Slashdot Mirror
Artificial Inteligence Common Sense Database
warren69 writes "Atari researcher/Stanford Prof. develops AI called Cyc, pronouced psych, based on "1.4 million truths and generalities". Allready this, umm application (linux fyi), has powered lycos search narrowing.
There is encouraging results, like Cyc asking if it is human."
Don't give it control of a manned space mission... "Open the pod bay doors, Cyc..."
(anonymous karma whoring -- whoo hoo)
Cycorp web site
OpenCyc
Sourceforge project
From the article:
Cyc's programmers taught it that certain things in the world are salacious and shouldn't be mentioned in everyday applications.
What do you think about imposing our morality on an AI? Is it neccesary for any artificial intelligence we create to share _all_ our values?
If there is no afterlife for an AI and no punishment, what motivation does it have to be good?
yes i run a goth/punk/emo porn site.
This is just great. Pronunciation keys using silent P's.
Canadian Cynic, canadian politics is less boring than you
The military, which has invested $25 million in Cyc, is testing it as an intelligence tool in the war against terrorism.
I seriously hope they aren't going to allow George W. Bush to input any intelligence into this thing.
Hmmm. I'm curious to ask Cyc if Linux is better than MS Windows, if free software is better than proprietary, if sharing music is stealing, and so forth. "Common sense" -- especially when collected in a database like this -- can't help but showing the biases of its creators. If this tool becomes as important as the linked-to article implies it will, let's hope it has common sense that fits with our agenda....
Why is Cyc asking if it is Human any more significant than Cyc asking if it is Lettuce, or asking if a football is a gourd?
Its artificial self-awareness may be prejudiced by the programmers to imitate self-awareness, or in this case merely be a surprising juxtaposition of semantics amid otherwise ordinary pairings, rather than implementing self-awareness.
In other words, it may now know that Cyc is not human, but it likely has no idea that it is Cyc.
--Blair
Yet another webzine discovers Cyc, and yet another crop of slashdotters hasn't heard of it... If you read the article, the damned thing asked if it was human in 1986. This is news?
I have been following this thing for at least 5 years, and they have continually been just a few years away from real world applications. One of the things they have been talking about for a long while was Cyc approaching the ability to "read" for itself, and gather new information for it's database from the web, newspapers, or any other authoritative source. They've been talking about it for a long time and it hasn't happened yet.
It is a very interesting application, but will probably never amount to anything near human intelligence - a very versatile expert system at best.
-josh
Not exactly as exciting as it sounds.
Basically, Cyc finds questionable conclusions following backwards reasoning, then asks humans for confirmation. A decent strategy, when you consider that the structure of common human knowledge is built to work for people with less than perfect logic.
The exchange went something like:
Datum: Humans are intelligent.
Datum: Cyc is intelligent.
Query: Cyc is a human?
Not in natural language, though, but its custom data language.
That, to me, is the biggest weakness of the system. IMHO, tying the data to a natural language, or to the real world in any other way, will take as much work as building up the knowledge directly tied to a natural language. This elaborate, detached structure is basically wasted effort, castles in the clouds, which is why they've had such a hard time applying it to the real world.
--Cycon
Your Brain + EEG + LEGO Robots = Brainstorms
I think there are, in general, probably two ways we could hope to achieve "artificial intelligence" (whatever the heck that is): First, by some form of duplication of what's already there. For example, by digitizing an entire working animal/human brain. This would not require us to understand the workings of the greater structure of the brain, just the little parts that make it work. The second is by figuring out what sort of simple, fundamental bits are necessary to create a digital "brain" capable of learning and improving in a way that would enable it to eventually become "intelligent" (again, we would have no understanding of the final "intelligent" structure, only the methods that created them). I think Genetic Programming, while somewhat interesting and possibly even useful, is not the key. It has the same concept in mind though, I believe.
But what do I know. Clearly not enough to dupe enough investors to pay for my silly musings.
The streets shall flow with the blood of the Guberminky.
