Edgelings

A New Kind of Search?

A NEW KIND OF SEARCH? by Charlie Martin

 Stephen Wolfram likes big ideas.

After starting out to be a mere child prodigy in physics and mathematics –  his first paper in particle physics published at 17, PhD and member of Cal Tech’s faculty at 20 – Wolfram left conventional physics to more or less invent a new area of research, that of studying complex systems with cellular automata.

In passing, he led a group at Cal Tech building a “mathematician’s helper” that became the hugely popular software program Mathematica, and wrote a wrist-breaker of a book called A New Kind of Science. A New Kind of Science could be read to suggest that physics had worked itself into a dead end, and should all be revised using Wolfram’s cellular automata.

Like I said:  Big ideas.

Wolfram’s newest big idea, which became generally available this weekend, is Wolfram|Alpha. The hype has been intense, with suggestions that it’s the next Google, even the Google-killer.

At first glance, Alpha certainly looks very Google-like: a simple, clean web page with one line in which to type a phrase. What it does with the phrase is very different, though.

Say you enter the phrase “distance to Neptune” in Google. You get a long list of web pages that have that phrase, ranked by Google’s notion of importance. The top entry from Google is from the University of Colorado’s Windows to the Universe site. A nice page, a nice site, and it tells you the minimum distance from the Earth.

Type the same query into Wolfram|Alpha, and instead you get this: the current distance between the Earth and Neptune, in astronomical units, kilometers, meters, and miles, the time it takes light to travel that distance, and the radiation pressure of sunlight at that distance.

The nerd’s instant reaction is “Oh, that’s cool.” (Certainly that was the reaction of this nerd.)

So what’s the difference between Google and Wolfram|Alpha? Simply, Google tries to find pages on the web; Alpha is using what it knows to compute the answers when you ask. (Apologies to John Searle, but I’m going to use the word “knows” with no scare quotes. It’s close enough.)

Of course that’s not the only kind of question you can ask Alpha. Enter “$250+15%” and you get back not just “$287.50” but also the steps of the arithmetic. Enter “sin(x)/x” and it will show you a plot of the equation, the Taylor series expansion, the first derivative, the indefinite integral, and just bunches of other stuff. (Follow the links, listing it gets dull real fast.)

Here are some other fun examples:
• “Solve d=1/2at2 for t”
• “Broomfield CO to San Francisco”
• “oracle sun” – which it interprets as stocks, and shows you more quantitative information about the two stocks than … well, plenty of information.
• Enter “1,1,2,3,5,8,13,…” and it recognizes that as the series of Fibonacci numbers, and shows you various facts about them.

All in all, Alpha’s pretty impressive. I can already predict that calculus classes are going to need to cut people off from the internet — because Alpha looks like it would work any problem in first-year calculus without a blink.

But Alpha does have its limits. Ask it “distance to mars” – no capital M – and it basically doesn’t know what a “mars” is. It doesn’t understand addresses (yet?) and it’s sensitive to other spellings. It’s easy to find the edges of what it knows.

But the problem I have with Alpha is that I’ve yet to ask it a real question, one I wanted to know something about, and get a useful answer. Ask it about “John Searle” and you get his full name and place of birth. (Here in Denver, in fact — I didn’t know that.) Ask about “John Searle Chinese Room”, the link I wanted above, and you get “Wolfram|Alpha isn’t sure what to do with your input”.

So, for now at least, Alpha appears to be sort of the ultimate desk calculator – many things you might want to calculate, it will calculate. Used with Google, it’s better. If, say, you want to compute how many golf balls would fit inside the Earth, you can use Google to find the diameter of a golf ball (1.680 inches), use Alpha to find the volume of the Earth in cubic inches (6.61×1025) and finally compute the answer (about 2.5×1025.). It can’t do it all at once, though: you have to look up the diameter (Google), find the volume (Alpha), save the volume, find the volume of the Earth in cubic inches (Alpha) save the volume in its plain text form because you can’t use the display form, then do the division.

In other words, rather than being the Google-killer, Alpha for the moment is really mainly a curiosity, plus a way to get access to some of the simple parts of Mathematica without paying $2500 for a license. Wolfram and his team are working every day to add more knowledge, but others, like Doug Lenat and Cyc have done similar things; as with most all AI projects, it’s easy to start something like this, and hard to really make something of it.

Whether Wolfram will be able to make something of Wolfram|Alpha, we’ll have to see.