May 21, 2009

Don’t Throw Wolfram’s Baby Out with Google’s Bath Water

Michael Feldman

Time to break out of the search engine mindset. Wolfram Alpha is not a Google wannabe.

With a geeky flourish, Mathematica creator Stephen Wolfram launched his “computation knowledge engine” last weekend, wrapping a live webcast around its debut. The launch of Wolfram Alpha attracted a lot of attention from the digerati, who, by and large, trashed the new Web app as a feeble attempt to top Google. The critics latched onto two general arguments:

  1. You’re not like Google. How can you possibly be relevant?
  2. You’re trying to be like Google. But it doesn’t work right.

Some reviews, such as this one in Slate, managed to encapsulate both concepts, contradictory though they were. The lead-in complains “Wolfram Alpha is yet another pretender to the search giant’s throne” and follows up with “Wolfram Alpha, the latest alleged Google killer, is not a search engine.”

Wolfram himself has been careful about characterizing his invention as anything but a computational tool, so it’s unfortunate that so many reviewers saw Wolfram Alpha through Googley eyes. Obviously, you’re going to be frustrated if you try to use a calculator to do a search engine’s job.

Some of the more specific complaints had to do with Wolfram Alpha’s inability to understand certain types of questions and with the tool’s incomplete database. These are certainly more valid criticisms. In fact, we pointed out a few problems in our coverage this week. But the tool wasn’t advertised as a complete and perfect system. It is, after all, just a few days old.

As with any new type of Web app, users will learn to negotiate the input language over a period of time. Just as Google trained people how to use search terms, Wolfram Alpha will teach people how to frame questions for quantitative-based knowledge. The process is bound to take longer with Wolfram Alpha since there are many computational domains and many kinds of possible outputs. By contrast, with a search engine we just expect the simplest kind of database transaction, albeit blazingly fast.

Linux Magazine’s HPC senior editor, Douglas Eadline, wrote a more coherent review of Wolfram Alpha this week, and in it, mirrors my take on the press coverage almost exactly. Says Eadline: “There is a tendency to compare everything to Google these days — and conclude Google is better and will smash these pesky sites in short order.” He goes on to say that he believes AI-type tools along the lines of Wolfram Alpha will become the “killer apps” for HPC down the road. Again, I think he’s on to something here, although I suspect the AI label will be dropped once these applications become more generalized.

In fact, many of the first green shoots of 21st century AI have their roots in HPC. Besides Wolfram Alpha, which is currently built atop five HPC-style clusters, we have IBM’s new System S stream computing software, which can tap Blue Gene and other expensive IBM hardware to do heavy-duty real-time analytics. Intel’s version of AI is RMS (recognition, mining and synthesis), a suite of advanced terascale software that will use manycore processors, optical interconnects, and 3D memory. All of these software platforms attempt to move beyond the simple database transaction model of computing.

But this is probably not a great time in history to be promoting “the next big thing.” People are understandably jaded right now — it being the end of capitalism and all. A wary (and weary) public is looking for instant gratification. Hopefully Wolfram Alpha will stick around long enough to prove itself in saner times.

Tags: , ,
Share This