Is Siri Smarter Than Google?

storagedude writes "Google could go the way of the dodo if ultra intelligent electronic agents (UIEA) make their way into the mainstream, according to technology prognosticator Daniel Burrus. Siri is just the first example of how a UIEA could end search as we know it. By leveraging the cloud and supercomputing capabilities, Siri uses natural language search to circumvent the entire Google process. If Burrus is right, we'll no longer have to wade through '30,000,000 returns in .0013 milliseconds' of irrelevant search results."

Yeah but?

[BradyB]

Didn't even Wozniak say that Siri isn't as good as the advertisements?

Steve quoted on various news sites:

I have a lower success rate with Siri than I do with the voice built into the Android, and that bothers me. I’ll be saying, over and over again in my car, ‘Call the Lark Creek Steak House,’ and I can’t get it done. Then I pick up my Android, say the same thing, and it’s done. [...] On the 4S I can only do that when Siri can connect over the Internet. But many times it can’t connect. I’ve never had Android come back and say, ‘I can’t connect over the Internet. [...] Plus I get navigation. Android is way ahead on that.

Re:Wait a minute

[Bruce Perens]

Right. As you and the other two have reminded me, this should have been "The search engine companies can't put their ads in there without paying Apple". And you can imagine that any constraints and regulation that are put on Google will make their way to Apple eventually. Will this protect the users? Absolutely not. Nothing can protect Apple users, because the problem is protecting them from themselves.

Re:Wait a minute

[Bruce Perens]

The difference is Siri gives you the answer to your question if it can. Regular search gives you a list of web pages that may or may not have any relevance.

You're not keeping up. Go on google and type "UA 647". You will see the flight status, properly formatted, right at the top. There are a significant number of questions that are answered this way, and it will only increase.

Re:Simple answer: no

[Anonymous Coward]

It is important to note that the submitter is a asshat retard as are slashdot editors for allowing the submission.
Once the asshattedness improves we will no longer have to wade through _irrelevant post results_

Re:Is she?

[Pf0tzenpfritz]

"I liken biggen assen and I cannot lie."

gern geschehen.

Re:Is she?

[Paradise Pete]

the cherry-picking to which he's referring is not done by Siri, it's by the original poster in choosing that example.

By the way, the answer that *I* get from Siri on that is "This might answer your question: (data not available)". I like Siri a lot and use it every day, but come on. It ain't that smart. And in many places outside the US its usefulness drops off markedly. If I ask it for a place to eat, for instance, it says "Sorry Pete, I can't look for restaurants there."

Re:Is she?

[Vintermann]

Try googling "plane map".

By comparison, try asking siri "which ships are in port right now?" and googling "ship map".

Consider the possibility that you're impressed by hardcoded displays, rather than sophisticated algorithms.

Re:Voice recognition

[kurisuto]

Sorry, but this is bull. Your statement that "voice recognition is at its limits phonetically" is just wrong. I work in the voice recognition industry, and in the past five years, I've seen the recognition error rate markedly and measurably go down, and this trend is continuing.

There are actually two kinds of models involved in voice recognition:

1) the acoustic model (which has to do with looking at a sequence of time slices of the acoustic signal and working out what sequence of phonemes could most likely have given rise to it). You say that voice recognition is at its limits phonetically, but these models are actually getting better over time with larger sets of training data, and the improved models measurably result in a lowered word error rate.

2) the language model (which has to do with specifying which words exist, and in what order they are most likely to occur). These language models can be very simple, as in the case of a yes/no question in a phone-based app (your model might accept "yes" and "yes ma'am", but not any arbitrary English utterance); or they can be very large, as in the case of a general-purpose dictation application.

In conjunction with the recognizer, what these two models give you is a raw string of recognized words. What sort of processing you do on that string is a separate question. There are obviously all sorts of things you can do with the string. The parsing and processing techniques are getting more sophisticated, and are getting integrated with other systems in interesting ways. This is largely a separate question from the accuracy of the string itself, which is the output of the recognizer (I say "largely" because your application might activate a different language model based on the current context, which does affect recognition accuracy).