Jeff Hawkins is best known as the father of mobile computing. His inventionsthe PalmPilot and its trademark Graffiti writing system, and the Treo smartphonerevolutionized the way we organize our lives. But these days, the 51-year-old is focused on a much bigger project: using advances in neuroscience to help create truly intelligent machines.
Hawkins's new company, Numenta, has just released a free software platform for developing intelligent applications based on a new theory of the brain called hierarchical temporal memory (HTM). The theory looks at how the neocortexthe wrinkly, outermost layer of the brain that accounts for about 60% of brain volumefunctions. The neocortex is responsible for all of our higher functions, including speech, vision, logic and motor control. Although these functions are in different parts of the neocortex, remarkably, the anatomy of the neocortex doesn't change depending on what it's doing. This has led Hawkins and others to believe that the brain does what it does using the same tools. In other words, if we think of our brains as a computer, then we don't need to have separate software for language, math or vision. Hawkins's enthusiasm for this new venture is clear. "Intelligent computing could be a bigger industry than general computing," he says from his California office. "We want to be the catalyst for that industry."
To date, intelligent computing systems have been single-purpose constructionscomputers that have mastered chess, but can't play backgammon or bridge, let alone understand simple sentences. Hawkins and his team are taking a different approach. Instead of writing a program to deal with a specific problem, an HTM system is instead "trained" at a task. For example, in a demo on Numenta's website, an HTM system is shown a series of characters and simple drawingsa ladder, letters of the alphabet. It's then shown those same images over and over again, but in distorted forms: bigger, smaller, stretched, with random pixels added on. At the end of the training process, the system is able to recognize freehand drawings of the images it has learned. It doesn't use a specific algorithm to recognize a ladder; rather, it has learned the concept of what a ladder looks like. By building a platform around a model of the brain, Hawkins's team plans to create machines that could learn to do just about anything.
But the platform has a long way to go. And Hawkins has his criticsthose who say the brain is simply too complicated to understand, that it operates on some unknown properties of physics, or that there is just "something mystical" about it. Hawkins has only this to say: "There is nothing magical or mystical about the brain. I'm a no-magic person."