Unsupervised AI makes up some pretty funny jokes




Funny thing: the first thing that comes to mind is the computer’s attempts to understand humor in The Moon is a Harsh Mistress.


If they program the joke writing AI to follow up each joke with the phrase “you see what I did there?”, the AI could take over for Craig Ferguson.


Spellcheckers are basically engines for creating puns: Freeway emergency telephones are for your connivance and safety …was one classic from my former workplace.


“i am funnybot”


“Unsupervised AI makes up some pretty funny jokes”. Actually no. No, it doesn’t.


A little bit clever but not what you would call funny.


If unsupervised AI makes a joke and nobody laughs is it still funny?


Anyone who has ever told a joke knows that whether it is funny or not depends as much on delivery as material. Measure that computers!


“Okay, the pod bay doors are open, Dave… Ha! Gotcha! Okay, SRSLY, they’re open now, Dave… Ha!”


Only if it falls in the woods.


It’s particularly amusing in that, while their second model came up with jokes that were viewed as funny by humans only 16% of the time, similar jokes collected on Twitter were found to be funny by humans only 33% of the time.

One has to wonder whether their model is not terrible when coming up with funny jokes as compared to humans, or whether jokes on Twitter are themselves actually generated by rather deficient algorithms.


It could be that, no matter what, any individual will only find 1/3 of jokes funny. If one had an algorithm that did 1/6 it would be a simple thing to introduce an human or two into the chain to filter out the unfunny ones. So the 1/3 number is probably some kind of maximum. A comedian could lift it higher with good delivery.


So are the NSA going to be surprised when their sys admin replacement fills up all the storage with jokes.


Except that’s not artificial intelligence. It’s exactly like punching numbers into one of those old-fashioned calculators with gears and a motor in it, and then pressing the ‘calculate’ button.

“The researchers wrote code…” Exactly. Where’s the AI part?

AI would be: the computer, on its own, reads the net, discovers that humans like humor, discovers the ‘I like my x like I like my y’ meme, discovers Google n-grams on its own, and (all of this without being told what to do) starts cranking out ‘humor’ and checking for applause.

Read about Clever Hans again. Soon.


More than one “kind” of AI. This would be in the realm of expert systems.


“Half as funny as the average twitter joke” is nothing to be proud of.


Man, leave the AI unsupervised for a minute, and it just starts screwing off. Stop cracking jokes, AI, and get back to work!


Another I remember from days gone by…“Pentium is not a word, did you mean penis?”


But how does it know if the jokes are funny once or funny always?