Mattel has announced Pictionary Vs. AI, a version of its visual guessing game where an AI model does all the guessing, using computer vision with a smartphone as the medium. The company just announced the game at the 2023 Code Conference, which is co-hosted by Verge editor-in-chief Nilay Patel. The physical board game will cost $24.99 starting on September 27th at Mattel’s website, and it will head to retail on October 2nd.
The idea is that players draw cards that assign a given word, then sketch whatever that word is before letting a web app guess what they’ve drawn. The app uses your camera to look at the artwork. You win (and move forward on the board) either when the AI identifies the image or by successfully gambling — er, I mean, correctly predicting — if the AI will figure it out. You have to bet before showing the drawing to the AI model.
You can also trigger special challenges, like one where each person draws a single line on another person’s art or everyone draws with their eyes closed.
Mattel’s VP and global head of games calls Pictionary Vs. AI “a new, modern way that fans can engage with the brand, combined with the hilarious quickdraw gameplay they’ve always loved.”
Ah yes, engage with the brand. I’m always telling my partner how I wish we did more of that.
Mattel provided The Verge with some images of the game:
The game has 224 word prompts in total, which does make me wonder how hard it will really be for the AI model to guess what you’ve drawn. While Mattel hasn’t said whether the AI model can recognize more words than that, 224 is far fewer than the 2,500 words of the original Pictionary.
The AI model actually comes from Google, along with the “millions of user-submitted drawings” used to train it. Remember that Google Quick, Draw! game that made the rounds online a few years ago? The one where you clumsily draw with a mouse on your PC or by dragging your fingers around your phone screen and it guesses what you’re sketching? Yeah, it’s literally that model but in board game form.
And in case you were curious if this is just a sneaky way for Google to further its model’s training, Mattel says the AI model “can only recognize images from the Pictionary Vs. AI preset database.” People attending Code Conference can try the game in person at the Pictionary exhibitor’s table during the conference — here’s how to get tickets to the show.