Mountebanks like Taleb sell their wares by making the regular jamoke reading his books and essays feel fiendishly intelligent for understanding the concept of fat tails at the expense of all those pointy headed Ph.D.’s in the back room with their slide rules and white laboratory jackets.
So true.
I tried Wolfram|Alpha today.
Color me unimpressed.
Wolfram|Alpha bills itself as a new kind of search engine. It’s new, I suppose, in that it provides an unprecedented level of analysis of the data. You get charts, lots of them, and answers to questions you didn’t even ask.
Wolfram|Alpha is basically an analytics engine, built with Mathematica, sitting on top of a data warehouses. Data warehouses are not new; nor is the idea of extracting information from a data warehouse and running the data through some algorithms, or performing any other kind of data analytics. I am even sure that there are people who access their company’s data warehouse via Mathematica. Still, Mathematica is Wolfram|Alpha’s special sauce – it’s a robust, powerful tool that can do some pretty cool things. And they’re making select features available online, in a pretty logical extension to the Mathworld portfolio.
The point, however, is that the weakest part of Wolfram|Alpha is also its greatest challenge: the data warehouse. Wolfram|Alpha needs a tremendous amount of “facts”, because facts are what make the engine useful. But where do the facts come from? If there are multiple contradictory facts, which one do you choose? What level of quality control do you impose? And so on.
The question is whether or not one company can afford to amass a database like that, and then maintain it. The larger the database becomes, the larger the maintenance costs become. And, arguably, the less marginal value each additional piece of data has. Is there an alternative? The obvious one is Freebase; an open data repository. Of course, Wolfram|Alpha will probably import all that information – though I don’t want to be on the fact-checking team for that.
Wolfram|Alpha, no doubt, will do everything it can to build its database. It’s already announced a “Pro” version, which will allow users to “upload and download data from Alpha”. OK, but who precisely is going to take advantage of that? And: will the data that people upload mysteriously turn up publically? Academics who have access to data should have access to Mathematica; and you can do a whole lot more in Mathematica. Unless people don’t want to do the work… which is entirely possible for, say, exploratory research.
Will it be successful? Quite possibly. But not, I think, likely; and hardly inevitable.
Saul Hansell has an article on NYTime’s Bits blog that goes into the “cost of downloading.” Unfortunately, we have to wait until the last four paragraphs for the “meat” of the article, and even then the point isn’t clear:
All these costs, by the way, apply whether or not anyone on the system is actually surfing or downloading anything. I asked Mr. King to help me figure out what a cable company pays per gigabyte used by its customers because Time Warner wanted to charge customers $1 for every gigabyte they used over a certain monthly allotment.
He told me that telecommunications providers will not sell bandwidth by the gigabyte to businesses, even though many customers want to buy it that way. For example, some movie studios that send large files to DVD manufacturing plants, don’t want to pay for connections they only use from time to time.
“The network providers almost always say ‘No,’” Mr. King said. “As long as the bandwidth is open for business, it will cost you the same whether there is data running or not.”
In other words, the cable and phone companies want to charge consumers per gigabyte even though they refuse to sell it to business customers on the same basis.
Let’s be clear: cost to an ISP is based on peak usage, not GB/month. That means, essentially, that if Time Warner wants to add “bandwidth caps” and charge users who go beyond that, it is not actually reducing costs.
Now, Time Warner may (legitimately) argue that bandwidth caps discourage certain types of bandwidth-intensive usage – for instance, watching YouTube videos. A 5-minute HD YouTube video might be 100 MB. It also, logically, discourages P2P traffic. This may reduce peak bandwidth at certain times, which would improve service for all nodes.
However, the pricing structure remains arbitrary – and operates solely as a way of changing user behavior without changing Time Warner’s cost structure. In other words, Time Warner is “nickel and diming” people to get them to use less of their internet connection.
Time Warner – and other companies – also remain vulnerable to the charge that they’re attempting to prevent the internet from delivering some content – namely video. If Time Warner sells TV service, and also sells TV (and uses separate networks for each), then by imposing a cap that makes watching videos online prohibitively expensive, Time Warner can prop up TV use. This is a particularly important point since companies have been putting shows up on demand (Hulu, etc).
Bandwidth caps may make business sense, at least in the short term. A generous interpretation would be that bandwidth caps will give ISPs time to increase peak bandwidth capacity on their end. A less generous interpretation is that Time Warner is looking both for additional revenue, and is trying to prevent other businesses from being undermined. And, on the plus side, the RIAA & MPAA is happy since people will download less.