In 2007, Silicon Valley's SRI International formed Siri Inc. to commercialize a virtual personal assistant technology born out of the institute's DARPA-funded CALO (Cognitive Assistant that Learns and Organizes) artificial intelligence project. A free app for the iOS platform was subsequently launched as a public beta in early February 2010, and just a couple of months later, Apple acquired the company. Spin forward to October 2011, and a conversational search assistant called Siri was launched as a new feature for the iPhone 4S.
A little while later, Google premiered its own digital PA in Android 4.1 (Jelly Bean). In addition to providing Siri-like search and assistance using natural language, Google Now delivered information and suggestions based on actions or decisions that the user had previously taken. SRI's latest project, bRight, progresses beyond both systems as an answer to what's been dubbed cognitive overload, where the tidal wave of information that can flood in during emergency situations can prove to be just too much to deal with effectively and, perhaps more importantly, rapidly.
The research prototype uses face recognition (though more secure biometrics, such as iris scans, will likely be implemented in the future) and gaze monitoring systems, along with proximity, gesture and touch sensors, to build detailed user profiles. In a similar way that modern computers might make valuable performance gains by effectively taking a shortcut when certain conditions are met, bRight's powerful AI software uses this information to anticipate what might be needed so that only data that's relevant to the job in hand is presented to the user, necessary tools can be literally placed at a user's fingertips, and repetitive tasks can be fully or partly automated.
For example, at a fairly simple level, if a user highlights a word in a document, the system can guess which menu items might be needed next and present the user with likely choices. Or if someone's writing a specific kind of email, such as a staff newsletter or performance bulletin, bRight may be able to determine its recipients based on previous activity, and pre-populate the Send To field. It might also detect potential errors or breaches of standard protocol.
"If bRight recognizes a user's action to be of a certain class, then it could provide corrective action," explains Dr. Grit Denker of SRI's Computer Science Laboratory. "Say I am writing an email about new bRight ideas and I am sending it to a bunch of people. bRight could recognize that I usually first send this to an internal team, before sending it to outside folks. Thus, if I am about to send such an email without having first sent it to my team, bRight could notify me whether this is on purpose."
"bRight combines semantic markup in the application layer with sensors at the observation layer (e.g., touch, gaze, gesture, etc.)," says Denker. "This combination provides higher precision for prediction, especially in an environment where you do not necessarily have days or months of training data. In order to be useful, it has to have high accuracy. This can only be achieved if the cognitive models we intend to build are tuned to the applications. We are currently working on developing a cognitive model of users in the cyber domain using our tools. We are very interested in finding partners who would work with us to instantiate bRight for domains that meet at least two of the following criteria: information overload, rapid decision making and execution, and the need for collaboration."