Slashdot Mirror
AI Is Funny - a Generative Joke Model
RemyBR writes "Can computers tell a good joke? Is comedy just a matter of statistics or is there something only a human can bring to creating a joke? A joke generator created at the University of Edinburgh (PDF) suggests that AI can be funny. Some AI generated jokes: 'I like my relationships like I like my source, open,' 'I like my coffee like I like my war, cold,' 'I like my boys like I like my sectors, bad.'"
I like my slashdot like I like my like like like...
Loop detected, aborting.
Bazinga.
Am I the only one who read the headline and thought of Al Gore?
ob. joke.. I like my coffee like my men - strong and black.
...I like my computer generated joke examples like I like my MSDN how-to articles...terrible?
I like my AI like I like my zombies, mancery.
I like my women like I like my sectors, industrial.
Someone flopped a steamer in the gene pool.
... and then I built a supercollider.
*Obligatory Futurama reference*
... milky white, weak, and tasteless.
Do they still sell those party joke books from the '70s? That's the level of humor we're talking about here.
I like my jokes like I like my AI, statistical.
give us a website with access to this joke maker! not just a journal article
Covered in bees!
Reading the article, they have a (human-created) statistical model for the specific words people will find funny in this one, exact type of joke. The only thing the "AI" is doing is analyzing word frequencies against this model. I suggest calling these "statistically-generated" jokes, or similar.
Well, in the movies anyway. Remember the first robot who could NOT get a joke? (Robbie)
And I think the first wise-cracking robot? (Johnny 5 in "Short Circuit")
And then of course there was Data .. with mixed results in reference to humor and jokes.
They built Funnybot? Do we have to worry about it exterminating the human race?
I like my AI like I like I like all my intelligence: downloaded from my opponent's servers in the dead of night.
I like my women like I like my AI joke generator. Inaccessible to most of the interested geeks.
Al is very funny, I like the way he deadpans"I don't think so, Tim" after Tim suggests a particularly bizarre way to go about a home improvement project. Glad to see Slashdot is finally giving him proper recognition.
Okay, did anyone else think of the one or two episodes where Data was trying to be funny and do stand-up? This story reminded me of that. So did the jokes.
Somethingawful beat them to it: http://www.somethingawful.com/flash-tub/laff-bot-beta/
frigid, expensive, bitter and eventually dumped in the gutter.
The fascinating thing to me was that the funniest jokes it managed to come up with had a definite misogynistic streak. Is it because misogyny is inherently amusing, or because sexist jokes are low-hanging fruit? Link to more coverage of the same story.
I think http://twitter.com/Horse_ebooks has a significantly higher success rating than 16%. Granted its tweets are not "jokes." But I hardly think "I like my coffee like I like my war - cold" qualifies as a joke. Especially since they are trying to make the AI "funny" out of any sort of context.
If you ask me it takes two people to make a joke. If the teller wants laughter AND the listener laughs it's a joke. In other words joke creation is a training exercise where the teller tries something and the listener gives feedback. Here it seems the AI has some joke building knowledge but that knowledge will expire when people get bored of what it has to say and it will need to adapt.
So from what I can read, this particular joke generator uses pretty straightforward word association and some Bayesian weighting. This article describes model that's a bit more complicated (having to do with graphs of word associations and forming loops of optimal length), and I wonder if it'd produce better (that is, funnier) results.
The problem will arise when the computers recreate the Funniest Joke in the World, sealed under a monument in 1950 labeled "To The Unknown Joke" After all, the computer will not know it has created the Killer Joke, leaving us with the danger of mass hysterics breaking out.
Usually people mix up humor and powerful humor. So they think of all the ingredients that make it work. The best example of humor is then the one that makes you laugh more.
But here's another angle: Just think of humor as having a humor part and a booster that makes you laugh more. The humor part is just the perception switch. It can be pretty mild. But add the naughtiness, the meanness , the embarassment and they provide a boost to the humor.
Taken that way, the AI examples in the topic article are really touching the essence of humor.
Read the article again and look at their premise. They wanted to show that semantic data was not important. That the AI-joke problem was simpler than previously thought.
Sure, these are simple, formulaic jokes, but they requires less domain knowledge. I find that as interesting as any other topic we've reduced by science, such as music to recorded waves.
adjective.
FLAC - Free Lossless Audio Codec
Just like at the movie theater!
Oops, wrong reference...
An enigma, wrapped in a riddle, shrouded in bacon and cheese
...Turing Test passes YOU!
You need Dr. Strangelove to walk into a coffee shop, and for the barista to ask him how he likes his coffee. Then MAYBE it's funny. Furthermore, the deliver and timing matters. You can deliver that line and kill or die. Finally, the person who sees this might not get it, or they might get it and just not think it's funny. Yeah, yeah, Dr. Strangelove likes the cold war. Not funny... to that guy; but maybe funny to you.
For all intensive purposes, "whom" is no longer a word. That begs the question, "who cares"?
Ask Siri where to bury a dead body - or where to buy a digital camera...heck, just say 'tell me a joke'...
IRC chatbots have been a source of hilarious topics for years
I like my machines like I like myself, Intelligent.
Mapping a common property between two different subjects literally isn't remarkable in of itself. A comedian invents far more jokes than they tell. You see, the trick isn't in coming up with jokes, it's knowing when they're funny enough to repeat.
In Mind Wide Open Steven Johnson points out that "Laughing is not an instinctive physical response to humor, the way a flinch responds to pain or a shiver to cold. It's an instinctive form of social bonding that humor is crafted to exploit."
