- Views 26
With today’s introduction of Apple’s iPad 3 or iPad HD or whatever the hell they end up calling it, I think we’ll be entering a pretty Siri-ous phase when it comes to mobile Artificial Intelligence (AI) technology. Apple has a winner in Siri, its iOS digital assistant app, and knows it, so we’ll soon be seeing all-Siri, all the time in Apple products to come this year including, no doubt, Cupertino’s own big-screen TV. But this is not to say that Google’s Android will be far behind. There are stories popping-up about Google doing its own Siri-like app. But I expect Google to go significantly beyond Siri capability and I base that belief on the fact that Google has been working in this area since at least 2008, when they hired one of the scientists who did the basic research behind this category-shaping product.
Siri, you may recall, was spun out of SRI International, the big contract research outfit in Menlo Park, California. What became Siri began a decade ago as a DARPA project called CALO for Cognitive Assistant that Learns and Organizes. The five year, $200 million CALO project involved 22 organizations and was the largest non-commercial AI project in history. CALO’s job was to locate, relate, and find a way to use all the information in a military unit. It was a cyber Radar O’Reilly. The project was CALO and the application it produced was CALO Express.
But that wasn’t enough. CALO looked inside the unit but didn’t look outside. That part was iLink, which applied CALO technology to external networks eventually including social networks. With iLink, CALO knew not just about the military unit but also about the war and even about the world. Siri is CALO and iLink downsized to work with mundane information like finding a tow truck.
“Ah, but a man’s reach should exceed his grasp — or what’s a heaven for?” wrote the poet Robert Browning. What keeps Siri from being even more terrific (this is just my understanding, mind you, and may well be wrong — I am an idiot, remember) is this problem of reaching and grasping. When you have a bunch of potential correct answers to a natural language problem, how do you decide which one is most correct or even which ones are worthy of further consideration? One way is to very tightly define the subject area, which is why Siri doesn’t give relationship advice. But what if you actually wanted a Siri for relationship advice? Well that’s possible, too.
This solution harkens back to a previous column about Siri in which I compared it to the old Architext search engine technology from 1993. Architext just grabbed the N nearest web sites to the target question. Siri, which has to come up with just one answer, limits the subject matter enough so the problem is a lot easier. For harder problems you need something called the Minimum Bregman Information (MBI) Principle to both define the answer space and select the answer. This is where Apple and Google have to eventually go and where SRI has apparently already gone, reportedly thanks to a fine fellow there named Hung Bui.
Among the lead authors of those SRI CALO and iLink papers is Sugato Basu, who left SRI for Google Research in 2008. The presence of Sugato at Google (and his list of subsequent research papers) makes it virtually impossible that Siri was in any way a surprise to Google. They had to have seen it coming for at least four years.
So what’s going on here? Only Apple and Google can know for sure and neither is open to talking about this stuff, but I have a theory. I think Steve Jobs saw AI technology like Siri as key to future interfaces and future product categories including television (“I finally cracked it,” he told Walter Isaacson). Siri wasn’t really ready for prime time, but Steve was dying and couldn’t wait — hence Siri on the iPhone 4S.
Google was willing to wait. So when Google later this year rolls out its version of Siri, that Google product will be BIG Siri, fully MBI-enabled, tightly linked to Google data (not Wolfram Alpha), and able to take on a far broader range of topics, like how to balance the SU carburetors on your old MG sports car.