A research scientist named Janelle Shane trains neural networks, as stated in her "About" section in the links below. She's revealed a few of her projects, including a neural network framework that produces pickup lines and her most recent creation, "Candy Heart messages written by a neural network."
In her "About," she says she trains neural networks to "write unintentional humor as they struggle to imitate human datasets. Well, I intend the humor."
 |
| How romantic. |
She also does other cool stuff with Pokémon and recipes, and has been written about before (this article discusses a neural network she trained to tell knock-knock jokes), among other things.
It's kind of incredible how a lot of these messages are just as or even more romantic and coherent than regular human-made pickup lines and candy heart messages. SWEET PEAR is infinitely more romantic than GO FISH. I wonder if all our romantic candy heart slogans will be written by AI in the future.
ReplyDeleteLOVE 2000 HOGS YEA
Maybe that's more a reflection on the incoherence of the real conversation hearts than it is on the coherence of the neural network's output...
DeleteIt definitely is.
DeleteBut actually it does raise some interesting questions about the legitimacy of the Turing test as an approximation of artificial intelligence. If we set the bar for imitating humans extremely low, does it really count if a machine successfully imitates us? Do we need to redesign the idea of the Turing test? Or do we only consider it legitimate in situations where people speak clearly and coherently?
DeleteI rather think that was what Alan Turing had in mind (imitating people who speak clearly and coherently), although the chatbot creators have tended toward other directions sometimes because they are easier to accomplish.
DeleteI wonder if applying the same sort of things we did to the NLP library would help this improve over time, or if people would just react more positively to the ridiculous ones and throw the whole data set off balance. I think it would be interesting to rank the pos/neg sentiment of each of these and see which ones are the fan favorites.
ReplyDeleteJoe I think you bring up an excellent point when it comes to training AI using the public's involvement in any capacity. Most people will not take it seriously right off the bat, and even if they do, the variation of human being's likes and dislikes are so varied, we might end up with even more ridiculous candy hearts. We wouldn't be able to blame the AI either. Humanity is too whacky for its own good, and there are consequences. Like unsatisfactory candy heart slogans.
ReplyDeleteHow cool would it be this neural network could help you find your soulmate?! If each candy heart had some feedback survey, the neural network could use this feedback to produce more successful candy hearts. Yes, I doubt it's going to take more than a candy heart to win over someone, but maybe A.I. will be computer scientists saving grace when it comes to their love lives. Happy Valentines Day CSC-320!
ReplyDeleteI really love these things. I know it's just a system throwing together words at the moment, but it's almost like an early era is actual intelligent AI. I mean, I'm not sure I'd ever swoon over an AI system giving me a candy heart, but the thought is so exciting!
ReplyDelete