12 December 2008

A short digression about information as a thing

Because I'm going to want to talk about information as a thing in some survival-of-government posts. Consider this a giant footnote.

The formal mathematical definition of information is that it reduces the probability of uncertainty.

This has the advantage of being rigorous and correct, and the disadvantage of being extremely hard to use. (Ever tried measuring the uncertainty in an unknown specific customer in the future?)

The much more usable heuristic is "information causes change". If it doesn't cause change, it's just data.

One easy example of this is a computer log; it's full of things the people don't care about, more or less "the computer says it spun the disk!" and "the computer says it talked to the network!". This is data; it's when it says things like "the graphics card is on fire!" or "crackers keep trying to get through port 40" that, suddenly, people care and that log entry is information, instead of data.[1]

Another example is a bank or credit card statement; the stuff you care about is the stuff you don't already know. If you recognize the line item, nothing changes. If you find yourself wondering "when did I start paying 28% annualized interest on this card?" that's information, in that it causes change. (Maybe only change in your mental state, but change.)

So if I start talking about information as a thing, what I mean (usually!) is that there's enough information to make substantially the same changes you would if you had complete information. (Not perfect information; perfect information involves the information that's in the future, and we don't have any practical expectation of getting that.)

[1]the most reliable method yet found for writing a "logwatch" program, one that is supposed to provide a system administrator with just the information in the computer logs, is to throw out everything known to be boring. Whatever is left over should at least get some cursory human attention. This is both much easier and much more useful than trying to teach the computer to figure out what the human will care about.

No comments: