What is Wolfram|Alpha and Why is It Important?

Every once in a long while, a new internet service is born that has the potential to revolutionize the way we do things online. Examples include Google for finding web pages, Facebook for keeping in touch with your friends, Wikipedia for collaborative knowledge sharing, and YouTube for the way we consume video media. The recent launch of Wolfram|Alpha is has the elements going for it to just maybe become another such innovation.

WolframAlpha is the brainchild of Stephen Wolfram. Wolfram is famous for developing the Mathematica computational software package which helps researchers and scientists more easily work with advanced computational models. He also wrote the book A New Kind of Science, which attempted to make the science of chaos theory digestible to normal humans (he didn't really succeed here imho) - in this book he draws parallels between complex computational systems and nature, and argues that much can be learned about natural systems by analyzing computational patterns. His thesis draws from and is related to the field of fractal theory.

What does WolframAlpha do?

WA is an answer-retrieval system much like Google, but instead of helping you find web pages as Google does, WA provides you with direct answers to specific queries. For example it will answer math questions (like "134 + 123 * 993") or statistics (like "mean {3,6,4,10,20}"), nutrition (like "vitamin A in one lb lettuce"), geography (like "zip 60637"), and many other categories. It's actually very similar to Google shortcuts (simple answers provided for queries like "weather new york" or "3 pounds to kilograms"), but greatly expanded in terms of query types and data sets available.

How does it work?

WA has a huge database containing sets of data of many different types, from global population data, geography, units and measures, physics and chemistry data and formulae, life sciences data, and more. In fact they claim to have 10 trillion pieces of data, 50,000 types of algorithms and models, and linguistic capabilities for over 1,000 domains, all running on a cluster of supercomputers which can run computations and return answers in a matter of seconds.

How well does it work?

WA has an intuitive and clean interface, and the site comes with plenty of examples to help you get started understanding which types of queries it can handle. From a quick set of tests, I found it to have useful and detailed answers to many queries, but still many cases when it could not understand my query (it couldn't answer "average home price san francisco"). In its current form, I don't see enough differentiation for an average user to switch over from Google (Google handles many of these types of queries quite well). But for an advanced user who needs advanced statistics and data sets from various fields of science, I'm sure it will become an essential resource.

Does WolframAlpha fulfill the promise of natural language search?

Natural language search is often thought of as the holy grail of internet search, a search engine that can understand and provide direct answers to human-language queries like "How many active volcanoes exist below the Earth's equator?" Many approaches have been developed to try to solve this problem, from algorithmic approaches like PowerSet to human-powered sites like Yahoo! Answers. Although many of these sites have made improvements in the way we can find information, the promise of natural language search remains unfulfilled.

It's my opinion that the ultimate solution to the natural language search problem will be found in between the separate spheres in which humans and computers communicate. What I mean by this is that computers will get better at understanding human language, but humans will also become more adept at speaking computer language. You can see it already happening in how we use Google. Many of us don't realize it, but we've all adapted some new patterns in how we communicate with Google. We use a particular syntax which we have learned will return us the best answers, like "hotel recommendations paris" or "history christmas wikipedia" to help steer Google right to the information we're looking for. I'm not sure if there's a linguistics researcher studying these patterns, but you might call it speaking Google.

I think that WolframAlpha will play a key role in this fundamental linguistic co-development process. WA doesn't promise to understand any sentence you throw at it, but it makes an effort to analyze the words you choose to best ascertain your intent. It will ignore certain filler words, and use others as queues to your intent. Over time, Wolfram's team will continue to apply "linguistic curation" (his words) and add new algorithms to make WA smarter at deciphering an ever-growing base of query types. On the flip side, people will learn how to format their queries to optimize the performance of their results. So long as WA's query processing is based on some consistent underlying principles, people will learn over time how to "speak WolframAlpha" (as awkward as that sounds).

It will be interesting to see how WolframAlpha interacts with the outside developer community to enable collaborative development of new query types and inclusion of new data sets, maybe through some sort of API. Google, for example, finds its strength in utilizing the community of web publishers to expand its index and determine page importance.

Overall synopsis

I think that WolframAlpha is still in its rough early stages, and it may be a bit hard to use for most people, but over time I predict that it will evolve to become an invaluable service we all use. Stephen Wolfram has a reputation for his ability to tackling massive theoretical problems and design solutions to them. While we have seen many ambitious projects claiming to become the next Google (Wikia, Cuil) which have failed, W|A seems to have a sensible approach, a capable visionary leader, and a strong early proof-of-concept. Keep your eye on this one.