A video was posted to YouTube on August 8, 2018 depicting a seemingly simple robot, armed with—well, an arm, with a human-like miniature hand made of synthetic materials. Just like its simple appearance, the robot’s function is equally as simple, but amazingly accurate and effective: powered by artificial intelligence, the robot is able to accurately pinpoint the location of Waldo (or Wally, as he was originally known in the UK), the well-known character in the series of children’s puzzle books named “Where’s Waldo?” first published in 1987.
The video is a demonstration by Redpepper, an ad agency based in Nashville, Tennessee to test out Google’s AutoML Vision service. Google announced this service in January 2018 as a suite of tools for people with limited AI experience to apply the company’s tools and technology to projects.
The robot and its AI is called “There’s Waldo”, and the way it works is by first having the “arm” taking a photo of a page from the book, which is then scanned to detect faces in the page, then matching the image on the page that looks most like Waldo through machine learning, and then finally having the arm move to point out Waldo on the page itself. The machine’s intelligence was trained by having 62 images of Waldo uploaded to it from Google Images, of which the data was used to help identify Waldo among up to 300 cartoon faces on a single page spread in the books.