Wolfram Alpha is based on a rather neat idea: making a website that can actually deal with information in an intelligent way, rather than simply search for words of things in existing pages. Put in ‘1 kg gold‘ and it will tell you that it would form a sphere 2.313cm in diameter, a cube 3.73cm to a side, and cost US$32,520. Put in ‘running 10 km/h 60 minutes 6’0″ 185lbs age 25 male‘ and it will estimate the number of calories expended. It doesn’t know about cycling yet, unfortunately. It doesn’t seem to be able to do calculations on greenhouse gas emissions yet, either, though it will tell you that mixtures of air and methane in which the methane is between 5% and 15% of the total will explode if exposed to a temperature of 595˚C. It also knows that Apple has 35,100 employees and a current P/E ratio of 23.1. It can search for base pair sequences within the human genome.
The biggest limitation of the site is phrasing things in a way it interprets properly. Indeed, most of the searches I try produce only the message: “Wolfram|Alpha isn’t sure what to do with your input.” For the time being, Wolfram Alpha is less of an open-ended vehicle for computations and data access, and more a set of discretely made tools for existing tasks. If you know which tools exist and how to format input for them, it works well. If you are trying to get it to do something its designers didn’t anticipate, it probably won’t work.
In short, Wolfram Alpha is a fun thing for statistics nerds to play around with, and could be genuinely useful for research. The best way to appreciate its current capabilities is to watch this introductory screencast.