3 ms·

Botnik, an AI-assisted humor application

33 points by gk1 9y ago

ukulele 9y ago
Humor should be a good target for machine learning, since it's less intuition based and more based on repeatedly trying to be funny and gauging people's responses all the while. Though it seems more "human" than some of the other ML tasks (eg be the best at Go), I wouldn't be surprised if comedy ends up being a highly successful application for these algorithms.
- theoh 9y ago
  By "intuition" you must mean instinct -- but this isn't a separate category from learned behaviour for an AI that can be trained over the equivalent of many generations. In any case, botnik is not an autonomous AI, and humans are responsible for the important decisions (is this incongruous word funny?) when composing a text with it. So it's a long way from AI humour.
slazaro 9y ago
I found the fake reviews surprisingly laugh-out-loud funny. I think it's because it retains the structure, it's not total nonsense because close words are linked properly (like markov chains), but at a higher level it makes no sense, making it absurd in the end. So it sounds like a parody of the original texts it's based on.
King-Aaron 9y ago
"Ever since the 1990's, when the Internet was a very small book with buttons, about the size of a coconut, we were thinking of robots. Robots that can tell stories about animals swimming through media industries. Robots that are designed to get published online. Robots that are on fire and in love." Perfection
dogruck 9y ago
This is cool. A few reactions: 1. Comedy writing is much more skill-based and formulaic than many people realize. Most people could learn how to write good jokes, just like most people could learn to play basic Go. (Beating Lee Sedol or Paul Mooney is another thing) 2. The resulting comedic quality is subjective. I agree with other commenters that several were LOL worthy. 3. Comedy has a long tradition of building on prior comedy, so this feels like a perfect area for building on curated corpuses.
- YeGoblynQueenne 9y ago
  With a bit of training I'm fairly certain I could beat Paul Mooney at Go. Also, I'm pretty sure that this kind of quipping can't be easily automated. You need to be able to figure out why beating Paul Mooney at Go is not such a big deal as beating Lee Sedol at it _and_ why this could be considered a funny association to make in the context of the above sentence. This is actually hardcore AGI stuff, well beyond our current reach.
  - dogruck 9y ago
    Hehe, indeed. Long way to Go.

33 points by gk1 9y ago

Humor should be a good target for machine learning, since it's less intuition based and more based on repeatedly trying to be funny and gauging people's responses all the while. Though it seems more "human" than some of the other ML tasks (eg be the best at Go), I wouldn't be surprised if comedy ends up being a highly successful application for these algorithms.

theoh 9y ago
By "intuition" you must mean instinct -- but this isn't a separate category from learned behaviour for an AI that can be trained over the equivalent of many generations. In any case, botnik is not an autonomous AI, and humans are responsible for the important decisions (is this incongruous word funny?) when composing a text with it. So it's a long way from AI humour.

slazaro 9y ago

I found the fake reviews surprisingly laugh-out-loud funny. I think it's because it retains the structure, it's not total nonsense because close words are linked properly (like markov chains), but at a higher level it makes no sense, making it absurd in the end. So it sounds like a parody of the original texts it's based on.

King-Aaron 9y ago

"Ever since the 1990's, when the Internet was a very small book with buttons, about the size of a coconut, we were thinking of robots. Robots that can tell stories about animals swimming through media industries. Robots that are designed to get published online. Robots that are on fire and in love." Perfection

dogruck 9y ago

This is cool. A few reactions: 1. Comedy writing is much more skill-based and formulaic than many people realize. Most people could learn how to write good jokes, just like most people could learn to play basic Go. (Beating Lee Sedol or Paul Mooney is another thing) 2. The resulting comedic quality is subjective. I agree with other commenters that several were LOL worthy. 3. Comedy has a long tradition of building on prior comedy, so this feels like a perfect area for building on curated corpuses.

YeGoblynQueenne 9y ago
With a bit of training I'm fairly certain I could beat Paul Mooney at Go. Also, I'm pretty sure that this kind of quipping can't be easily automated. You need to be able to figure out why beating Paul Mooney at Go is not such a big deal as beating Lee Sedol at it _and_ why this could be considered a funny association to make in the context of the above sentence. This is actually hardcore AGI stuff, well beyond our current reach.
- dogruck 9y ago
  Hehe, indeed. Long way to Go.