Taking theories seriously

[Hans Blom, 950907d]

(Bill Powers (950905.0830 MDT))

I think we have to be careful not to translate "it's all perception" to
mean "everything is optional." We can recognize that the reason most
people see colors in the world of experience is that we have human
sensing equipment and neural processors, so color is indeed a perception
and not a property of the outside world. However, almost all of us have
color-perceiving systems that are so similar that what we learn about
one person tells us something about how most others perceive.

Do you mean that science is a _social_ enterprise, a search for things that
we, humans, can agree upon, rather than a search for "truth" / reality? Or
is that the same thing?

It's very easy to interpret "it's all perception" to mean "I have a
right to my own way of seeing the world."

Sure you have the right to see things any way you want. I can't stop you.
Moreover, you do already. You have no alternative. But in science, we do
not talk about rights. That is law, or ethics, maybe.

                                        From there it's a short step
to saying "Any theory is as good as any other theory"

No, it isn't. That's quite something different. At least, when we remain
within the domain of science. Science has criteria: goodness of prediction,
internal consistency, how well does it work? Even in the "science" of daily
life, nothing is more practical than a good theory.

                                                  and from there to
saying "The universe is whatever I believe it is."

But it _is_, in my opinion. Your world-model _is_ your world. You cannot do
better. Model-based control theory says that world-models can be "correct"
only if they incorporate sufficient uncertainty. When the true x is 3, an
assumed model-x of 2 +/- 2 gives stable, although -- in a controller that
takes the uncertainty into account (in my demo, the controller didn't) --
careful, control.

                            Even in your own model, Hans, there is a
tendency to converge toward some final set of parameters

Yes, a tendency. But full convergence takes infinite time, even if the laws
of the world do not change in the mean time. In the mean time, the world-
model depends on the actual history of its perceptions. Which will be
different from person to person.

So one person is extremely limited. It was the great invention of science to
pool all the knowledge of many individuals to improve convergence, and
average out the results -- where this can be done. True convergence is only
possible in those areas where the laws of nature are constant and no re-
learning need take place.

                                                          , and while
there is some range of parameters that will work, there is an immensely
wider range that will not work. And I suspect that the more accurate you
want the final control to be, the narrower is the range of workable

Correct. High quality control demands a high quality model / a finely tuned
controller. So let's pool resources and move...