Cognition Technologies' Semantic Map paves the way for the robot uprising

Cognition Technologies' new Semantic Map lets computers -- and, conceivably, evil robots -- "understand" the English language in much the same way humans do, based on word tenses and context in a sentence. With this technology, a computer or search engine can understand virtually every word in the English language -- for a vocabulary about ten times that of a typical American college graduate. The system is already being employed in search engines, allowing people to ask questions in human-phrasing instead of unnatural, machine formatted word strings. Researchers say the ability to understand language is an important building block of the nascent Semantic Web, and will make the Replicants of the future extremely difficult to detect.

[Surgex @ Sep 20th 2008 12:30AM]

That's all well and good, but can it understand Jive?

[Chris Anderson @ Sep 20th 2008 1:39AM]

I say hey sky, s'other say I won say I pray to J I get the same ol' same ol.....you dig?

[BigD145 @ Sep 20th 2008 1:55PM]

Can it understand teh leet hax0r?

[Artie Lange @ Sep 20th 2008 12:43AM]

Ruh-roh. Someone get Deckard out of retirement and make sure he gets the latest version of the Voigt-Kampff empathy test.

[Strange Quark Star @ Sep 20th 2008 12:04PM]

He was quit before this article was posted. He's twice as quit now.

[Alex Keen @ Sep 20th 2008 12:43AM]

"Quite an experience to live in fear, isn't it?" Bladerunner

[Wolfticket @ Sep 20th 2008 11:49AM]

'I've seen things...'

[Dillon @ Sep 20th 2008 12:49AM]

Wait - so it is able to just figure out a word, which it doesn't even know, entirely from the context? Wow, I must say, that is quite epic.

[Will @ Sep 20th 2008 12:58AM]

No, it's really quite a 'dumb' system. They've "tought" it about words (it looks similar to the free WordNet), and combined it with a Word Sense Disamiguation system (WSD is a field of NLP research). So it can't understand words it's never seen before, nor can it be applied to any language other than English. Most NLP researchers moved on from this style of system a very long time ago because it's not scalable to other languages or unseen information. Most state of the art NLP nowadays is based on statistics, making it easier to apply to other languages.

[jeremy @ Sep 20th 2008 1:59PM]

Will ,
So how do you think this tech compares with Hakia's? J

[Skinjob @ Sep 21st 2008 3:16AM]

I believe Will is incorrect.

It's similar to WordNet, but MUCH larger. Compare the stats on the two systems (both have published their data on their sites). It's also not combined with a simple WSD. It's combined with a lot more than that. Read the white papers on the site and you'll see that. As to languages other than English: the structure and architecture of the map is completely transferable to other languages, but obviously the words and their meanings need to be "curated."

If "most" NLP researchers moved on from this then it's because it's difficult. Cognition's site says it took their linguists 24 years to build this map. When you're hand-coding 10,000,000 semantic connections, it's going to take some time. NLP based on statistics is less accurate and doesn't understand words, as it is based on patterns.

[Jonathan Allen @ Sep 20th 2008 12:50AM]

It is the distant future, the year 2000.

[ADAM @ Sep 20th 2008 1:04AM]

there are no more elephants!

[John @ Sep 20th 2008 12:55AM]

so the average college grad knows 10% of the words in the English language? I'm not sure if that's good or not...

[Reader @ Sep 20th 2008 2:05AM]

Eh, play freerice.com and I think you'll be amazed at some of the random ass words that come up. "umiak = Inuit skin boat"

[gonintendo @ Sep 20th 2008 9:34AM]

yeah, the English language is monstrous, and freerice.com is a prime example of this.

[Duke @ Sep 20th 2008 11:17PM]

I think umiak can be spelled with a Q as well, if memory serves. Scrabble is a decent way to learn a lot of random ass words.

[Sisyphus @ Sep 20th 2008 1:49AM]

This is sort of misleading.

The technology is in no way creative, merely reactive to queries. The significant part of understanding language is participating in it in a creative and responsive way (that's a big paraphrasing of whats current in philosophy of mind / phenomenology, even robotics). Oh -- thanks for putting "understanding" in quotes though -- just noticed.

Understanding requires involvement (the embodied kind). I'm guessing the best that this system could do is analyze patterns in the way people use words on the internet and "evolve" over time, otherwise it won't be much use in 10 years given human adolescence...

