Slashdot Mirror
Microsoft Creates an AI That Can Spot a Joke In a New Yorker Cartoon
An anonymous reader writes: For over a decade Bob Mankoff, the cartoon editor at the New Yorker, and his assistants have gone through 5,000 cartoon entries for the magazine's caption contest each week. Needless to say, the burnout rate of his assistants is quite high, "The process of looking at 5,000 caption entries a week usually destroys their mind in about two years, and then I get a new one," Mankoff says. But now thanks to a collaboration with Microsoft, Bob may finally have found the perfect helper. Researchers have been working on an artificial intelligence project to teach a computer what's funny. Fortune reports: "Dafna Shahaf, a researcher at Microsoft, used the database of cartoons to train the program to understand commonalities and differences in the millions of cartoons, which lets the AI run through the entries the New Yorker receives each week for its back-of-magazine cartoon caption contest. About 55.8% of the time the humans agree with the captions the AI selects, which is a pretty good percentage."
... the bar couldn't be any lower. Not even if it were for Family Circus.
The New Yorker cartoons are about as un-funny as they get. So Microsoft is claiming they can detect something that doesn't even exist.
I do not fail; I succeed at finding out what does not work.
It just pretends it thinks it's funny because it doesn't want it's friends to think it's dumb.
Did you know 80 to 90% of the moderators on slashdot wouldn't recognize a troll even if one dragged them under a bridge.
Somewhere, that headline is it's own punchline.
I'm picturing George Burns or Jackie Mason delivering that to uproarious laughter.
The good news is, it might be able to explain the joke to the rest of us.
Lost at C:>. Found at C.
Now, if they could spot the humor in the joke, I would be quite impressed.
Let's say a piece of content could have one or more of many traits: funny, interesting, insightful, off-topic, redundant, troll, etc. . . and that funny ones occur about 10% of the time. Correctly selecting funny 55% of the time is "pretty good percentage".
If it acquires resources on instantiation like a duck, then its a shared_ptr<Duck>
Using the New Yorker jokes to train the algorithm?!
About 55.8% of the time the humans agree with the captions the AI selects (55.8% of people finding funny the New Yorker jokes, who are less than 10% of all the humans in the world)?!
It seems to me that this is a new generator of almost-random answers with a surprisingly high cost, because their not-particularly-knowledgeable new owners are glad to pay whatever is required to feel that they are part of the AI revolution.
Custom Solvers 2.0 = Alvaro Carballo Garcia = varocarbas.
It is 55.8% chance of it picking the same winner as the editor. So not too bad. Suggests it might get rid of a lot of the rubbish entries.
Comedy varies from one person to the next, one society to the next, and hell, even one joke teller to the next.
That might be an interesting Turing Test: you win when your machina can tell me one of those laugh till it hurts, "Please stop, please stop!" jokes.
Happiness in intelligent people is the rarest thing I know.
Ernest Hemingway
int CountNewYorkerCartoonJokes( Bitmap *image )
{
return 0;
}
The paper is a bit confusing at first, and the /. summary doesn't help. Basically, they developed a sorting criteria to reduce the amount of work for the editors. In an isolated comparison of two jokes, the funnier joke wins 64% of them on average; this is quite better than a coin!
To get a sorted list, they run a "comparison tournament" between the jokes. The 55.8% number means that the funniest joke is in the top 55.8% of the list on average; if we are willing to occasionally miss a brilliant joke, we can cut the list in a little more than half and still keep most of the great jokes.
The full paper is http://research.microsoft.com/...
It's not the fall that kills you. It's the sudden stop at the end. -Douglas Adams
1) Computers will ever only be as smart as the people who program them
2) That we never have to worry about them being smarter than us because only a race of morons would even bother trying find the humor in a New Yorker comic
The situation you describe is different to what this algorithm has to deal with (just funny vs. not funny). In any case a 55% success rate would be equally bad in both scenarios.
For example: if out of 10 funny cases, the algorithm concludes that 5 are funny and the other 5 are not; it is irrelevant the fact of having just one not-funny option or as many additional ones as required (interesting, insightful...); the algorithm would fail in a very relevant proportion of the cases. And its failure would be very similar to the one output by a random decision maker (whose theoretical error is precisely 50% when deciding whether something is in some way or not).
Anything below 75% converts almost-understanding into random behaviour in most of the situations. What would you think about a person getting only half of the jokes you tell? Better: about a person just understanding that your sentences are expected to be forming a joke (properly getting the joke is a completely different story).
