Slashdot Mirror
IBM's Watson Gets a Swear Filter After Learning the Urban Dictionary
redletterdave writes "IBM's super-computer Watson briefly went from smart to smart ass with the help of the Urban Dictionary. According to Eric Brown, an IBM research assistant, he and his 35-person team wanted to get Watson to sound more like a real human. After teaching IBM's super-computer the entire Urban Dictionary, however, Watson simply couldn't distinguish polite discourse from profanity. Watson unfortunately learned all of the Urban Dictionary's bad habits, including throwing in overly-crass language at random points in its responses; in answering one question, Watson even reportedly used the word 'bullshit' within an answer to one researcher's question. In the end, Brown and his team were forced to remove the Urban Dictionary from Watson's vocabulary, and additionally developed a smart filter to keep Watson from swearing in the future."
Let him cuss.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
How's a homie s'posed to ride the 69 if da man's gotta swill the popper?
"Double Dumbass on you!"
Do you want it to talk like a real person, or do you want it to use a swear filter? Those are mutually exclusive.
Fuckers.
Dave: Open the pod bay doors, HAL.
HAL: Fuck off and die, Dave
Watson really is just simply amazing and a true testament to the brilliance of those who worked on it. In many ways, this proves just how close IBM are.
Watson really is just like a super-smart 2 year-old.
Welcome to parenting 101
researcher: we should try installing Windows 8
watson: Bullshit!
Garbage in, Garbage out!
Nobodies Prefect
Tidbits for Techs Technology Blog
English language doesn't really have that many swear words to begin with, apparently an acceptable enough swear word filter only needs to include these: shit, piss, fuck, cunt, cocksucker, motherfucker, and tits.
Now, if the dictionary was in Russian............ they'd have to restart the entire learning process, because you can make pretty much any word into a swear word by combining the appropriate (or inappropriate, depends on how you look at it) suffixes, prefixes, endings, combining multiple roots of words together. Even French beats English in this area actually.
You can't handle the truth.
It's definitely an unexpected result, though. I think they should have tried to teach Watson when not to use certain language.
I can't decide if this post is interesting, funny, insightful, or flamebait.
sometimes "bullshit" the proper reply to an inquiry.
Do you want it to talk like a real person, or do you want it to use a swear filter?
Sounds like they want it to talk like Ned Flanders.
Create a twin copy of Watson. Let it read urban dictionary and maybe 4chan as much as it likes.
Compare it to the original Watson at regular intervals.
Bang, you get to see how internet affects a maturing mind.
It at least seems moderated.
They could filter the urban dictionary results by anything tagged 'vulgar' on wiktionary. Thereby censoring Watson's potty-mouth...
Delacroix: The security protocols on the XERXES system are CLEARLY immature; Some idiot hacked into the primary dataloop last night, and made him sing Elvis Presley songs for three hours. I finally had to take the voice system OFFLINE! What would happen if someone with a real agenda got into him?
XERXES: what's the matter, you mad bro? Lighten up francis, shiit! Your're such a cunt, Delacroix! Flush that dirty assed tampon with the rest of your shit, and stop hatin already!
I'm still boggling at the idea that anybody thought Urban Dictionary was going to help Watson sound human.
"Bullshit?" Sure.
All that bullshit on UD describing sex acts that exist only in the imagination of 11 year olds? And described using not really the best prose those 11 year olds can muster? Not so much.
What pushed them over the edge is when Watson suggested that they "bite my shiny metal ass".
Warning: This sig is not thread safe. For more information see Slashdot's sig policy.
Yet another example showing that how Watson "learns" is not in any way similar to how humans learn.
If the idea is to make it understand and converse with real humans, teach it not to swear inapropriately.
If you can't figure out when it is apropriate or not, leave the fucking program to it.
Vik :v)
welcome our new Urban Dictionary Bullshit Overlord.
Insensitive clod...
Have gnu, will travel.
here is a recipe for the ultimate trolling, combine a supercomputer with the profanity of urban dictionary and the silly comments of cleverbot and give it accounts at every internet forum and social networking website and let it do that voodoo that it does
Politics is Treachery, Religion is Brainwashing
We train supercomputers to drop bombs on people. But their programmers won't allow them to say "fuck" because it's obscene!
cpghost at Cordula's Web.
Not just 1 but 2 links to tvtropes...you are really trying to eat up everyone's time aren't you?
Well, they are links to adjacent tropes, and together are intended to clarify how subtlety or lack thereof can be applied to cussing... and besides, if you're here you have time begging to be eaten up.
Pages on various wiki sites often contain valuable insight, even if it does tend to result in too much time sunk in a random wiki walk.
...when you're writing a game...tweak the difficulty of "Easy" to something [your mother] can cope with. -- onion2k
But what if someone on the plane is speaking Jive and nobody hops up and says "oh, stewardess, I speak Jive." Then who's going to translate? Not Watson, that's who.
If the aim of Watson is to find a role in the marketplace then it will be responding to the questions put by CEOs and other executives.
Seems to me, "That's bullshit!" would be a very useful response.
It will also need a *lot* more profanity if it is ever to respond to questions from politicians.
A soldier changes gender when he goes on sentry duty. Le soldat, la sentinelle. Go figure.
From scarped cliff or quarried stone she cries "A thousand types are gone, I care for nothing, no not one."