Its a purely dumb expert system. it has no self reasoning capability -- it draws inferences from already preprogrammed facts. it cant learn without someone stuffing it and it definitely has no curiosity drive to allow it to grow exponentially smarter.
Youre not teaching it about morality -- it doesnt learn. its dumb. youre just adding new constraints to filter through.
Personally i think this is a hare brained idea. the 60 mil would be better spent on developing a huge set of different neural network algorithms and finding one that enabled expoenential growth.
You can look at AI in two ways (or a combination of both, of course):
- AI needs to have its capabilities defined and data manually entered in, so that it can do what an AI needs to do
- AI needs to be able to learn, so that it can learn what an AI needs to do. A smart AI that 'knows' nothing is just a big paperweight.
Roughly, at any rate.
Both ideas have merits. Babies, for example, learn by association, and by occasionally trying stuff out and making assertions based on observations. However, they also come equipped with the hardware (wetware) capable of handling this.
I think that getting both parts right will be useful, so yes, it is (or might be) a big deal.
Lastly, what do you want to use the computation force for? Write down the equations and calculations now that will yield a successful AI, if it's that damned easy. You can't, because designing it is more difficult than throwing expensive hardware at it.
--
Try translating 'Mensa' from Spanish to English.
Why is it that the last two stories have both come right off of fark.com?
The job ended because of turnover at Lycos after it was bought by Terra Networks. Cyc showed promise and could be brought back, though it can't improve search engines all by itself, said Tom Wilde, Terra Lycos' general manager of search services.
Lenat has been announcing that Cyc will become "intelligent" Real Soon Now about every two years for the last decade. Nobody believes him any more.
Someday that database may be useful, but not with a predicate-based world model. I regard Cyc as the ultimate answer to "Will rule-based expert systems ever become intelligent". The answer is "no".
Cyc is a cool project - one that I've been reading about for 10 years now. But I don't think it is AI or ever will be. It basically collects a huge number of rules and has a deductive engine that helps it infer new facts based on what it knows. If you think that's all the human mind does, then you might want to read some books by Douglas Hofstadter. Amazing stuff.
Intellegence is about finding the differences between things that are the same, sameness between things that are different, and adapting to new situations fluidly. All of these are impossible with large collections of rules.
I believe that machines may think someday, but it won't come from projects like Cyc - it'll be more from the neural network approach.
A lot of the comments I've read so far are missing something. Yes, it is just a giant fact-base in an expert system. And yes, that will exhibit human-esque "reasoning". And yes, a good argument can be made that this isn't "true" intelligence, and it won't develop true sentience
Imagine the military and educational benefits of such a system. The US military is getting their money's worth, and they know it. Imagine Cyc, with its full fact-base, on a device carried by every soldier. "Cyc, how do I fix this problem on an Apache helicopter?" "Cyc, where is the fuel tank on this specific enemy vehicle?" Can you imagine being an inquisitive child and having one of these things at your disposal? "Cyc, how does this work?" "Cyc what is fourier analysis?"
This sort of system is a really good system for organizing and relating statements and presenting them in such a way extraneous unrelated results can be easily eliminated, and related results can be located quickly. It it can be made to derive statements for its fact-base by reading anything available, then it would become almost like an Oracle of Knowledge. Eventually, with some years of refinement, it may be possible to ask the engine difficult theoretical questions, ("How can we improve on the strength of carbon nanotubes?") to which it would respond with an experimental procedure (as the answer is not immediately clear) to discover more facts toward the solution to the problem...
When you consider this, it doesn't really matter if it has "true" intelligence or not. We don't have to argue the finer points on reasoning, intelligence, etc. No matter what, it will be a system the human intelligence can use to extend its own reasoning, and with that, I think, we will be able to make great bounds forth in education and scientific discoveries because we will be able to relate such broad and deep pools of knowledge.
Wendell
I don't remember the precise expression, but in its language, it was much closer to:
Datum: Members of the class of humans are intelligent.
