Wired magazine has an exclusive about Viv, a new product from the creators of Siri, who left Apple after Steve Jobs died. Viv is supposed to deliver what Siri promised, and more: a ubiquitous digital assistant able to answer complex vocal requests such as "Find me a flight to Dallas with a seat that Shaq could fit in," the article says, with a nod to former basketball giant Shaquille O'Neal. While the dream of artificial intelligence lives on, implementations will keep suffering from a fundamental disconnect between how software engineers think and the way the real world works.
In February, Apple persuaded a judge to dismiss a 2012 lawsuit accusing it of deceptive advertising, arguing that "Apple made no promise that Siri would operate without fail" and that a reasonable consumer couldn't expect commercials to "depict failed attempts." By any other standard than Judge Claudia Wilken's, Siri has been a failure. Apple added it to iOS7, the current version of the company's mobile operating system; a survey of U.S. iOS7 users last year showed that 85 percent hadn't even tried it, and almost half of the remainder judged it a disappointment.
The problem with Siri is that she's not that smart, and she's not getting smarter. In 2011, soon after Siri arrived as a built-in service on the iPhone 4S, Matt Honan complained on Gizmodo.com: "I love John Coltrane. I've got many of his songs in my music library, yet when I ask Siri to "play some Coltrane" (just like in the commercial!) it tells me it can't find any `coal train.' Is that my fault, Siri? Or do you just not do homophones? Are you not intelligent enough to contextualize sound?" Three years later I tried the Coltrane test -- and got the "coal train" response, too. Meantime, Google Now on my Samsung phone happily identified the jazz master.
Siri's founders, who sold the service to Apple for $200 million in 2010, know where they put the blame. "I do feel if Steve were alive, I would still be at Apple," Wired quotes one of them, Adam Cheyer, as saying. They weren't, however, and still aren't the only team working on outsourcing secretarial work to artificial intelligence. Microsoft's Cortana and Google Now are at least as good as Siri, and better at some tasks, but they still don't cut it.
You have to speak as if to a child, and in your best, most grammatically correct English. Google Now will respond correctly to "Check in, FourSquare" but not to "FourSquare Check in." They're all awful at foreign accents and the realities of life outside the U.S. I asked my Android about the next train from U-Bahn station Schwarzkopfstrasse in Berlin, where I live, and it did a web search for "the next Lebon train from Schwarzkopf Professional." Siri searched for "the next train from bunch watch Ghostbusters."
Enthusiasts claim that you have to be patient while artificial intelligence systems learn. As time passes, they get better at recognizing your speech. Also, as data accumulates from a multitude of users, they gain a better understanding of what might be required. Most people, though, demand immediate results; anything less means they stop trying and revert to previous methods.
Judge Wilken might say I'm demanding too much from what is just a computer program, not a real assistant. Well, the Viv team says their new creation, accessible through the cloud to every application and service on earth, will accelerate the learning process to achieve an uncompromising imitation of a human assistant -- only much smarter and faster.
Wired gives an example of "someone unsteadily holding a phone to his mouth outside a dive bar at 2 a.m. and saying, `I'm drunk`." Without elaboration, Viv would contact the user's preferred car service, dispatch it to the address where he's half passed out, and direct the driver to take him home." That's still the system's future, and it's another example of software-engineer thinking vs. real life. What if the drunken phone owner wants to hear his favorite lullaby, get a cup of coffee or call an ex-girlfriend rather than get a taxi home? What if he wants a taxi, but isn't headed home?
To an engineer, real-world interactions are commands. They're not, really: both verbal and non-verbal communication carries too many undercurrents of meaning. Some invaluable human assistants know how to interpret these requests; others are good at more straightforward stuff, but even that requires being able to figure out exactly what's required. While we'll keep hearing about increasingly intelligent machines, there's no reason to be scared since they won't have our kind of intelligence; they're just Siri with different names.
To contact the author on this story:
Leonid Bershidsky at firstname.lastname@example.org