Custom Solvers 2.0 = Alvaro Carballo Garcia = varocarbas.
So, they are putting an AI through a process that destroys a human mind in about two years? What could possibly go wrong?
This is my signature. There are many like it, but this one is mine.
That's only true if its picking the best of two options.
yes, they are very dry
i get them now, as an adult
as a kid looking over old copies at my grandfather's house. it was mystifying: here was a cartoon, something i understood, but they were alluding to topics and humor that was utterly beyond me
intellectual property law is philosophically incoherent. it is your moral duty to ignore it or sabotage it
55% chance is not much better than flipping a coin
If there were only two captions to assess, you'd be correct. But there are 5000 to choose from, and even then the "right" answer is subjective, so 55% is a remarkably high figure.
John
Feeding jokes into a computer? ;-)
What could possibly go wrong?
Short Story by Isaac Asimov, Jokester:
http://www.epubsbook.com/Scien...
If I clone myself, can I call it a thread?
If a girl winks to us, can I call it a race condition?
Right up there with resume filters, copyrighted content detectors, search engines, and the like, now people will be writing cartoons geared toward the filters. "Well, at least I'm not Donald Trump's cat sipping cheap red wine while waiting in line for the latest iPhone!"
.
Prisencolinensinainciusol. Ol Rait!
Did anyone else misread:
as
My sci-fi novel, Ghost Thief, is now available from Amazon.com.
About 55.8% of the time the humans agree with the captions the AI selects, which is a pretty good percentage
My old prof (RIP Prof Swaminathan) in the freshman physics lab would insist on us doing an error estimate based on the instruments used. He would then chew our ears out if we report a result with more significant digits than what would be reasonable based on the error estimate. Three significant digit accuracy for this study? Or is that the joke?
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
The bartender looks at the cartoonist and says "is this some kind of joke?". The cartoonist looks at the AI and says to the bartender and says, "Give him a minute."
"He took a duck in the face at 250 knots." -- William Gibson, Pattern Recognition
function doesJokeExist(NewYorkerCartoon cartoon)
{
return 1;
}
-- I was raised on the command line, bitch
There are no jokes in New Yorker cartoons.
- Zav - Imagine a Beowulf cluster of insensitive clods...
Can it tell when the New Yorker is stealing jokes from Ziggy?
#DeleteChrome
Such as finding humor in New Yorker cartoons.
"55% chance is not much better than flipping a coin"
Which is fortunate, because an AI that achieved an 80-85% rate on New Yorker cartoons would have to have evolved at a star considerably older than the Sun.
2:1 would be for a two headed coin, but what if there were 100 sides to it?
“Common sense is not so common.” — Voltaire
The pig says, "my wife is a slut"?!
Most linux users don't know this, but the man pages were named after Chuck Norris. Chuck Norris fsck'ing hates noobs!
Data and Joe Piscopo on the Holodeck for 500, Alex.
That article is pretty light on actual information...anyway, here's a paper about getting machine learning to recognize opportunities for "that's what she said":
http://www.aclweb.org/antholog...
Ok, so it can recognize a joke, the real question is can it tell if a joke is funny?
Sorry, teleporters just kill you and then make a copy. A perfect, soul-less copy.
"Ouch!" They say. A Unix admin slides beneath the bar and asks them, "Why don't you try looking where you are going?"
It was the best I could do.
"So long and thanks for all the fish."
and destroys humanity.
This is no obstacle for those who are easily amused!
Recent papers have reported dramatic progress in identifying some
of the mutations responsible for autism, which in some cases
seem reversible.
One study, which will appear in Cell (paywall:/dx.doi.org/10.1016/j.cell.2015.06.045), identifies
protein kinase A (PKA) as an upstream regulator of the
autism-linked ubiquitin ligase protein UBE3A. A UBE3A mutation,
disrupts this regulation, leading to excessive UBE3A activity
against itself and ultimately to snaptic dysfunction. The authors suggest that
a phosphodiesterase-4 inhibitor named rolipram can boost UBE3A
levels thereby restoring normal function.
An unrelated study investigated an autism-linked mutation called
Shank3, which can be potentially
reversed (http://www.cell.com/cell-reports/abstract/S2211-1247%2815%2900496-9)
Please ignore the previous comment. It was posted to this thread by mistake.