SRI, the creators of Apple’s voice-controlled assistant Siri, is developing a new computerized assistant that watches and analyzes your actions in certain situations, and then helps you by predicting what you will do next. This assistant, called Bright, will overcome the physical bottleneck with human-machine interaction. As an example, Bright might know from experience that you like to watch cute cat videos on YouTube with your morning cup of coffee — and so when it sees you approaching your computer with a cup of coffee, it would automatically bring up YouTube. Less intrusively, it might instead make the Chrome/Firefox button your toolbar bigger and easier to click. The initial purpose of this assistant, dubbed Bright, is to cut down on the steadily increasing cognitive workload faced by workers in high-stress, computerized environments — but in the future, it’s easy to see how Bright might find its way to gadgets, PCs, and smartphones.
SRI International, originally the Stanford Research Institute, is a non-profit research institute and business incubator founded by Stanford University in 1946. Technologies developed at SRI have been spun-off to form such luminaries as Siri (which was acquired by Apple), Symantec, Verbatim, Nuance Communications, Vocera, and Intuitive Surgical. Much of the work at SRI is funded by various wings of the US government; Siri, for example, was developed as a result of DARPA’s CALO (Cognitive Assistant that Learns and Organizes) project.
Bright is intended to overcome the bottleneck that is currently present with human-machine interactions — or last-meter bandwidth, as SRI calls it. In short, computers are capable of displaying and processing much more information than we can feasibly handle. Bright is essentially a piece of software that tracks every action that you take on a computer, and at the same time watches you with state-of-the-art sensors, including gaze-tracking cameras, touchscreens, and gesture detection. By correlating this information with what’s actually happening on the computer, Bright builds a cognitive model of how you behave in various situations. With enough learning, Bright can predict your behavior, and thus make decisions for you, or at least simplify the decision making process.
In the videos above, you can see that the technology is still in an early stage of development — and also that Bright is much more than just an assistant for high-stress computerized environments. As always with research institutes, the team behind Bright probably doesn’t know exactly how the technology will turn out, or who will try to commercialize it, and so a scatter-gun approach ensues. As it stands, with its huge tabletop display and multi-user support, Bright is being developed for cybersecurity and emergency response, where huge amounts of data need to be visualized and responded to rapidly. It seems the interface is the most complete part of Bright; now SRI needs to add the prediction and automation elements (the hard bit).
According to Bill Mark, SRI’s vice president of information and computing sciences, and one of CALO’s principle investigators, learning user behavior is a “small-data problem.” Whereas big-data problems might sift through millions of records to find patterns, a single human doesn’t generate that much data — and we tend to change our behavior regularly, too, which really upsets the machines. “We’re not putting in that much data. These machine-learning algorithms like to generalize over very large amounts of data,” Mark tells Technology Review.
Moving forward, once SRI’s researchers have worked out how to predict your next action, Bright could revolutionize how we use computers. In much the same way that Google Now learns where you work, and then displays the ETA when you’re about to leave home, Bright could learn how you react to incoming email, automatically scroll websites based on your reading speed, or mute everything except mission-critical applications when you have a deadline to make — and that’s just the beginning.