The algorithm came up with some bizarre results after learning from genuine candy heart messages.
After craft beer, Halloween costumes, and Christmas carols, research scientist Janelle Shane is back with another mission for her neural network.
She is out to prove that artificial intelligence, which can redefine the future, can also be hilariously weird on occasions.
Shane took up her neural network and fed it with a massive chunk of data.
The process involved training the machine learning algorithm, but instead of costume or beer names, the researcher used romantic candy heart messages this time, something fitting for Valentine’s Day.
The algorithm learned patterns in random messages like “Love You” and “Be Mine” and imitated the data to give out its own results.
While some of the messages like “My My”, “Cute Kiss”, “LoveBun”, and “My Bear” did appear good enough for candy hearts, others were hilariously weird and reminded us why machines are still several steps behind humans. They cannot understand the nuances of different words, at least as of now.
The neural network generated bizarre messages like “Stank Love”, “Sweat Poo”, “Look Big”, “How You Hot”, “Love 2000 Hogs Yea”, “Chert Face” and “You a Goo”.
“There was yet another category of message, a category you might be able to predict given the prevalence of four-letter words in the original dataset,” Shane wrote in a blog post detailing the entire list of results. “The neural network thought of some nice new four-letter words to use.”
Describing the weirdness of the answers, Shane told Gizmodo that the neural network might have created those messages because they seemed good pronounceable. In a previous experiment with Christmas carols, she had said that the system doesn’t have any built-in knowledge of English or clue what Christmas carols are.
You can find the complete list of AI-generated messages on her blog.
I trained a neural network to generate new candy heart messages – some more successful than others. https://t.co/tPZyGV3upg pic.twitter.com/9J8HjiMshV