Friday, May 22, 2009

Wolfram Alpha: Not Ready For Prime Time

No doubt you've heard of Wolfram Alpha (they style it with a vertical bar, thus: Wolfram|Alpha). It's a new search engine that's nothing if not wildly ambitious:
Wolfram|Alpha's long-term goal is to make all systematic knowledge immediately computable and accessible to everyone. We aim to collect and curate all objective data; implement every known model, method, and algorithm; and make it possible to compute whatever can be computed about anything. Our goal is to build on the achievements of science and other systematizations of knowledge to provide a single source that can be relied on by everyone for definitive answers to factual queries.
Really? "...To make all systematic knowledge immediately computable and accessible to everyone." Wow! I had to check it out.

I clicked over, read the intro, watched the video, and dived in. I thought I'd start with something simple, objectively knowable, and (frankly) not hard to figure out. I asked "What is the operating cost per hour of air force one?"

I figured it would be interesting to see what the software did. Google would just deliver pages that contained the answer. Maybe Alpha would just do that, but maybe it would do something interesting, like start with the fact that Air Force One is a Boeing 747 and work the calculation that way, from known raw data. I hit Enter. Alpha replied:
Wolfram|Alpha isn't sure what to do with your input.
OK, that's a little disappointing. Maybe the question wasn't as easy as I thought. So I opened another tab and asked the same question of Google. Google instantly replied with 1.3 million hits, the top one of which was this:
Air Force One: $68000 an hour?: The Swamp - May 19
"Does it still cost $68000 an hour to operate Air Force One? ... million in annual operating costs, Air Force Secretary Michael Wynne said this month. .... thousands of New Yorkers suffering from PST, and all at a cost of 68k per hour. ...
www.swamppolitics.com/news/politics/blog/2008/05/air_force_one_68000_an_hour.html - 67k - Cached - Similar pages -
Skimming the rest of the entries on the Google page shows multiple cost estimates from other sources in the same dollar range, so right away I have a pretty good ballpark answer.

Maybe my test question was too weird. One of the examples Alpha gives involves stock/share prices. So I asked "microsoft original and current share price." That's just two pieces of data--- the IPO price, and the current price. I figured Alpha would report that and (as their demo implies) go on to calculate the difference, show a timeline, and work other mathematic magic. Alpha said:
Wolfram|Alpha isn't sure what to do with your input.
Oh, come on. I entered the same thing in Google. The #1 result from Google was for a site offering "The latest Microsoft Corporation share price." The #2 result from Google was for a site offering "Microsoft Corporation Historical Share Prices." Geez, Alpha, you're not impressive so far.

OK. Alpha is supposed to be good with dates; another type of known, objective, numeric data.

I live in Northwood, New Hampshire. There's a historic marker a mile or two from my house that commemorates a goodwill tour of the newly-created United States by Revolutionary War hero, General LaFayette. OK, Alpha, tell me what "date lafayette visited Northwood NH"

This time, Alpha didn't give up. Great! It churned for a moment and then proudly offered to show me... the distance between Northwood NH and Lafayette Louisiana. What?

Google's top result for "date lafayette visited Northwood NH:"
New Hampshire Historical Markers - A Waymarking.com Category
LaFayette's Tour - Northwood.
http://www.waymarking.com/waymarks/WMC9X_LaFayettes_Tour
Clicking on the embedded link produced this:
"Upon invitation of President Monroe, issued at the request of Congress, Marquis de LaFayette, Revolutionary War hero, revisited the United States for a goodwill tour which included an extensive visit to New Hampshire towns. He passed this spot June 23, 1825, traveling between Concord and Dover."
Two clicks, and I had the info.

Why is Wolfram Alpha so sucky? Stephen Wolfram is no dummy. I actually met him a couple times a thousand years ago when I was Editor of Byte Magazine. He'd come out with Mathematica, a software package that was unprecedented at the time, and which remains a category leader today.

Maybe Alpha needed plain math to shine.

Using an example from the Alpha site (talk about a slow pitch over the center of the plate...) I entered $250 + 15%. Alpha immediately responded with the correct answer:
287.50
Google can do some simple math, so I typed in the same thing. Google's top reply:
$250 + 15% - Wolfram|Alpha
Input interpretation: $ 250 (US dollars)+15%. Result: $287.50. Calculation:Show details. $250.00 | $250.00\n+15% | $37.50\n | $287.50 ..
.
Now that was interesting. Google didn't do the math itself, but reported Alpha's handling of the same query. I still had the answer, harnessing Alpha without ever having to leave Google. Hmmmm.

Let's try some harder math. I entered an open-ended equation, also based on an example lifted from the Alpha site: x^2+sin(x)

And indeed, Alpha shone. It actually used the engine from Mathematica to parse the equation every which-way, delivering a page full of graphs and related information.( http://www77.wolframalpha.com/input/?i=x^2+sin(x)

Google went in a different direction when I entered the same equation. Because it's an open-ended equation, Google's first result is an instructional video showing how to solve "integration by parts" equations. The second result is a PDF paper with similar information. The third result is... the Alpha page. (The Google results page: http://www.google.com/search?q=x^2%2Bsin(x))

After this, I'm hard-pressed to see any reason to go to Alpha. Google still produces better results in the kind of categories I use, and when Alpha would be better, Google points me there.

Maybe I'm missing something, but Alpha seems to be NOT a general repository of all manipulable, quantifiable, systematic data, but a very narrowly focused, special purpose engine.

In a word: Meh.

3 comments:

  1. I got about the same reaction when I was given his book, The Making of a New Science as a gift. I got halfway through it. It was entirely variations on the theme "when you take even a simple iterative system and make a small tweak in the right place, the results go all over the place."

    You just got three inches of book in one sentence. I have to give him this: I got the book when it first came out, and I still remember what it was about.

    ReplyDelete
  2. Yup, so true, definitely a waste of time. In addition to my own "strike-outs" here's more:
    http://www.newscientist.com/blogs/shortsharpscience/2009/05/are-you-smarter-than-wolfram-a.html?DCMP=NLC-nletter&nsref=blog4
    Can you say hype?

    ReplyDelete
  3. Google doesn't do a Wolfram search with maths type queries. x^2+sin(x) seems to be a special search that Google knows about from previous experience. Try some more simple equations. You won't get any Wolfram results from Google.
    Wolfram is too slow to respond for Google to actually access in real time. And Wolfram is very picky about how you phrase your queries, so it is unlikely that the same phrase would suit both Google and Wolfram.
    I'm afraid you still have to search both systems separately. As you say, the Wolfram database is still very small compared with Google, so for a while you will have to select only queries that Wolfram knows about.

    BillK

    ReplyDelete