Don’t have a Halloween costume yet? Well, you’ve got 24 hours, and if you’re tired of the same ol’ costume ideas perhaps a new AI tool developed by computer vision researcher Janelle Shane and published by the New York Times can help. The experiment, which sought to determine what someone might dress up as this year based on past Halloween costumes, uses an algorithm along with a list of 7,182 costumes to give costume suggestions. The AI was self taught, which means it struggled at first to come up with legitimate costumes. Really early efforts spit out gibberish, but eventually it came up with some more amusing recommendations like, “Dorothy, from Grouch,” “Frankenstein’s Bunny,” “Sexy Banana,” and “Cyborg baby man.”
While the experiment has definitely improved beyond gibberish, I personally got “Spank pope,” and I’m not sure what to make of it. You can learn more about the experiment and get your own costume suggestion here.
For fun, here are some of the highlights of the other AI generated costumes that were featured in the article with illustrations from by Jessia Ma.
If you found these amusing there are a few dozen more in the original piece that you’ll likely get a kick out of, especially their collection of “sexy” costumes generated by the AI. These were just some of my favorites. If you give the tool a whirl, let me know on social media what you get, especially if it turns out funnier than “Spank Pope.”
