[FRIAM] Unintended effects of intentional systems

[Eric Charles]

Delayed continuing:
The example of the three kids with the magnifying glass demonstrates a
slippage that happens sometimes in "point of view" talk. The reason for
playing with the magnifying glass is different for each child, and what
each child thinks of as "the reason" the other two see as side effects. One
could thus say that what is a reason and what is a side effect depends on
your point of view, and that makes the whole thing seem a bit wishy-washy
and "subjective". But it isn't, because we are not, at any point, trying to
say that anything is inherently "the reason" or inherently "a side effect."
What we are trying to do is point out that each child has an objective,
verifiably reason for picking the magnifying glass; they objectively,
verifiably did not select it for those other reasons. For the sake of
simplifying the example, we assumed they could provide accurate verbal
reports, but the ultimate proof would be found by decoupling the naturally
co-occuring features of the toy.

Per Glen's question: We can provide magnification with a cell phone. We can
use multiple lenses to create a device that focuses rays to heat surfaces
and which magnifies without inverting the image. We can provide
fire-starting devices of all sorts. So, the three functions co-occur in the
particular device we started with, but it is certainly possible to decouple
them.

The magnifying glass example isn't very "FRIAM" though..... how about:
We are observing a critter living in the forest. The critter apperars to
looking for things to eat around dawn and dusk. There are leaves and twigs
and berries and rocks and dirt and grass. Even with its dichromatic vision,
there are a lot of colors and smells. The various types of things come in
lots of different sizes. Obviously those variables are correlated to
various extents.

Observing over an extended period, we see that the creature eat mostly
long, thin, light green leaves, that (to us) have a mildly sweet fragrance.
We become interested in why the organism does that, and start introducing
all sorts of new leaves into the environment:
* We take the current target leaves and coat them in a few different types
of oils to *alter the smell*.
* We find a variety of long, *wide*, light green leaves with the same
fragrance and introduce them to the environment.
* We find a variety of *short*, thin, light green leaves with the same
fragrance and introduce them to the environment.
* We find varieties of long, thin, mildly fragrant leaves of *a few
different colors* (dark green, yellowish, etc.) and introduce them to the
environment.

We find that the critter's eating pattern is not affected by the smell,
eats the wide leaves, but does not eat the short leaves or the differently
colored leaves. (By the way, we can do separate experiments if you want to
verify that the critters can distinguish all the variables in play, that's
pretty easy, it is just an extra step.) So, by altering the environment to
include objects not normally present, we have learned something about what
the organism *wants* to eat, which we could not have learned without such
manipulations. It wants to eat long green leaves, regardless of smell or
width.

We could obviously get more fine grained in our manipulations, and
introduce other variables as appropriate. (There may yet be an extreme at
which the organism does care how long or short the leaves are, but we have
at least confirmed a lack of caring within the range tested.)

One might presume that those preferences arose as the result of natural
selection, because the leaves that meet that criterion in this critter's
ancestral environment provided an advantage if eaten preferentially.  There
are several caveats, however:
1) We cannot rule out that those preferences arose due to genetic drift
(i.e., that it provided no advantage and no disadvantage, but became
fixated due to chance).

2) If it was selected for, we cannot be certain that the selective pressure
is still operating today.

To get at these issues, we could perform a *different* set of experiments,
manipulating the food the animals were eating in various ways and seeing
how that affected their success in life and in reproduction. IF we found
that animals eating other leaves found in their native environment did
worse, THEN we would have evidence for retracting the caveates above.

Continuing the caveats:
3) We would definitely not have evidence that the preference-attunement
system would provide advantage in any other environment that had different
leaves present, because we almost always find that you can develop
"super-stimuli" that the organism prefers over the actually-advantageous
thing. For example, we might find that, given the choice, the organism
would preferentially eat an even thinner variety of leaves, even if those
leaves provided almost no nutritional value, and the critters who ate those
even-thinner leaves never successfully reproduced. (In general terms, we
would expect to be able to identify ways in which the preference-mechanism
was over-fit to the ancestral environment, and could work against the
organism in a slightly different environment.)

4) IF we found that organisms eating those leaves had a reproductive
advantage over those that did not, AND we started to experiment to
determine why, we would expect the answer to be something about the
chemical make-up of the leaves (broadly speaking). Once those factors were
identified, we would expect to be able to produce visually identical
leaves, which the critters prefered equally to the naturally occuring
leaves, but which did not provide the functional benefit. That is, we
should be able to dissociate the variables describing the *want* of the
organism from the variables explaining the *evolutionary function* of that
want. It is not literally impossible that they would be identical, but it
would be highly unusual for the studies to determine that. Eating long-ness
isn't a thing evolution cares about, but if long leaves, in the ancestral
environment, provided a certain amount of carbs and crucial minerals,
evolution can avoid having to create delicate mineral-identification
mechanisms, and just create a critter with a preference for eating the
things that happen to have those minerals in it.

So what you have is a mess of correlated variation between several
variables, *some* of which the goal directedness of the organism points at,
and *some* of which have been responsible for the multi-generational
arising of that preference. Without doing experiments, it is impossible to
determine what variation within which variables fits into each of those
categories.

And... Depending on what variation is isolated during the course of those
experiments, we could use a bunch of different types of math to describe
the relevant variation. Maybe preference (or function) is described
adequately by a single variable linear equation. Maybe it is best described
by a multivariate equation with exponential effects, interactions, and
limits. Maybe it is best described as a manifold. Maybe it is best
described using low dimensional topology. Etc., etc.



P.S. In 4, Even if we demonstrate reproductive advantage, it is not
guaranteed that it is something about the chemical make-up of the leaves.
It could be something about where they are / how they can be reached. It
could be something about a beneficial bacteria that forms on the leaves
often enough to provide benefit to the organism. It could be something
about how eating those leaves creates a crucial response from the plant,
such as exuding a liquid with a scent that attracts prospective mates.
Etc., etc.

P.S. It is possible that we would see some developmental adjustments in
organisms during some of the above experiments, and those could be
investigated in other studies.




On Mon, Aug 17, 2020 at 5:42 PM Eric Charles <eric.phillip.charles at gmail.com>
wrote:

> The more we are discussing ideas, the more interesting the discussion is.
> The more we are fighting over word choice the less interesting. (Unless,
> say, we have already established full agreement about ideas, and are
> working on idea-marketing.) To get past this we need to drop as many
> problematic words as possible,  with "problematic" defined by their
> creating resistance from those in the current conversation. Then we try to
> talk about the ideas, to see if we can forge agreement using different
> words.
>
> Glen and others are right that, as presented, the "epiphenomenon"
> discussion can be easily replaced with a discussion about
> degree-of-modularity and about statistician coupling of variables. And yet
> Nick can't seem to see it that way. Either Nick completely doesn't get the
> alternative, or there is something he hasn't quite articulated clearly that
> makes the offered alternative not fully sufficient to meet his needs.
>
> I propose that Nick has failed to adequate include two factors in his
> presentation:
>
>    1. The things he is talking about are all speculative labels until
>    the-fact-of-the-matter is determined by experimentation.
>    2. He is talking about systems that are the way they are for reasons
>    (which we might reference with the word "intention" with a "t"). This gets
>    a bit muddled because:
>       - Some of the systems are human systems, i.e., systems of humans
>       doing things, and we could figure out the reasons they are doing those
>       things through a bunch of experimentation or by getting the relevant people
>       to give accurate answers to "Why are you doing that?" questions.
>       - Some of the systems are systems designed by humans, and in those
>       cases we could get at the reasons by doing a bunch of experiments on the
>       designers, or by getting accurate answers to "Why did you build it that
>       way?" questions.
>       - Some of the systems arose through natural processes, without
>       involvement of human in the past or the present. In those cases, we don't
>       have a verbal community to interact with, so we can only get at the reasons
>       why the system is the way it is through experimentation.
>
> Example:
> We see three kids playing in the yard, each playing with a magnifying
> glass. We ask each the reason why they are doing what they are doing (and
> for the sake of this example, we assume they all answer honestly). Kid 1
> says, "I am playing with the magnifying glass because it can make small
> things look larger, allowing me to visually explore them better." Kid 2
> says, "I am using it to invert images, because it is funny to look at
> things upsidedown." Kid 3 says, "I am playing with the magnifying glass
> because it can focus light rays towards a spot on the ground, to heat it
> up."
>
> Of course, as the kids play, all of them do all three things at some point
> (make image bigger, invert image, heat ground). But for each, only one of
> them is the reason they are playing with the magnifying glass, and the
> other two effects are coming along for the ride. If we could provide a
> device that magnified without inverting the image, that toy would work just
> fine for Kid 1. If we could provide one that inverted images without
> magnifying, that toy would work just fine for Kid 2. If we provided
> something that heated the ground with one of those other factors, that toy
> would work just fine for Kid 3. Etc.
>
> If you aren't interested in that distinction, that's fine, but surely it
> is a legitimate distinction for someone to be interested in.
>
>
> <echarles at american.edu>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20200821/00c8fdb3/attachment.html>