Wired has a nasty tendency to see incremental improvement and say "quantum leap." Check this out: (This article) (Wired, October 2011) What's interesting is what they don't say. All this stuff started with Semantic Forests, an NSA patent on contextual analysis filed in 1997. From there, it branched out to Total Information Awareness, the DARPA program that yielded CALO from SRI, where one of the founders cut his teeth. At this point things split - DARPA spun off CALO which eventually yielded Siri and now Viv. The useful bits, however, got rolled back into the Information Awareness Office, most specifically EELD. EELD had, in 2002, the ability to read a newspaper and use it to predict (with enough accuracy to make the NSA happy) what individuals were likely to do - it was an attempt at psychohistory. EELD, of course, ended up rolling into PRISM as Semantic Forest was rolled into ECHELON before it. So what you're looking at here is a method of AI that either parallels what the NSA is doing... or was rejected by the NSA in favor of a bigger approach. I lean towards the latter because the NSA has never seen a problem it can't throw more computers at. Which brings us full circle: Google's approach, the NSA's approach, Microsoft's approach has always been MOAR DATA. The idea being that the mean of a crowd will give you a closer approximation than any one individual. Google didn't run out Google Voice and Google 411 'cuz they were being nice; they wanted a bunch of data points as to what words sound like so they can parse any dialect. Apple is a decade behind on data collection when compared with Google so they have to go a different way. They need something nimble that does the heavy lifting on your phone rather than back on the server. Will they succeed? Well, Siri is a piece of shit. I think that's pretty much universally acknowledged. Siri Mark II is likely to be less of a piece of shit, but still a piece of shit, just like Apple's maps no longer require you to take an offramp that doesn't exist but will still get you lost 9 times out of 10. Either way, the article is about a refinement, not a revolution, but I guess that makes for boring copy. didja notice how their infographic invokes MAPQUEST? srsly? when was the last time you got a useful result out of Mapquest? “The vision is very significant,” says Oren Etzioni, a renowned AI expert who heads the Allen Institute for Artificial Intelligence. “If this team is successful, we are looking at the future of intelligent agents and a multibillion-dollar industry.”
Rather than just issuing the app commands or Google-style search phrases, you interact with it through conversation. Saying something like “I’d like a table for six at Flour and Water” would prompt the app to make a reservation using OpenTable. And if you haven’t provided enough information for it to complete a task, it will prompt you to elaborate. Siri then uses information about your personal preferences and interaction history so it can better accomplish specific tasks. As you use it more, it learns your preferences and improves its performance.
Their geocoding service is much better than Google's if you have messy/incomplete/misspelled addresses. I've never compared their route finding, but to the extent that knowing where you're going is a requirement for finding a route there, and they're getting the addresses via speech recognition, Mapquest might very well perform better. Also, Google's license terms forbid you from using the maps API to do anything but draw maps. Mapquest's is much more permissive.didja notice how their infographic invokes MAPQUEST? srsly? when was the last time you got a useful result out of Mapquest?
Full disclosure - a good friend of mine works for Mapquest. I really do wish them well. By contrast, my feelings for google are similar to my feelings for Goldman Sachs: they are a vampire squid bent on getting rich off the backs of the proletariat. That said, Google is Goldman Sachs. Mapquest is my friendly neighborhood credit union.
I had to do a retake when I saw Mapquest there. Really? Part of the reason why I'm not yet convinced, the other part being Wired's overhyping. I do think this project has merit, if only because it might produce a working result that for once depends on efficiency and data control instead of data gathering.
Yeah, not to mention the killer query that's brought up ("what is the population in the city where Lincoln was born?") is already possible on WolframAlpha. Quantum leap my ass.