The two main focal points for the hype around Wolfram Alpha have been the notion that it's a rival for Google, and the claim that it represents some significant development in Natural Language Processing (NLP), neither of which are strictly, or even slightly, true.
First of all, and perhaps unsurprisingly, the application isn't a serious rival to Google at all - it doesn't actually search the web. It rather provides a sophisticated interface to a collection of data, much of which seems to relate mainly to academic science. The data presented and used by the system is chosen and prepared by people (i.e. humans!) involved in the project.
What the program does that is interesting, is facilitate a level of interaction with the data, using computation/ calculation to create a new way to engage with the 'knowledge' contained within. Whereas Google provides an interface to the Web itself, Wolfram Alpha will be more like an interactive encyclopedia, a source of information in its own right, albeit one that is available over the Web.
As for NLP, far from representing some advance, the Wolfram Alpha project seems little more than dismissive. The approach sees language as a necessary inconvenience rather than any sort of focal point for its model of interaction. With Stephen Wolfram referring to 'linguistic fluff', it's clear that dealing with real natural language is no priority.
The program relies on the fact that humans are accustomed to tailoring their input to a computational model (such as search engines, with which we all now instinctively reduce our queries to conceptual terms), rather than attempting to replicate human linguistic interaction - which is what the field of Natural Language Processing is all about.
So, it sounds as though it'll do some pretty exciting things, just not what the hype dictated, but you have to grab the headlines somehow.
Written by Sue Smith