MIT Model Helps Computers Think Like Humans

The researchers' algorithm lets computers use multiple approaches to examine data, much like the way humans size up the world.
In a development that will extend the eternal quest of creating computers that think like humans, two researchers working at Massachusetts Institute of Technology have developed a model that helps computers recognize patterns in the same way as humans do.

The two researchers, associate professor of brain and cognitive sciences Josh Tenenbaum and recent MIT PhD recipient Charles Kemp, produced a broad algorithm that examines several different approaches of looking at data that is similar to the way humans typically size up different situations.

"Instead of looking for a particular kind of structure, we came up with a broader algorithm that is able to look for all of these structures and weigh them against each other," said Tenenbaum, the senior author of the paper.

Realizing that humans naturally tend to sort out order from different sets of information, the researchers knew that computers, on the other hand, typically don't know where to begin when faced with large and varied data sets -- unless, of course, the machines have been programmed to seek a specific structure like a hierarchy, a cluster, or linear order.

The model developed by the MIT researchers offers various data structures and then finds "the best-fitting structure of each type for a given data set and then picks the type of structure that best represents the data," according to this week's MIT announcement explaining the researchers' work. The researchers said that humans -- even young, inexperienced children -- carry out similar tasks every day, often unconsciously.

"We think of children as taking in data, forming theories, and testing those theories with experiments," said Tenenbaum. "They're like little scientists. Until now there's been no good computational model for how children can, like scientists, grasp the underlying global structure of a set of data."

Kemp, now an assistant professor in the department of psychology at Carnegie Mellon University, developed the algorithm while he was at MIT. On his Web site he notes that he is "particularly interested in high-level cognition, and [has] developed models of categorization, property induction, word-learning, causal reasoning, similarity, and relational learning."

The researchers believe their work can impact and further the field of artificial intelligence.