An interesting little piece in The Guardian about a competitor to Siri, the voice assistant on iPhones from Apple. It's called Evi and comes from a small company in the UK:
Apple's Siri has a new British rival – meet Evi
Cambridge firm True Knowledge outsmarts tech giant with phone assistant that copes with variety of UK accents.
I'm not at all surprised that a UK developed software engine deals with UK accents better than one developed in the US does. I'm also not surprised that a company not intimately tied to Apple releases its software on both Apple and Google Android hardware.
The interesting little bit to me is the way that they've taken rather different architecure choices. Siri doesn't, really, store any information herself. Not even on the servers at HQ. Siri is really an interface which, when asked a question, then interrogates other sources for answers to that query. Those other sources are the usual search engines, Wikipedia and so on.
Evi works slightly differently. Sure, the ultimate sources are pretty much the same but they upload the information into Evi. So Evi is really interrogating herself in order to answer questions.
Now there is a bit of blurring here: it's going to be difficult to state that Siri stores none of this information when she obviously does given the existence of caching.
Which architecure is best though? I have a feeling that it rather depends upon what you mean by "best". In a static world I can imagine that Evi is. For whatever data is in there has at least been checked before it has been loaded. In a dynamic world it's a little different. If the universe of knowledge is expanding no faster than it's added to Evi then Evi might still come out ahead.
However, if your aim is to be able to add new territories, entirely new languages, as fast as you can then I can imagine that Siri's model is better. For you just link up Siri to the usual data sources in the new language and you're good to go. You don't have to do what Evi does:
While Siri relies on external databases for answers, such as the online encyclopedia Wolfram Alpha, Evi has a painstakingly amassed collection of 635 million facts on 28 million things. Her creator, the Cambridge start-up True Knowledge, is adding to the database all the time by scouring Wikipedia, Yellow Pages and other sources.
To get, say, a French language Evi you've got to start all over again. To get a French language Siri you just direct her queries to the relevant French search engines.
Further, if it turns out that Evi is really the right way to go then of course you can do that with Siri at some future date. After you've stopped people creating local language versions of Evi by being first into that market, something you can do much faster with Siri than you can Evi.
On the other hand there's the entire Android universe to conquer with Evi so perhaps they aren't worried about Siri at all. All the best of luck to them of course but I do wonder whether that structural issue is going to limit their expansion ability.
