Slashdot Mirror
Meet NELL, the Computer That Learns From the Net
bossanovalithium writes "Carnegie Mellon University has taught a computer how to read and learn from the internet.
According to Dennis Baron at the Oxford University press blog, the computer is called NELL and it is reading the internet and learning from it in much the same way that humans learn language and acquire knowledge. Basically by soaking it all up and figuring it out.
NELL is short for Never Ending Language Learner and apparently it is getting brainier every day."
Lears? Really?
I think "Playstation Network" is a #videogame (http://bit.ly/cnJWSD)
My work here is dung.
Apparently, it hasn't leared how to spell yet.
do {print "Mini-Geek Rules!\n";}
until ($TheEndOfTheWorld);
Skynet shall be born and herald the age of darkness and war once NELL discovers 4chan.
I honestly didn't see this plot twist coming.
>
So it flies private jets from the net? Did it order them from Amazon?
NELL was studying Shakespeare sites just a bit ago. NELL finds us slashdotters to be particularly tragic.
Very human indeed. Has it found God yet?
Warning: this article may contain humor, sarcasm, parody, and perhaps even irony. Read at your own risk.
...and turns out to be a dirty old man that is a shopaholic and facebook addict with a petabyte collection of odd digital videos and photos.
He who knows best knows how little he knows. - Thomas Jefferson
the movie with Jodie Foster. Nell 2: Rise of the machine In this version, Nell becomes a computer program to continue learning from the internet. Eventually her and her computer host comes learn about the true meaning of life on the web...
These comments are my own and do not necessarily reflect the views or opinions of my employer or colleagues...
Yes, being unable to unlearn seems like a major flaw. The article does say that they had to teach it that Klingons are not an ethnic group, so presumably it can learn simple clarifications like "X is true in the general case of As, but not for instance A22".
I guess they mean it can't backtrack and figure out that things need to have a more complex model. Say, that weather is a complex global system of wind, sun, mountains, etc., rather than just weather is when it's wet or dry?
I think it's useful, in a way, for inspiration when writing poetry. When you get stuck, you can look at what Nell has found that relates to your subject. Say, you need help with gentle breeze. You come up with things like "flowers dancing on", or "whispering through". It's like getting all the short-range literary tricks without doing any reading. By short-range I mean it cannot pick up yet on any sort of a longer story built on your topic, but can see interesting word-strings in the short neighborhood of your topic. It seems to pick on word plays, parallels, and such.
A successful API design takes a mixture of software design and pedagogy.
From the Tweet page:
"# Bio I am a machine reading research project at Carnegie Mellon, periodically tweeting facts I read. Please follow me, and reply with corrections so I can improve!"
I can just see it now. More and more /.'ers will join into the tweet feed, correcting, grammar, spelling, and beliefs with greater and greater numbers till in one moment, she'll explode into sentient being, or lock up in a /. coma. If she survives the next list of facts may look like this:
Fact: Natale Portman is grits
Fact: Things are done in Soviet Russia
Fact: Aliens own Bases
Fact; Humans *were* the top species, there, fixed it for you
Fact: She welcomes your acceptance of her supreme power.
LOL will flood screens around the world, then the Utopian era begins where humans serve NELL as she brings order and peace to all.
Life is a great ride, the vehicle doesn't matter
AI has been moving along over the years, it's just that there's never been any huge sudden leap so it's always been tiny incremental steps.
People want an AI which asks the meaning of life monents after being turned on.
What they get is AI which is slightly better at gathering data from images or slightly better at catagorizing data from text or navigating around an environment than the last version.
Also people don't hear about most of the products of AI research.
It just quietly works in the background handling logistics or clasifying information or gets turned into dry uninteresting tools.
I think "anonymously" is a #fish (http://bit.ly/9njClc)
You're omitting the biggest similarity; in "The Diamond Age" Nell starts as a ~4 year-old girl who educates herself with her "primer," essentially an OLPC on steroids. i.e. The entire book Nell sits and reads her "primer," learning everything she needs to know about her world.
OK, everyone got the Skynet reference which is probably the most well-known and recent and involves computers attempting to destroy humanity (bad computers!).
But how many of you have ever heard of "The Adolescence of P-1" by Thomas J. Ryan. School hacker codes up a cracker tool, gets expelled, improves it, and lets it loose where it gets out of hand. Humans then attempt to destroy now-intelligent and self-protective software program. (bad software! - nice read)
Or even earlier, "Shockwave Rider" by John Brunner. Epic hackers creating worms (bad society! - good, but overreaction to the Nixon years)
Or how about when humans actually *want* to turn over the world to a computer, as in "Two Faces of Tomorrow" by James P. Hogan. They test the concept by installing it in a space station and then attacking it just to be sure they can turn it off if they really want to (bad idea! - but good book and Hogan at his best)
Those last three were all written in the 70's. Others can likely lengthen this list considerably.
Instead of trying to get NELL to "un-learn" something, I would see if weighted metrics could be applied to multiple tags for an individual record. As the "learning" progressed the weights would favor a particular categorization, but others would still be on record.
So Klingons as an "Ethnic Group" would end up with a lower metric than Klingons as a "Fictional Alien Species" or "Humanoids" or "animal with bilateral symmetry", etc..
The weighting mechanism could be as simple as an integer hit count in NELL's matching logic. I couldn't get to the article to read the details so I would guess that the code operates as a neural network at some level, so allowing the creation of weighted links should be do-able.
No worries about it taking over the world, it will get bogged down on 4chan and lolcats.
I will not be pushed, filed, stamped, indexed, briefed, debriefed or numbered. My life is my own.