Apple’s new “intelligent agent”, Siri, is taking a page from Wolfram Alpha and trying to answer questions in a funny manner – leading to an amusing blog entitled Shit That Siri Says.
Intelligent agents have been a dream for a long time. And by “dream” I mean “not reality”. Consider SatireWire’s classic Interview With The Search Engine.
Now, here’s a thing. Ask Jeeves launched in 1996 or so. That interview was probably closer to 2000. So… let’s just compare hardware for a moment, shall we?
In 2000, Intel’s processors were the Pentium III, Pentium III Xeon, and Celeron, roughly. The exact ranges varied. Realistically, you could probably have a 600MHz processor by then. Or a couple of them, in a very expensive server.
In 2011, the iPhone 4S, which is what Siri is really aimed at, is coming out. It’s got a dual-core “A5” processor running at, rumors tell me, around 800MHz.
Now, MHz aren’t really an apples-to-apples comparison, but in a lot of ways, the advantage is often with modern hardware, even when that’s specialized low-power hardware. It’s a tradeoff, but there’s certainly a reasonable comparsion to be had. Which is to say. That five ounce phone is, quite likely, faster than the servers with which Ask Jeeves was running even in 2000. In 1997, they’d probably have been running with a Pentium Pro or maybe a very early Pentium II, at maybe almost as much as 300MHz.
Just think about that for a bit. You know why phones and tablets are so good now? Because they are dramatically faster than the machines you got used to. And that’s why intelligent agents, like Siri and Alpha, are starting to be only sometimes funny, and are actually sometimes useful now – they have enough speed and power to do things like “keep a bit of context so their answers aren’t completely illucid”, and they’ve got a lot more information, and power to process it, than they used to.
I’m still not entirely equipped to handle how fast things are these days. We are living in the era of infinite computing power. Getting used to this, and its implications, takes a while.
From: Dave Leppik
Date: 2011-10-14 16:06:08 -0500
Except that much (if not all) of Siri runs server-side. They’ve limited it to the 4S (perhaps because of the beta nature of it), but that might be to limit the beta test group. (When Siri was an independent company, they had an iPhone app.)
My guess is that certain queries run client side, e.g. “make an appointment.” At least that’s what it should do, if they recognize that certain things ought to be done without an Internet connection. But for open-ended questions and ones where the answer requires the Internet, there’s no reason to do the processing client-side.
Date: 2011-10-14 20:10:05 -0500
Question for me is, what exactly are they sending across the virtual wire? Are they sending the raw speech, to be analyzed on the far end? Are they running the speech recognition locally, and just sending the text query?
If there’s any queries that can be handled locally without an internet connection at all, how far does this boundary reach?