This bot is called "
Faces are captured and then shot out to Google's Auto ML Vision service. That service can compare all the faces sent its way to the one face it has been trained to find. In this case, it is looking for our pal Waldo. The system feels confident in a match that is 95% or higher. Once that match is made, the robot hand proceeds over the photograph and points a finger at the matching face.
The bot can find multiple faces if Waldo is hiding in more than one spot. It's been years since I've hunted for Waldo, but I rather thought the point was that there was but one Waldo per page. Machine learning can do all sorts of cool things for humans today from making NCAA Final Four predictions to fixing your grammar.