by Tom Kaneshige

Siri’s Steep (and Long) Learning Curve

Jun 15, 20124 mins
Car TechConsumer ElectronicsData Center

Siri, the iPhone's voice-enabled personal assistant, has a big problem: It doesn't really work the way Apple says it does.

It’s graduation season, and teenagers across the country are donning caps and gowns and walking onto center stage to receive their diplomas. They should be proud of their achievements after years of study.

Too bad Apple’s pride and joy, Siri, won’t be graduating anytime soon.


“If Siri worked, it would be great—but it doesn’t work,” says Kyle Wiens, an iOS app developer and co-founder of iFixit. “It’s a hard problem. If Apple is lucky, five years from now we’ll say they solved it.”

Billed as a virtual personal assistant, Siri, the voice-enabled artificial intelligence engine baked into the iPhone 4S, made a dramatic entrance eight months ago. At Apple’s Worldwide Developers Conference (WWDC) this week in San Francisco, Apple claimed Siri has learned quite a bit since then, particularly in its ability to more accurately answer questions about local sports, restaurants and movies.

Automakers such as BMW, GM, Toyota, Audi, Honda and others are gearing up to integrate Siri into cars in the next 12 months. A button on the steering wheel will turn on Siri for eyes-free communication.

Siri will also be available on iPads, Apple said at WWDC.

So Apple has big plans in store for Siri. The idea is that it will become the preferred way people interact with computers and replace text-based Google searches. At the AllThingsD D10 conference earlier this month, Apple CEO Tim Cook called Siri “profound” and said Apple is “doubling down” on the technology.

Slideshow: Apple Innovation: 10 Future Tech Ideas

But Wiens isn’t convinced. “I think Apple is getting ahead of itself,” he says.

Apple’s claims of a “significantly enhanced” Siri at WWDC comes amidst a recent spate of lawsuits claiming Apple over-sold Siri’s abilities in television ads. Indeed, YouTube is awash with videos of iPhone 4S users trying to replicate Siri’s responses on the television ads with little success.

Siri’s problems can be traced back to a combination of spotty server availability, big expectations, and a steep learning curve.

On the server side, Siri processes requests on Apple servers, not locally on the iPhone 4S. This means Siri needs a fast connection and available servers. When Siri says that it cannot help with a request, it’s usually a network connection or server problem.

“Apple has had massive data center scalability problems,” Wiens says.

On the consumer expectations front, Siri faces an uphill battle courtesy of Apple itself. Apple’s television advertisements show actors such as Samuel L. Jackson and John Malkovich having casual conversations with Siri and asking off-beat questions, to which Siri responds quickly without a glitch or pause.


All of this makes Siri look on par with Star Trek’s fictional talking computer—an unrealistic expectation.

Natural language recognition and true artificial intelligence (AI) understanding as depicted by Star Trek’s computer isn’t easy, says Lars Hard, founder and CTO of Expertmaker, an AI software company based in Sweden. Such a system would likely require an interconnected network of several AI systems.

Blog: iPad’s Starring Role in Sci-Fi Movies

“The available capabilities of AI today are related to very specific functions, such as making a reservation at a restaurant or booking transportation,” Hard says. “Going beyond that, into any knowledge domain, requires vast amounts of general data and domain-specific information.”

CTO Lars Hard of Expertmaker

Siri’s education is in the early stages. Apple admittedly released Siri as a beta, in part because voice recognition software requires a large data set—that is, people have to use Siri so that it can learn and become better via a feedback loop.

“Problems arise when attempts are made to add a learning function for more open-ended knowledge domains,” Hard says. “A general model with those capabilities does not exist yet. Human knowledge and experience is so extensive that we will not see the beginning of real machine learning for a few more years.”

In other words, it’ll be awhile before Siri comes close to meeting lofty consumer expectations and speaks to you like it speaks to Malkovich.

Tom Kaneshige covers Apple and Consumerization of IT for Follow Tom on Twitter @kaneshige. Follow everything from on Twitter @CIOonline and on Facebook. Email Tom at