Think about how often you laugh at references, the more obscure the better. You're sharing a bond with the person making that reference—and once you start looking for that, it becomes increasingly obvious (at least it did for me).
That's probably why "I like my X like my Y, Z" style jokes are funny—they make us think, "Wow, you and I both see that X and Y have that relationship, possibly based on abusing a synonym, which doesn't immediately spring to mind when you think of them."
The more I think about humor as an exploit of laughter as social bonding behavior, the more I notice it. And the more I notice people laughing when things aren't funny, but when it's appropriate to reconfirm a social bond (like when someone does something embarrassing that might take them out of the social norm, and the people around them laugh to reassure them that the social bond has not been damaged... much).
This is where I would make a joke about how geeks are not good at social bonding, but I'm too much of a geek to relate to such things.
Building Better Software
Which is why I hate TV shows with laugh tracks; particularly the ones that go off on every pause, even for non-jokes.
Well, surprise isn't the same as a perception switch, although there's overlap. Because I describe humor as a perception switch I'm saying it's the same mechanism as insight. Suddenly understanding something. Not that I came up with that myself.
The "human-generated" jokes were harvested from a corpus, using, as far as I can tell, a regular expression (the very fact that the paper spends all of half a sentence on this is quite unfortunate), and were therefore not necessarily intended to be funny in the first place. We can quantify: the human evaluators found them funny only 33% of time.
Another problem is that the algorithm is hand-fed some criteria which, in the authors' pre-conceived opinion, correlate with joke's funniness. As it turns out, the correlation is significant, yet weak; 16%, compared to 3.5% for random selection. This tells me as much about the hand-fed criteria as it does about the validity of this entire approach.
In my reading of the paper, the authors do not have a firm grasp of how perception of funniness is triggered in humans, and are ill-equipped to study this phenomenon. It seems I like my coffee how I like my computer scientists: not!
Did they just troll a list of prolog statements for things that match up? There may be some diamonds in the rough, but must will be duds.
I am not a robot. I am a unicorn.
Here I am, brain the size of a planet, and they ask me to make puns. Call that job satisfaction, 'cause I don't.
Have gnu, will travel.
I like my jokes like I like my symphonies
-
- - You can't take something off the Internet! That's like trying to take pee out of a swimming pool.
It's a bit of a letdown, but still interesting. I made an inspirational quote generator that certain types of people enjoy. But only certain types. Humor is not easy to get right, least of all automatically right.
He was the best. :)
If laughter is a social bonding behavior, why do people laugh at something they are reading when they are completely alone?
File under 'M' for 'Manic ranting'
"Dave, your joke is busting me up, literally."
Table-ized A.I.
If laughter is a social bonding behavior, why do people laugh at something they are reading when they are completely alone?
One's never alone with a good book...
(i.e. at the very least you're engaging socially in some sense with the author, and likely with the characters in the story as well)
I don't care if it's 90,000 hectares. That lake was not my doing.
I don't care if my computer can tell a good joke. I want it to laugh at my jokes.
ok, I'm going to pay attention to that now, thanks.
"First they came for the slanderers and i said nothing."
Odd... That was supposed to be "Reference" but reverence works as well.
This story seems inspired by Asimov's "Jokester", in which a humor "Grand Master" is attempting to find the source of humor by feeding Multivac a curated set of jokes. I won't spoil the ending, but let's just say the mice need a new maze at the end.
Defending IP by destroying access to it? That makes sense, RIAA/MPAA. Go to the corner until you can play nice!
I wonder if it could have been generated using the joke generator. Or maybe the slashdot post was generated by it as well. Or this post. Who knows ?!?
aside from those jokes really not being that funny, is that they are also counter intuitive, with perhaps the exception of the first. Assuming the AI lives in the computer, the last thing it would like is bad sectors. In fact, they often grind the operation to a halt. The same goes for coffee. Assuming a computer would, in fact, want coffee in any form, wouldn't it want it hot?
If(punchline="hot" || coffee="cold"){return "fail"}
All AI since the 70s is just statistics. They just don't call it statistics because somehow "AI" never fell out of vogue, despite it being the term used by a bunch of philanderers who never delivered the connectionist utopia they promised.
AI my ass. I'm pretty sure I could write that in SQL.
SELECT 'I like my ' || TRIM(a.noun) || ' like I like my ' || TRIM(b.noun) || ', ' || TRIM(a.adjective) || '.'
FROM a.AdjectivesAndNouns, b.AdjectivesAndNouns
WHERE a.adjective = b.adjective
AND a.noun b.noun;
Or something like that.
Of course you'd need a table of nouns with common adjectives, but that would not require AI to build either.
A few of them might even be funny. This is at least as good as the AI version, because clearly the AI is doing a bad job at recognizing what is really funny.
Given that it was a computer talking about its preferences, I really don't see it. Computers don't have genders, which to me adds extra layers to these jokes that wouldn't be there if a gendered human was delivering them.
There shouldn't really be any sexual meaning at all in there, unless you brought it in there yourself. So if you saw misogyny in there, you are reading those jokes a lot differently than I am. Nothing nessecarily wrong with that, but its *you* who did it, not the computer.
http://www.dailymail.co.uk/sciencetech/article-1328640/Why-laughing-instinctive-learn-cry.html
File under 'M' for 'Manic ranting'
Computer Generated jokes you say?
What do you think Windows is ?
-- 29A the number of the Beast
Edinburgh is the location of one the biggest comedy festivals https://www.edfringe.com/ running right now....
More like a meme generator. I hope they won't release this beast on the Internet.