Bench Press

Computers can do some pretty incredible stuff. We’ve seen computers analyze evolutionary trees, master the game of Go, and even take on the challenge of Jeopardy!. However, as amazing as computers are, the one glaring deficit of computers is its inability to extrapolate and make connections, something that comes easy to humans. The culprit is how computers are designed. Computers are built to follow algorithms: a strict set of guidelines which, when executed faithfully, will yield the correct answer. However, ask a computer to build upon these algorithms is a whole different issue. Humans, on the other hand, quickly learn to develop schemas and a general model of understanding. From these basic concepts, we can use deduction to answer a question we’ve never been asked before by combining different skills together to achieve this goal. For example, humans will quickly identify that a certain object on a street is a car, even if we’ve never seen that certain model of car before.

Enter Cognitive-Level Annotation Using Latent Statistical Structure (CLASS), a project that is trying to push computers to recognize specific classes of objects the way humans can. Luc Van Gool of Belgium’s Leuven University (KUL), a member of the CLASS team, sees great promise in CLASS and has already found a marketable use for it in cell phones. Through a mobile service, CLASS will be able to let its users take a picture of an object, say a famous monument, identify what it is, and provide further information to the user via the internet.

“It’s like the object itself becomes the link to further information,” observes Van Gool. He expects the application of this technology to expand rapidly. For instance, cities and museums may offer interactive guided tours or guide books through kooaba.

While CLASS is still a long way off from being complete, it is clearly a giant step towards extending the capabilities of computers and furthering the field of artificial intelligence.