"If Waldo is identified the arm will move to that position and point the little creepy hand at Waldo on the page," says Reed.
Reed trained the machine intelligence by uploading 62 images of Waldo from Google Images.
"I thought that wouldn't be enough data to build a strong model but it gives surprisingly good predictions on Waldos that weren't in the original training set. It detects false positives roughly 5 percent to 10 percent of the time so it's not perfect. Maybe I could get that lower if I built up my Waldo training set more," Reed says.
"There are a surprising amount of faces on a Waldo spread. Did you know some of those Waldo pages have over 300 faces? I didn't, but some do! Crazy," Reed tells CNBC Make It. "The AutoML quota is 600 predictions per minute so I can really only process a page or two per minute."
Redpepper built There's Waldo to demonstrate the creative agency's ability to quickly generate ideas based on new technology.
"We do 'Sprint'-based innovation work with brands like Mars Petcare, Verizon, Kirkland's etc. and are constantly exploring new technologies so we can leverage them throughout our creative process," Reed says.
The same technology could be applied to other use cases, like identifying how many pairs of Nike shoes or Gucci handbags are being worn at the mall, count the number of cans of LaCroix or Coke are seen in a photo from an event like a music festival or identify a missing child on all public cameras when an Amber alert is sounded, Reed suggests.
Other ways that Redpepper has played with technology include building a QR code with Oreos and having humans sing Christmas carols written by artificial intelligence.