The company is called Siri, and it's so serious about it's stealth status that it's even registered the domain stealth-company.com. We chatted today over the fact that they've announced their $8.5 million Series A with Morgenthaler and Menlo Ventures.
Still, even with everything I know about the project, it's difficult to frame it in just the right way without revealing more than I'm supposed to. Rather than focus on precisely how this works, I'll talk a bit about the project from which the company originated. It's a small outfit you may have heard of called DARPA, also known as the place where the Internet was conceived.
What Siri is Up Against
One of the holy grails in AI research, something I'm pretty familiar with from my own personal experience, is putting together a personal assistant that is both friendly, useful and reliable. There are a couple of factors that tend to impede these well-meaning projects from being actually useful. The first part is the semantic interpreters - the English language is a tough thing, and so far most projects still function on the Turing principal, which is more about fooling judges into thinking the machine is sentient rather than attempting 'actual sentience.'
As a result, most personal assistants will break down under heavy usage either revealing their limitations or by the other major shortcoming common to these projects: inefficient and frankly 'dumb' information storage methods.
Most of the chatterbox AIs, for which most personal assistant AIs revolve around, function off the Eliza methodologies (or if you need something more recent, Alice). In Loebner/Turing tests where the job is to produce good conversations, their information storage methods work wonderfully. When it comes to specific information storage and retrieval, asking the same question with slightly different wording can produce wildly different answers.
That's Quite a Feat
These were the sorts of technical hurdles that CALO faced when they were assigned $200 million and a mandate to make AI work. These issues have been present in every AI that's attempted to tackle the personal assistant problem for the last thirty years. There have been one or two exceptions, but no breakout stars have emerged into the public eye.
So much of what we have in Web 2.0 and social media is waiting to be used in this way. Every tool we interact with daily has an API, and we've been manually hooking them into just about every service and social network we've signed up for in the last six months. Back last November, I jokingly suggested some potential definitions to the successor technology for Web 2.0, and referenced Eric Schmidt and Ken Rutkowski's Web 3.0 definition: "applications that are pieced together."
Chances are, you don't need yet another search engine to go out and find content on the web. You need bridges between your own content and communications. You need your calendar to be aware immediately when you schedule a meeting via your email. If there's a scheduling conflict, it wouldn't be completely unwelcome if your personal assistant contacted you and the other participants to find a common time that works for everyone.