This might be the droid you’re looking for

May 13, 2009

You have to hand it to Stephen Wolfram. His new Alpha knowledge searche engine, if it fulfills the planned mission, will be to Google and Wikipedia what C3PO is to R2D2 (no offense to maintenance droids). On the one hand you’ve got the world’s most extensive encyclopedia, a passive collection of free documentation, hyperlinks, and the occasional public domain photo, and the world’s most comprehensive search engine, specializing in elaborate indeces for finding the closest match. Direct input yields direct output with a dizzying array of roadsigns pointing to further outputs, some of which you might want—say, the real location of the Northern Hemisphere’s Pole of Cold—and a bunch of others you might not. That is, unless you are myself and are sucked into a list of weather records from around the globe. Whether or not you believe Wikipedia has value as a site of record, it is indubitably one of the world’s great time wasters.

Alpha, on the other hand, is supposed to be a sort of Mathematica of all human knowledge. Unsurprisingly then, the beta tester I’ve been toying with for the past day or so is full of holes. Ask it mathematical questions, such as x^3 cos(y), and it shines, instantly supplying various plots and alternate expressions, readily available in “copyable plaintext” for use in a research paper or computer program. Veer toward scientific topics and the program is hit-or-miss, failing to recognize The Second Law of Thermodynamics. But it does respond to the search term “entropy”. Forget about reading on the term’s etymological and cultural background. Alpha delivers the basic goods: “[mass] [length]^2 [time]^(-2) [temperature]^(-1)”.

Looks a lot like a passive database doesn’t it? Not exactly. If I accept that this “knowledge engine” has a learning curve and pursue suggested queries supplied by Wolfram I find that the incomplete engine is pioneering a new level of interactivity. Simply typing the names of stars in the Centauri constellation and separating them with commas, for example, causes Alpha to assume a comparison request. In addition to spectral class, surface temperature and whatnot, the system delivers positions relative to Earth-bound locations (near Newark, N.J., in this case). This is real-time information, best illustrated with a query to the International Space Station. Alpha uses external sources to track the station, reporting current velocity for example.

The approach reminds me of a few standout software innovations from our R&D 100 Awards that rely on real-time information combined with interpretative software. One, Sensor Web 2.0, used satellite and ground-based sensor information to build a virtual environment for monitoring wildfires. Another, the Biomimetic Search Engine from Lawrence Berkeley National Lab, sought to emulate human thought patterns to the extent of continuing to learn.

Alpha splits the difference, going for a generalist approach with barest hint of machine learning. This pseudo-C3PO of software might not find immediate acceptance in the scientific community, but it shows potential. Yes, it’s difficult to give up on a loyal R2D2, but protocol droids do have an advantage in communicating knowledge.