IT LETS you check the weather or make an appointment simply by asking aloud, but is Siri, the 鈥減ersonal assistant鈥� on Apple鈥檚 newly released iPhone 4S, really such an advance?
Yes, says , an artificial intelligence (AI) researcher at MIT. He says Apple has created a 鈥渧ery impressive piece of engineering鈥� by combining established techniques from fields such as voice recognition and natural language processing.
, who researches machine learning at the University of Oxford, stresses that Apple hasn鈥檛 just put together existing techniques. But he has reservations: 鈥淭he difficulty is that each one of these systems makes errors, and when they are fed into each other the errors multiply.鈥�
Advertisement
Apple won鈥檛 talk about Siri鈥檚 underlying technology, though a patent application it filed earlier this year reveals that the software manages these errors by restricting queries to specific areas like dining or the weather. Apple calls such themes, for which Siri has access to databases of information, 鈥渁ctive ontologies鈥�. For example, the dining ontology contains databases of restaurants, cuisines and dishes, along with information on the concept of a meal 鈥� that it involves one or more people gathering to eat.
The active ontology idea is not new 鈥� , one of the inventors of Siri, formally defined it in 1995. What is unusual about Siri is that, unlike earlier grand AI projects, it is 鈥渧ery specifically focused on helping in particular domains鈥�, says Philip Resnik, a computational linguist at the University of Maryland in College Park. 鈥淚f you go out of those domains, all bets are off.鈥�
Siri listens out for keywords such as 鈥淢exican鈥� or 鈥渢aco鈥� to identify the subject area. It also works out whether to prompt for more information 鈥� such as what time to book a table 鈥� or whether it has enough details to access a reservations website and make the booking. This final step is possible because most web services now offer application programming interfaces (APIs) that let apps feed information to them. 鈥淭hat鈥檚 one of the reasons Siri is possible now when it wouldn鈥檛 have been five or 10 years ago,鈥� says Resnik.
The ability to make sense of requests phrased in ordinary language sets Siri apart from competitors such as Android鈥檚 Voice Actions, which requires commands in a certain format 鈥� saying 鈥渘avigate to鈥� will elicit directions, but 鈥渉ow do I get to鈥� ?鈥� will not. It doesn鈥檛 look as if Google is planning a Siri competitor yet. 鈥淚 don鈥檛 believe that your phone should be an assistant,鈥� said Andy Rubin, who heads Android development at Google, last week.
Siri will only get better. All queries users put to it are processed by Apple鈥檚 servers, giving the company a wealth of data it can use to improve the app. Katz suggests Apple could mine this data to discover commonly asked questions that Siri cannot yet handle. That鈥檚 simple enough, but what about asking it to 鈥渂ook a meal for my family when we鈥檙e all available鈥�?
聯Siri will only get better 鈥� all queries are processed by Apple, giving it lots of data to improve the app聰
鈥淪iri 2 might involve taking advantage of the fact that many of the tasks you attempt to solve have a social aspect to them,鈥� says Resnik. So, for example, the Siris on your family members鈥� iPhones could all work together to organise the meal.
Blunsom says Apple must try to keep expectations realistic, otherwise people might dismiss Siri because it 鈥渃an鈥檛 answer esoteric questions, despite the fact that it can find you a good sushi restaurant nearby鈥�.
