Taking a first bite out of Wolfram Alpha

The knowledge engine that knows the answer is 42

Is it the newest rival to Google, likely to knock Google off the top spot? No it isn’t. Does it provides a single answer to complex questions – unlike traditional search engines? Nope. Could it possibly be "a natural search engine"? Not quite.

The newest game in town, if Twitter is anything to go by, is the Wolfram Alpha computational knowledge engine. There is a good chance that in the long term it will revolutionise how we interact with the internet. But not yet. Right now, it is a very good start, and belongs firmly in the area loosely described as semantic web development.

The greatest threat to its success is likely to be the enormous hype that has broken out across mainstream press and media this last weekend. If Wolfram Alpha engineered this bubble, then there is at least a minor question mark to be raised over the strategic nous of their PR department. For if any one thing is certain, the non-specialist press just love to tear apart new technologies the moment they fail to live up to claims they never made in the first place.

The Wolfram Alpha engine is the brainchild of Stephen Wolfram, renowned physicist and CEO of Wolfram Research. In addition to his work on complex systems – a field in which he founded a research centre and journal in 1986 - he was also responsible for the development of a technical computing system known as Mathematica.

Until recently, this was the mainstay of the company, counting millions of users worldwide, and applied to a range of technical problems, such as trading systems, pricing and optimisation models and Monte Carlo simulations.

This system now underpins the Wolfram Alpha engine, ensuring that at a technical level, it is reasonably stable. The engine contains a number of key components. At the front end is a natural language interpreter, allowing users to feed the system questions in an approximation of English.

Behind this reside a number of key data sources which have been captured and standardised by Wolfram staff over many years. The number of data sets included are, at present, limited to a few hundred, although, as a spokesman for Wolfram pointed out, as one data set includes "all current and historical weather", whilst another includes "the English language", this measure may be deceptive.

Feed the Wolfram Alpha engine a question, either in natural English or using a simpler set of search terms, and it will attempt to come up with information that provides an answer – or answers – appropriate to the query. A major part of Wolfram Alpha’s function relates to its ability both to disambiguate unclear terms and to make a stab at providing an answer relevant to the likely cultural stance of the individual asking the question.

Feed it "gold" and "lead" and it is likely to come up with answers that relate to the physical properties of metals: give it "gold" and "red", and material relevant to colours will appear. Feed it "Cambridge" from a UK-based IP address, and data will be returned relevant to the Fenland City of that name: do so from a US-based address, and the Massachussetts Town of that name will appear.

"Rabbit", in the UK, should return data on the animal both as biological entity and source of food: in the US, according to Wolfram Alpha, the second level of meaning will be overlooked, as Americans simply do not eat rabbits.

So much for the product. Wolfram Alpha is very clearly a knowledge engine that is heavily biased towards drawing down data around topics it has already included in its underlying database. Unlike Google, it is not setting out to rule the web. Its commercial model envisages four strands to future revenues: partner licensing, major brands sponsorship, plus professional and corporate versions. Cost structures are not yet determined, but one suggestion is that companies could end up paying $10 per employee for a license.

It is also envisaged that companies might decide to incorporate into their private version of Wolfram Alpha data sets that require separate licensing (such as selected geodemographic directories) or their own private data sets. An insurance company might wish to load dates and locations of claim by type of event, a credit card company might want to incorporate purchase data, and so on.

In a seriously original debunking of the product, the Guardian complains that "Wolfram Alpha has trouble answering that enduring philosophical question: 'Why am I here?'", responding merely "Wolfram Alpha isn't sure what to do with your input."

It does not know the airspeed velocity of an swallow – laden or otherwise. However, Wolfram Research have bowed to the inevitable by allowing their engine to answer questions about the meaning of life (the universe and everything) with the answer, "42". Although the engine itself is data-focused, it will also return links to the most relevant web sites in respect of questions submitted.

Stephen Wolfram says that he is "not keen on the hype". Wolfram Alpha sits smack in the middle of fast-expanding research into the field of semantic search, whose goal is to turn interaction with the internet into something much closer to the intelligent man-machine dialogue exemplified by HAL 9000 in Stanley Kubrick’s 2001. That study, at present, is focussed on issues of disambiguation and the creation of "ontologies" – the re-structuring of data sets to align them more closely with the information requirements likely to be placed upon them.

In those aims, Wolfram Alpha looks like being a resounding success. ®