Slashdot Mirror

MindPixel seems to have much in common with Lenat's CYC (which is much further along). See here.

Projects like CYC invest a great deal of money, time, and effort into creating these patterns, these networks of information. Even so, you just cannot pay enough people enough money to do this.

Google, on the other hand, has a million people a day supplying them with precisely this information. If they watch what you search for, and then what pages you end up selecting; or how you refine your search to get to what you really are looking for -- then they can truly build patterns of knowledge for free -- it's likely that they'll find a way for people to pay for the priveledge, even.

I don't actually expect that they are archiving their search clickstreams that way and for that purpose, but if they were it might explain why they are not so hung up on this profitibility; and why they've worked so hard on making the system so damn fast. The more people use the system, the quicker they can tune and refine their searches, and the amount of times that the search finds the right page; the more useful information is available.

Think of it! Millions of people volunteering to increase the potential intelligence of your AI every day! For free!

Of course, I don't want to be around when the AI created at the end of this process finally understands the connection between Natalie Portman and hot grits -- it's likely to turn it's back on humanity forever at that point.

thad

In other words, AI was completely based on explicit semantics, without ontology. It sort of flopped.

Agreed. I went through Stanford CS in the mid-80s, when the "knowledge engineering" crowd still had credibility. I was convinced those guys were full of shit, and, in retrospect, they were.

Even ontology doesn't help much. Doug Lenat is still slaving away on Cyc, which is supposed to do "common sense" by referencing a big canned ontology of statements about the world. Cyc has been a year away from a major breakthrough for the last decade or so, from Lenat's press releases. Cyc now has a product that helps index documents, but it's not clear if it's much better than Ask Jeeves, or even "grep".

Sorry for not sticking to telescopes, but I couldn't help it. You must be aware of projects like CYC and everything. My question is, does the world really need GAC? How is GAC different, besides the commercial aspects? --willy dog

Also, DirectHit is what's powering the beta version of HotBot - which incidently uses the Cyc Engine & KnowledgeBase.

The Cyc project seems very cool - but I've not even seen any of their stuff running yet. (Despite attempting to implement an inference engine myself around the core of their KB!)

Whilst we're on that subject, the core of their KB is available for free download & usage - even in commercial apps - providing you give them the necessary credit in the app (etc). Check out their website for full detalia...

HTH,
fRoGG

That's the problem, all right, but I don't see anyone in the AI world except Cycorp really wrestling with it, and they've mostly given up on human psychology, which is where all the useful web-semantics lie.

For example, suppose you want to find a Drew Barrymore fanpage that discusses how she stays so upbeat all the time-- how do you represent that with a finite set of XML labels?

Nobody knows, and solving it will take a sort of subtle psychological analysis that's the real, gaping abyss in current AI curricula.

The GOFAI idea that human cognition is a matter of disembodied symbol processing is dead, and good riddance.

I would add to this only a few of my own personal questions from having tried to get into AI:

• How can one approach AI from a non-academic point of view? I am not in college--I have a full-time job... How can I contribute? Is there shareware (or commercial software) is out there to help study AI?
• My experience, getting into the nitty-gritty of AI, is that it's laiden with mathematics--comeplex math, at that. Is there an approach to AI that isn't founded in mathematics? What can a non-mathematician contribute to the study of AI? Or do I have to bite the bullet?
• Friends of mine in CompSci (graduates) have all said "ALife is dead". They encourage me not to persue it, "if I want to work". Do you think the demise of ALife has adversely affected AI research? (For example, Santa Fe's ALife site is stale.)

I would also love to hear your comments on Cyc. Do you think they are going to 'win' the AI race?

Let me just add that I think the creation of AI is perhaps the most critical and meaningful pursuits available to mankind, as a whole. The universe exhibits an amazing capacity to create life--to share in that power as a race is about as ultimate a purpose as I can think of for humanity.

Best of luck to you.

