Wolfram|Alpha tries to be a computational knowledge engine. Sometimes it succeeds

by  Alan Zisman (c) 2009 first published in CUE BC Newsletter

Stephen Wolfram is perhaps best known for Mathematica- mathematical computation software that harnesses the power of computers to visualize complex mathematical calculations. Where Mathematica tries to help technical users find the answer to mathematical problems, Wolfram’s latest project- Wolfram|Alpha- uses the power of Mathematica to try to help the rest of us find answers buried deep in public databases.

Wolfram|Alpha, which went ‘live’ on May 18 (www.wolframalpha.com), is being touted as a Google-killer. Even though it looks like a search engine- a relatively bare start page with a field to ask a question or type a search phrase, it isn’t. Those with long memories may think it’s more like ‘natural language search engines’ like Ask Jeeves. It isn’t like that, either. Instead of a search engine, they’re calling it a ‘computational knowledge engine’.

Try this- go to Google, and type in “What is the weather in Vancouver today?” The result- a page of hits, 1.15 million in all. First hit is a weather forecast for Vancouver, BC, from theweathernetwork.com. It’s what we’ve gotten used to from search engines- a list of websites that may- and often, but not always, do- have some relevance to the terms typed into the search field.

Repeat the same question with Wolfram|Alpha. The results are very different. No list of websites. Instead, there’s an answer to the question. The temperature is 14 degrees C, wind speed is 1 m/s. There are graphs of the temperature today, and projected for tomorrow, and for temperatures for today’s date over the past 30 years. There’s a note that this is for Vancouver, Canada, with a link to click for corresponding information for Vancouver, USA. A link for Source Information notes where the data came from.

Another test… as a child, I lived in Hillside, New Jersey. Searching in Google gets me a map and the first page of 1.3 million hits, starting with a link to a Wikipedia article. Wolfram|Alpha gives me a single page of data- population, location on a map of the US, current weather, and distance from nearby major cities- 26 km to New York City, for instance.

In other words, ask a question and get an answer. Very different from what search engines do- ask a question and get a list of sources for possible answers.

Wolfram|Alpha’s approach is potentially useful and powerful, but it’s only going to be helpful for some sorts of questions. The answer needs to available in structured data available online or be computable from that structured data. Dr Wolfram has said: "Our goal is to make expert knowledge accessible to anyone, anywhere, anytime."

So you can find lots of statistics, and get graphs comparing them to other related statistics. But want a recipe- which is something I often search for online? Don’t bother trying. Typing my name into Google gets 13,900 hits- some of them accurate. Type it into Wolfram|Alpha and you get ‘Wolfram|Alpha isn't sure what to do with your input.’ (Apparently, when the servers in Wolfram’s five data centres are overwhelmed with requests, the error message will say thrill fans of the movie 2001 by saying: 'I'm sorry Dave, I'm afraid I can't do that').

So it’s not going to be a Google-killer. For some sorts of questions, it may prove more of a Wikipedia-killer- at least for questions that have a straightforward or calculable answer. Want to know about Stephen Hawking? Wikipedia offers a long article with a biography, lists of awards and selected publications, references, further reading, and links.

Wolfram|Alpha gives me a single screen page, showing his full name, date and place of birth, and a timeline running from birth to now, but with no entries other than years. Again, not the sort of question likely to get a useful response. The Wikipedia article on Hillside, NJ similarly offered more interesting and more useful data.

Wolfram|Alpha can, however, be fun to play with. I entered my birth date (January 5, 1951). In response, I learned that it fell on a Friday and that (as I write) it was 21,318 days ago. Nothing particularly notable happened on that day.

Note that it’s not just snipping out data from existing documents, as some theoretical future search engine might. Instead, it is able to calculate answers- like how many days from my date of birth to today- that don't already exist in indexed documents.

"I wasn't at all sure it was going to work," Wolfram said. "But I'm happy to say that with a mixture of many clever algorithms and heuristics, lots of linguistic discovery and linguistic curation, and what probably amount to some serious theoretical breakthroughs, we're actually managing to make it work."

It may take users some time, though, to get the information they’re looking for out of Wolfram|Alpha. Want the distance to the sun? That’s pretty straightforward- typing ‘distance to the sun’ returns a page with the answer in a variety of units. The same for ‘distance to the moon’.

How about if I want to know how much further it is to the sun than to the moon? Wolfram|Alpha isn’t sure what to do with “how much further is the sun than the moon?” And typing “sun distance moon” gets me a table with the distance to each- but in different units, not easily compared- astronomical units for the sun, and kilometers for the moon. There’s probably a way to get it to generate the ratio I want… certainly I could use a calculator and work with the data from the individual pages for distances to the sun and the moon.

Wolfram hopes that this project will become widely accepted and that then,  “it will raise the level of scientific things that the average person can do.  People will find that the world is more predictable than they might have expected.  Just as running Google is like having a reference librarian to help you, running Wolfram|Alpha will be like having a house scientist to consult for you.”

While neither a Google nor a Wikipedia-killer, Wolfram|Alpha, even in its early stages may be a useful tool for you and your students. Give it a try.