be able to see the relations between different thing but it should try out every thought and equation onto everything it knows. It will create alot of nonsense but it will learn to discard this. The result won't be a perfect intelligence. It might even turn out to be blond but it will be an AI for it will be doing exactly what we do.
But please contradict me for this page filled with crap was probably totaly wrong and i'd like to learn more about this topic.
AI already exists - it is just under appreciated. As for Larry's Google AI, progress is already being made.
W3C is working on the semantic web which has the ultimate goal of giving meaning to information on the basis of ai solutions.
This will end the Information Age. It will start the Knowledge Age where computers rather than exchanging bits of data will exchange actual knowledge!
To make A.I give it a body and the senses.
Post a Comment
<< Home