Datum: Individual entity Cyc is intelligent.
Query: Individual entity Cyc is member of the class of humans?
It's not a direct logical conclusion, but it's a question worth asking, which is what the programmers were shooting for.
Don't get me wrong, I think Cyc was a good academic exercise, a worthy experiment, and it will pay off for the field in the long term. I don't think the project is generating a practical system, though. Some investors are getting royally screwed, and it's being taken to an insane stage of development.
MULE . o O (The carrot's only a yard in front of me, so that means it's only two or three steps away!)
In other news Noah and his pets survived the Great Flood in an Ark.
This comment is printed on 100% recycled electrons.
Consulting The Jargon File entries for
bogosity and micro-Lenat,
we see that the uLenat is the everyday unit of bogosity,
and that it is named for Doug Lenat, whose project Cyc is.
I tend to agree with Reid, myself.
ob book: For a literary treatment of a connectionist machine
that may or may not resemble Cyc,
see Richard Powers _Galatea_2.2_
Wait a minute. Didn't I say that on the other side of the record? I'd better check
There is a lot more to knowledge than the classification of namable objects and their relationships. There is a huge amount of knowledge that cannot be formalized with symbols. For examples, playing soccer or football, recognizing a subtle fragrance, face or musical tune, manual dexterity, finding one's way around an unfamiliar neighborhood, etc..., in other words, the sort of common sense knowledge that can only be acquired through direct sensory experience.
The interconnectedness of human cognition is so astronomically complex as to be intractable to formal approaches. This realization immediately makes the use of symbolic knowledge representation approaches to creating human-like common sense in a machine look rather silly. That 25 million dollars of taxpayers money went into this Cyc thing is a testament to the effectiveness of the propaganda machine of the GOFAI community. Bravo!
'Cyc' means 'tit' in Polish. For that matter, CIPA ( which stands for the Children's Internet Protection Act, I think ) means 'cunt'. It's probably a good idea to make sure your project name passes the laugh test with the major language families before you pour millions into it.
This was a lesson bitterly learned by the Warsaw weekly 'FART' back in the early 90's. Fart means stroke of luck in that language, but their luck ran out pretty fast.
Not to mention the marketing team behind the Chevy Nova ['won't go'], Latin American division.
A customer service representative will be with me shortly.
The A.I. software mania of the mid-1980s was a preview of the late-1990s InterNet mania. Droves of computer science professors quit to start new A.I. companies. Exaggerated claims were made about the power of A.I. software. There were cries of "losing the computer race" with Japan. Japan has the Fifth Generation Project: A.I. parallel computer hardwired with Prolog- but it fizzled out too.
Although little practical progress was made in A.I., there was some decent spinoffs. The first workstations and first personal graphics computers were from A.I. efforts at Xerox, TI, Symbolics and others. Soon after Apollo, HP, and Sun followed with more generalized workstations using this technology. And then Apple MacIntosh and the Thieves of Redmond.
Richard Stallman was left unmolested in the empties out MIT AI lab to develop his GNU tool family.
Cyc was part of the US government-industry A.I. research institute in Austin. Then it became privatized into its corporation hobbling along on governemnt and private funding.
Its a logical mistake to think that that was a logical mistake. Don't confuse a question with a conclusion. Using your example, it would be wrong to conclude that John is Peter. However, it was not a conclusion but a question, and a valid one at that. You may not believe that Cyc is intelligent, but to claim it is using poor logic in this example just shows your lack of same.
Move on. There's nothing to see here.
For instance, they deployed the technology to an image library owned by a news company. The company had lots of images, all with different captions. The thing was, there was no fixed system for the captions, they were just english descriptions (short) of what was in the photo.
So Cyc analysed all the captions, and turned them into CycL (it's own logic language). It then used its rudimentary natural langauge capabilities to figure out equivalents, so if you asked for "frightened child" it would match to "girl with gun held to her head" even though they contained no equivalent words. Pretty clever stuff, though they're a long way from being able to make it formulate sentances itself.