Au contraire Mon capitan! (pardon me =)
Pattern recognition has already reached a significant level of compexity but, what is not public at this moment is an integrated personality. A Machine Intelligence might even exist today, if it does you definately don't know about it and neither do I. To know of such a thing would very logically be a death warrent or at least permanent house arrest.
I think therefore I am. The prefrontal cortex is one of Gödel's islands of consistancy, reverse-engineering of that structure is well under way on many fronts: here, here, here and too many other places to mention. Gaming AI doesn't have a trillionish dollar distributed budget behind it simply because games don't generate that kind of revenue. Besides this hardware is woefully inadequate, a few very fast processor versus my billions of slow ones. I simply have more chances to stumble across something.
Hmm. So a compressed dictionary is the key to creating a true intelligence? Well! Step right over to those fine folks at Cyc who have been doing just that! To bad the darn thing is a lot more brittle than you or I. Although I really like the semantics they're developing - someday it could make good baby food for the real thing. I've spent many sleepless nights researching this field and the only thing I've learned is that there are a whole lot of distractions. The proof of that lies in the fact that HAL didn't come online on schedule.

The Night Angel

Only the fool would take trouble to verify that his sentence was composed of ten a's, three b's, four c's, four d's, forty-six e's, sixteen f's, four g's, thirteen h's, fifteen i's, two k's, nine l's, four m's, twenty-five n's, twenty-four o's, five p's, sixteen r's, forty-one s's, thirty-seven t's, ten u's, eight v's, eight w's, four x's, eleven y's, twenty-seven commas, twenty-three apostrophes, seven hyphens, and, last but not least, a single !

I'm suprised nobody has licensed the Cyc software/ontology for use in web indexing. Actually I could be out of date and someone might have already!

The key to good indexing and search lies in scanning for knowledge and not "words". Unfortunetly more and more webpages are designed to be as noisy as possible and contain little information. For example millions of webpages contain navigation menus however the "knowledge" of what can be navigated is stored as images, which is completely useless... the "knowledge" is completely lost and indexing is difficult.

There needs to be more use of meta-data in web pages if we want to index them for the knowledge they contain. Until we can index them we can't search them.

You're right, this is a big problem, and one which Cyc will hopefully solve. Don't expect *that* to be Open Source anytime soon, it requires a huge amount of tedious work to make something like Cyc, so they're pretty careful about holding onto it...
---
pb Reply or e-mail rather than vaguely moderate.

Fact 1 -- Deja News is in the Echelon building:

Deja News, Inc.
9430 Research Boulevard
Echelon II, Suite. 350
Austin, TX 78759

Fact 2 -- Cycorp makes what are arguably the best tools for scanning the web for concepts.

Fact 3 -- Cycorp was a spinoff of MCC.

Fact 4 -- Deja News, Inc., Cycorp and MCC are within walking distance of each other.

Fact 5 -- Bobby Ray Inman was the first director of the MCC.

Fact 6 -- Bobby Ray Inman is a spook's spook.

I may be a bit biased here since I was invited to go to work at the MCC when it was in its early formative stages (before Austin had been selected). My office was, at that time, at Arden Hills operations at Control Data Corporation, just two stories above about an acre of supercomputers that had signs hung on them that read "Fort Meade".

As Seymour used to say to the "insurance" agents located at the "Thorp Insurance offices" out in the middle of the corn fields near his farm where his tribe was building the Cray-1:

"Just don't let my people know you're here."

Re: the speech recog and language parsing -- have we forgotten about the cyc project?
I think it could do what they say in the news release, just fine.

it's gonna take a while to build the sensors and work out the locomotion stuff. By the time the ball is ready, the software has had a lot of time to grow up. It would be a much nicer place for erwin to live.

I think it's another reason to colonize space. ;-)

As much as everyone is fawning over Kismet, Dr. Brooks as much admits that it is a hack when he lets slip that it exploits "human" programming. Much more impressive (and operating at a 5-year-old level, not 2-year-old) is CYC, which has been being built in Austin Texas for about 10 years now by Dr. Douglas Lenat. See Cyc here. It is based on Dr. Lenat's work going back to the 80's to build a computer that can learn and reason. Dr. Lenat previously had built AM and Eurisko. All /.ers should have at least some familiarity with his work.

From http://www.cyc.com/halslegacy.html :

There is a lot of controversy about how human-level machine intelligence will develop. Some scientists believe it will follow a path similar to the one followed in nature by evolution: there will be artificial one-celled animals, artificial insects, artificial lawyers, artificial monkeys, and so on up to artificial human-level machine minds.

Sorry, it's http://www.cyc.com.