Search is more than skin-deep. To most users, a search engine is only skin (or GUI) deep. And anyone who's taken Wolfram for a spin has judged it based on the results they get back. In a few cases, Wolfram's abilities are quite impressive. But that's not what makes Wolfram|Alpha important. For that, we look to what Stephen Wolfram has done with the entire concept of interpreting and analyzing information. Wolfram|Alpha doesn't search data, it calculates it. That's a fundamentally important distinction.
Unlike Bing, which is promising a revolution that barely qualifies as evolution, Stephen Wolfram knows this is the first step on a long, long road. He says so right on the home page: "Today's Wolfram|Alpha is the first step in an ambitious, long-term project to make all systematic knowledge immediately computable by anyone."
Words are not enough. Wolfram's previous work with Mathematica and NKS (New Kind of Science) shatters the paradigm that every search engine is built on, semantic relationships. As revolutionary as Google's introduction of the linking structure of the Web as a relevance factor was, it was added to a semantic foundation. PageRank is still bound by the limits of words. And words are slippery things to base an algorithm on.
The entire problem with words is that they're ambiguous. The word "core" has 12 different dictionary definitions. It's very difficult to know which one of those meanings is being used in any particular circumstance. Google and every other engine is limited by its need to guess at the meaning of language, one of the most challenging cognitive tasks we encounter as humans.
Potential advancements in relevance require gathering additional signals to help interpret meanings and reduce ambiguity. Personalization is one way to do this. Hunch, Aaron's nominee for the iPhone of Search, requires you to fill out a long and rather bizarre quiz about your personal preferences. All this is to learn more about you, making educated guesses possible. If you're going to stick with a semantic foundation, personalization is a great way to increase your odds for successful interpretation.
Another way to interpret meaning is to go with the wisdom of crowds. By overlaying the social graph, you can make the assumption that the one meaning people like you are interested in, is also the meaning you might be interested in. Again, not a bad educated guess.
Knowledge as a complex system. But what if you could do away with the messiness of language entirely? What if you could eliminate ambiguity from the equation? That's the big hairy audacious goal that Stephen Wolfram has set his sights on. If you look at the entire body of "systematic knowledge," you have a complex system -- and in any complex system, you have patterns. Patterns are abstractions that you can apply math against. In effect, knowledge becomes computable. You don't have to interpret semantic meaning, which is intensive guesswork at best. You can deal with numbers. And unlike language, where "core" has 12 different values, the number "3" always has the same value.
Wolfram|Alpha is not important because it provides relevant results for stocks, cities or mathematical problems. It's important because it's taking an entirely new approach to working with knowledge. It's not what Wolfram|Alpha can do today; it's what it may enable us to do tomorrow, next year and in the year 2015.
Wolfram|Alpha could change all the rules of search. Keep your eye on it.
Wolfram will never be more than a niche search engine, if that, unless it gets bought by Google, Yahoo or Microsoft.
Wolfram Alpha is indeed revolutionary but if you just jump in and try it as a conventional search tool you will be frustrated. To get an idea of how amazing it is, watch Wolfram's narrated examples. http://www89.wolframalpha.com/screencast/introducingwolframalpha.html
I played with it first, got annoyed, watched the video and am astounded. Everything you say Gord is true, this is incredible.
The main problem with search is that people lack facility with language, both per se, and specifically in computational contexts. This is a problem best solved by teaching people, not machines.
You don't type 'core' into Google to find authoritative info on apple Corers in the 18th Century American kitchen. The problem here is not with the ambiguities of language. It's with the searcher providing insufficient surround terminology to isolate desired results. The solution is to sit every kid down, starting shortly post-infancy, for a little class in database-query formulation (something that would be huge fun for the kids, for the teacher, and would have enormous side-payoffs in logic, problem-solving, math and science -- you could even use Google for this if you could reliably block adult content).
Once the basic mechanics of query-formulation are understood, greater facility with search comes from attention to written language: sensitivity to style, voice, context and epoch of communication. Understanding that, at the simplest level, academics in the social sciences (for example) phrase things differently from lawyers, who phrase things differently from right-wing bloggers, who phrase things differently from early 19th-century military theorists is a huge help in figuring out phrases that neatly, quickly return exactly the information you need.
IF that information is available. The other big problem with search is that people game it and fail to serve its needs, both for commercial reasons, and out of inertia. Imagine being able to sit down at a browser and enter 'Porcher Veneto toilet flush assembly' and actually getting results that put you a click and a credit-card away from buying. But you can't, because Porcher (a division of American Standard) has a deal with its resellers not to maintain a parts-site online, and most of its resellers - if they maintain websites at all - just post .pdfs of the Porcher parts catalog, unsearchable.
In this context, Wolfram Alpha takes exactly the right approach. It ignores the first problem in favor of tackling its second order: what happens when a smart person needs to compile data from diverse sources and create a single picture granting additional insight? And as your article suggests, it works by deducing and matching a set of known semantics (e.g., this is a stock price) to isolate data, then does useful and interesting things with the data -- the vocabulary of semantics and the pallette of 'useful and interesting things to do with data' will both grow with use and time.
However powerful this gets, however, it's not going to revolutionize search except for the small minority of people for whom search works fine already. People who type 'core' when looking for historical information on kitchen implements are still going to be out of luck. And we're all going to remain frustrated searching for toilet-parts until a combo of RFID and supply-chain automation and sensor networks and other technologies finish revolutionizing the economy of things, and every nickel washer in America has its own personal web page (and blog!)
Agreed, Wolphram is hugely important and will lead to further advances in search. Those who use it as a search engine will be frustrated and won't go back. Those who understand the value will use it as intended, for computing knowledge. Can't wait to see where it leads.