Back in June, I recorded a podcast with Tom Gruber of Siri. A week later I saw the company's 'Virtual Personal Assistant' put through its paces on an iPhone, and was impressed. Earlier this month I got on the phone with CEO Dag Kittlaus and VP of Engineering Adam Cheyer for an update, and today you can download Siri for yourself via Apple's App Store. Versions for Blackberry and Android will follow 'soon after,' and Kittlaus stresses that mobile is 'just the beginning.'
Today's iPhone app is the first consumer offering from a company that has spent a long time thinking about this space. Much of the core research resulted from the $150million CALO project at SRI, funded by DARPA. Siri itself emerged from SRI to close an $8.5million Series A round with Menlo Ventures and Morgenthaler back in October of 2008.
Explicitly described as complementary to web search, rather than a replacement for it, Siri seeks to move beyond a paradigm based upon keywords and links to embrace one that is personal, task oriented, and conversational in nature. Siri guides the user along a path, making query formulation iterative and relatively painless, and ensuring that the application gets the information it needs. Early use cases are optimised to 'help you get things done.' probably whilst mobile. You might, for example, ask (by speaking to Siri's Nuance-powered speech processing engine) for 'Sushi near work at 7pm.' That simple request is relatively straightforward for a human being to understand, process, and act upon, but requires a significant degree of intelligence on the part of a software agent. Where is 'work'? What is 'sushi,' and what do you actually want to do with it (find a restaurant where you can eat it, presumably)? When is '7pm'? Alternatively you could let Siri conversationally lead you through the 'right' questions to reach the same outcome, as the sequence of screenshots at the end of this post demonstrates.
At this point, it's worth mentioning that Siri (like so many location-powered applications on the Web, the iPhone, Android, or wherever) is currently only really effective in the United States. This is due to any number of factors, including consumer readiness and the easy availability of cheap yet comprehensive data, but the situation will
doubtless hopefully improve over time. Siri currently makes the point explicit, refusing to allow registration of a home or work address (or timezone) outside the United States. For the purposes of experimentation, my office has temporarily relocated to 1600 Pennsylvania Avenue, Washington DC, where there seems to be plenty of sushi available for tonight's dinner.
The speech processing works well on the whole, although I'm bemused that Siri interpreted my daughter's 'Where is the nearest Greek island?' as '18 me.' Whilst it's possible that this is an AI's attempt to avoid causing offence by responding 'Greece, you silly girl' it seems more likely that the engine got very confused by her non-American enunciation. Trying to sound like Hannah Montana just made it worse.
Even in the States, data is key to ensuring that apps such as this one deliver a rich and useful experience, as all the AI smarts and user interface polish in the world can't help an app that ignores the Starbucks across the street when you ask it to find you a coffee. Siri has lined up an impressive group of data providers including OpenTable, MovieTickets, TaxiMagic, Citysearch, Yelp, Yahoo Local, Gayot, Rotten Tomatoes, NYTimes.com, WeatherBug, AllMenus, StubHub, LiveKick, Maponics, Nuance and TrueKnowledge. Kittlaus celebrates the recent explosion of accessible APIs from sites such as these, claiming that Siri has acquired 'far more data than we've had time to integrate yet.' In a number of cases, revenue sharing arrangements mean that Siri gets a cut when money changes hands. A selection of test searches focussed on areas of the US to which I travel regularly delivered the sorts of results that I'd expect, and there's clear value in the integration of data from a number of different providers.
The Siri team looks forward to analyzing the logs once users start putting this app to work. Amazon's Elastic Compute Cloud (EC2) will handle the heavy lifting in the early days, allowing computing resources to scale with demand. Once the team has an understanding of real-world loading, Cheyer suggests that they'll pull much of the computing resource back in-house to lower costs. There is a clear expectation that Siri's responses will iterate rapidly as data become available to show how users use the app.
Further out, there's the ever-present need for more data. Kittlaus is also interested in increasing the opportunities for facilitating revenue-generating commercial transactions, and in allowing Siri to 'know you better.' Work, home, and current location is one thing. Why not favourite food, names and contact details for family (so I can have Siri 'tell my wife I'll be late home'), preferred airline, and more? It makes sense not to introduce these features from the outset, as consumers will need to both value and trust Siri before willingly giving up such detail. But you can be sure they'll be included soon.
Voice already plays a role on mobile devices such as the iPhone, perhaps most usefully in Google's search app. It remains to be seen whether consumers will really use two, or look for many of Siri's features to move across and enrich the voice-powered search experience they're already getting from Google, which presumably has many of the same data deals already in place.
Kittlaus stressed several times that Siri will deliver value on other platforms, suggesting a Siri email address (similar to email@example.com, presumably), a destination web site with which users might converse, or a Siri IM buddy that could be drawn into conversations.
By delivering value to users, and by building an ongoing relationship (backed by data) that's difficult to replicate, Siri seeks to offer a compelling and defensible business. Playing with the application from the other side of the Atlantic it shows clear promise, and I look forward to putting it through its paces on my next trip to the States.
And, of course, you can be pretty sure it'll run on the iPad.
This sequence of screenshots illustrates Siri's conversational approach to getting from my vague opening query about 'restaurants' to a reservation for specific people in a specific place at a specific time on a specific day. It would have been quicker to simply say what I wanted up front... but sometimes you just don't know until prompted.