virus: Occam's Razor II
Sun, 20 Oct 1996 23:41:34 -0500 (CDT)

Reed, I was pushing last night, so I didn't answer your Occam's razor
query immediately. [Professor X said on Wednesday, "We are not having
class Friday. You are to read Section 6.6, and on Monday I will draw names
randomly. The students whose names are drawn will have to explain
Section 6.6, but only to the point where they don't understand it." It's
always a good idea to start early; then the impossible stuff is what's
waiting for you the night before it's due ;)]

When faced with a system [such as the development of intelligence, but
there are many others; I'll use weather and the not-phenomenally reliable
'science' of weather forecasting] that:
I) has definite normalcies
Weather: climate [usual temperature, rainfall, etc. at time of
year]. Also storms, etc. have their paths predictable on short time scales,
once they form.
II) is unpredictable in detail at large scales [isn't the weatherman
doing good if he can name the temperature to within 2^o Fairenheit? How
about naming the amount of snow down to 0.1 inch--2 decimal places?]

The first problem is isolating relevant variables. There is a good
chance that we are looking at a chaotic system, which means that errors
we can't see in our data are magnifying into results we CAN see. The
best we can do at large scales, then, is learn how to predict the
"strange attractor".

We presume that major influences [the ones shaping the "strange
attractor"] are deterministic. Technically speaking, a chaotic system is
deterministic by definition, so indeterminists and quantum physicists
will be more comfortable with "pseudo-chaotic system". I refuse to
clutter terminology.

That is, we can reasonably expect to predict the "strange attractor". As
we refine our instrumentation, we can increase the effective resolution
we have on the "strange attractor" by increasing the detail we can see.

In the above example, climate is the strange attractor. We usually don't
have snow, in June, in England. [Krakatoa changed things, thus starting
off a quote about "Derby day" and some truly absurd prediction in an
almanac that year about snow on Derby day that was CORRECT!]

However, we never will have great resolution at long ranges. A
simplistic overestimate of the maximum length of time a weather forecast
can possibly be accurate is 14 days: after that point, quantum-mechanical
error now is sufficient to completely derange the predictions. [whether
a hurricane exists, or not, are very divergent qualitative predictions].

/ Towards the conversion of data into information....
/ Kenneth Boyd