...idk, my bff jill?

[jooster @ Sep 20th 2008 2:56AM]

Spot on Sisyphus, we rats, we rabbits, I mean humans, have still lots of ways to differentiate from other life or machine forms that makes us feel more competent or more powerful. That gap however in my human animal mind is getting smaller. Be this statement true or not, if this technology works as advertised it will give way to a better understanding of our species and that I find extremely valuable.
Why do we have the need to feel doomed by x?
Well, let politicians be, why not terminator?

[KarlW @ Sep 20th 2008 4:19AM]

Easy way to see if it works - any offers from Google or Microsoft?

[tulkas @ Sep 20th 2008 8:46PM]

you said it there,
but without sounding smart, cognition is patterns and nothing more.
we humans are molecular computers.
and are people born disabled lacking embodiment?
it seems that compelxity and intelligence (most of which are collective) emerge as a result of many contributing factors
accurate symantic maps of language are required for artificial general intelligence as in the learning process there needs to be the ability to check the accuracy of the original learned meanings. It is like having a dictionary and thesaurus to check from time when you get stuck.
but this is a technology that is aimed at narrow ai
making the internet computers and ads more personalised
It lends itself to live avatar news readers and videogame npc's
not to mention automated phone systems.

[GhostDoggy @ Sep 20th 2008 11:53AM]

The mistake is ok since Joseph Flatley doesn't realize that the character in the picture is not a robot, but a biologic (aka replicant). He'd been better off using the character Sonny from I, Robot, but we have to remember that a lot of those posting so-called articles are essentially Best Buy rejects.

[Andreas @ Sep 20th 2008 2:02AM]

Oh, so that's about... 100 words, right?

[Funke, Tobias Dr. @ Sep 20th 2008 5:46AM]

Wow, well it's a good thing I learned "jackass" then.

[Wwhat @ Sep 20th 2008 7:46AM]

Picked it up from MTV did you?

[EddieStarr @ Sep 20th 2008 3:19AM]

I don't like the paranioa tone this article is taking. Technology has to eventually advance to the point where we will be able to communicate fluently with hardware (and software) It will just be apart of our technological evolution, I am never going to be under the impression that robots would in any form retaliate against humanity, or alteast not in my lifetime. Though, I doubt robots would ever damage earth nearly as much as humanity will, (and already has)

[Dave Chappelle @ Sep 20th 2008 4:16AM]

wooo!! 2019 here we come!

OH!!!! NO1 HAS SAID IT!!!
Deckard is a Replicant!!!!!!!!!!!!!!!.... alll agree????

[Andreas @ Sep 20th 2008 4:30AM]

I disagree.

The whole point with the movie is that Batty is an extremely human robot and Deckard is a an extremely robotic human. - "More human than human"

[Dave Chappelle @ Sep 20th 2008 7:11AM]

ye but what about the origami unicorn at the end? that's the teller. Regardless of the 'what it means to be human' aspect. he is robot underneath, it is a bit confusing actually i think that was the case in book 'i dream of electric sheep' or w/e and i am sure Ridley Scott got into a dispute with Harrison Ford about this too.

sorry, i am being side tracked off the topic lol. i did not even enjoy this movie the first time i watched it @ school. (then i went online and all the comments were like u MUST watch it a second time so i may/may not just been thinking about it lately)

[Wwhat @ Sep 20th 2008 7:48AM]

It's not robots in that movie, it's artificial biological humanoids.

[effzehn @ Sep 20th 2008 5:08AM]

The joke called the Voigt-Kampff is pretty inapplicable here, since it's only software. I'd rather go with the classic Turing Test. Or some derivate of it. To create an "understanding" machine (i.e. a machine which creates the impression that it actually understands _something_) you need (amongst other things) contextual knowledge and a model of the world, which means it has an "idea" about how we live (latter) and "followed the conversation" (former). It is an extremely complex task for sure, but I think we can at least achieve it in a simple way, for now.

[LondonConsultant @ Sep 20th 2008 8:23AM]

So, is there any difference between a machine that understands something and a machine that only gives the impression it understands something?

[Frankenstein Black @ Sep 20th 2008 8:34AM]

There he is, ROY ("I want more life F'er") BATTY. Best flick of all time!
http://www.imdb.com/title/tt0083658/
Lets hope they don't f-up a remake...

[Frankenstein Black @ Sep 20th 2008 8:41AM]

Oh, for the n00bs and gameheads that don't from Replicant Roy, Bryan Fury from Tekken is his doppelganger
http://www.jun-shrine.com/characters/bryan/bryan.html

[bw @ Sep 20th 2008 9:43AM]

Understanding something is just making meaningful connections between several types of information, so machines do understand in their own "limited" way, for now...but the potential for them to understand certain things in a much more profound way than we do is enormous.

Keep the faith :)

[tatarynowicz @ Sep 20th 2008 9:54AM]

Finally! "mp3 players" is dead! Long live the "Hello, can you please tell me what are the most popular mp3 players at the moment?"!

[tasteslikechicken @ Sep 20th 2008 10:57AM]

my current model of the brain is that it doesn't work as a single unit. There are a bunch of subsystems (like the video card and sound card and possibly the physics card in your computer). This might be analogous to the sound card, the work that novamente or numenta has been doing lately would be the cerebral cortex, cognex or vision-systems work for the visual cortex. I'm unaware of a limbic system but it might not be necessary for an AI. I'm actually surprised to see how fast the pieces are coming together. It seems like we might have a couple AI's (and I think we'll need two, so that they can learn from one another and not just from our weirdly different perspective) sooner (4-8 years) rather than later 20+ years. So what do we humans do afterword? We moved from hunter-gathering societies to agricultural societies then to industrial societies, and now we're moving to knowledge/service societies. As far as an AI or robot uprising it seems a little funny. Human uprisings have occurred because of a lack of resources (and advanced from there by classism, racism, religious-isms, etc to justify the war beyond that point). A robot or AI will be smart enough to make unlimited money (Warren Buffet would rapidly look stupid to an AI aimed at making money in the stock market), have unlimited time to live, and it will realize that reasonably quickly as well. So, what do we do when we have (from our current perspective) unlimited resources. Robot workers plus AI's will finish a space elevator quickly, figure out inertial confinement fusion, biofuels, cheap solar, thorium nuclear reactors, etc in quick order. They will take up no room (they might exist within server farms moving around as they wish, with a million eyes in the form of every camera on every computer, a million ears in every microphone on a conected computer, they could be made to 'feel' data, I know people that (tell me that they) can 'feel' misspelled words, what happens when you can 'feel' all the scientific theories that exist in all fields simultaneously?

[Coby Tamayo @ Sep 20th 2008 1:55PM]

AI is still having difficulty dealing with the fact that computers and brains work in fundamentally different ways. I suppose if your sofware worked at a certain level of complexity, you could neglect the fact that brain hardware and computer hardware don't do the same thing, but that level of complexity has yet to be approached in any significant way. Brains are what Stephen Pinker called "blank slates." A brain comes into the world with little to no initial knowledge, but with a malleable structure that reorganizes itself based on patterns it recognizes within its experience. When people hear words, they don't just draw on some static dictionary that's been programmed into their heads; they're actually drawing on memories of how they first heard the word used, and subsequent instances of hearing it. AND, the new use of the word reshapes the experience-based definition that's already there, in only slightly. For a deeper insight into the matter, read Jeff Hawkins' book On Intelligence.

[charles @ Sep 20th 2008 6:40PM]

lol the way you people promote robots its quite clear your transhumanists. "i can make a better human" "next stage of human evolution" blah blah blah. I look forward to the HUMAN uprising against unethical scientists who think the world revovles around them. Do they honestly believe that "godlike beings" will treat normal humans as equals. what a joke. Scientists talking about beings that have freewill to do whatever they want yet somehow they know they won't hurt us. Nice leap of faith at our expense your taking.

[Paulmichael @ Sep 20th 2008 11:18PM]

I don't like asking search engines questions, I'm too used to using keywords by now. If I could talk to my computer in a way that was like talking to a person and then it would search for me, then I would like it.

[Str1ker @ Sep 21st 2008 2:26AM]

For some reason I do not believe this.

[ryanrule @ Sep 21st 2008 8:34PM]

can it understand lil jacob from gta4?

[ginandphotonic @ Sep 22nd 2008 5:42AM]

Sweet, now I can finally input commands into my computer just as futuristically as in Tron.

[Navneet @ Sep 22nd 2008 7:35AM]

wondering how come no one has acquired cognition yet?


Navneet
Do you blog? Visit www.contentxn.com