Whether it was laughing at iPhone personal assistant Siri as she repeated herself for five minutes straight or chucking as bad guys in video games ran into walls, we’ve all had our encounters with artificial intelligence that seemingly lacked a little something called “common sense.” As technology has improved, so has A.I., but for the most part, computer systems, game characters, robots, and other artificial intelligence models just haven’t reached the level of smarts that science fiction dictates as a danger point for robot rebellion (a la The Matrix and the Terminator franchise).

Now, however, computer scientists at Carnegie Mellon University are looking to change all that, working to teach a computer how to think and behave with common sense, and using the most unorthodox method to get there. The computer system is called NEIL (or Never Ending Image Learning), and has been set the task of browsing the internet for images since mid-July.

On the surface, four months of 24/7 image browsing may sound like mindless work unlikely to build up any intelligence at all. It certainly doesn’t sound like the kind of strategy that could clear the historically unprecedented hurdle for computer common sense. However, according to a report from the Associated Press, the new computer system is working through millions of images with the goal of it determining – on its own – what those images mean collectively.

Said in more simple terms, Carnegie Mellon’s NEIL is meant to analyze pictures in order to form connections between them. In theory, those connections can then be used by the computer to postulate theories about the different objects in said pictures. So far, the computer has identified some 1,500 objects and made 2,500 associations between them. Already, some of those associations are right on the money, like a theory that zebras and tigers share likenesses in their appearance.

Of course, considering the number of weird images on Google, NEIL is bound to make some pretty off-the-wall claims as well, such as one which postulates that news anchors look like Barack Obama. Still, NEIL is doing well after only four months, and experimenters always knew there were going to be challenges with creating a learning computer model.

If anyone is up to those challenges, chances are that it’s Carnegie Mellon. In the 1990s, researchers at the university developed a chess computer called Deep Blue, which managed to topple world champion Garry Kasparov in a six-game match. However, in that case, computer scientists merely had to develop a supercomputer that could learn and master the rules and strategies of chess. With NEIL, the goal is a much wider cross section of knowledge – as well as the ability to learn and think on the go – and that mission will undoubtedly take a few years.