# Artifactbusters IV

[From Rick Marken (941126.1000)]

First some corrections/changes to my previous post [Rick Marken
(941124.1430)].

Here is a quote from that post with corrections inserted in caps

If the previous action (tumble) produced an improved result (a
larger gradient of attractant) then the probability of doing this action
again under the same circumstances (the circumstances being the
value of the gradient before the tumble) is decreased (INCREASED); if
the previous action produces a worse result (the same or a smaller
gradient of attractant) then the probability of doing this action again
under the same circumstances is increased (DECREASED).

The following is a slightly changed but, it turns out, much improved
version of my "de-artifacting" code. It displaces the spot by a
fixed amount (+ or - 15 pixels) on each tumble:

if random (100) > 50 then
put y + 15 into y
else
put y-15 into y
end if
if random(100) > 80 then
put -inc into inc
else
put inc into inc
end if

I also said:

Without the "de-artifacting" code, the "law of effect" model works
just fine; p(t|up) converges to about .67 and p(t|down) converges to

Actually, in the one dimensional case, p(t|up) converges to about .99
and p(t|down) converges to about .49. These tumble probabilities do,
indeed, consitute a control model, as I noted.

Bill Powers (941123.1415 MST)--

I don't know what you [Bruce] will make of this, but in fact the logical
path you describe just above is NOT the effective path-- it works the wrong
way, with a probability of 0.37.

By a chance property of the situation, however, this turns out to be the
least probable path, so it is most likely that this probability will be
decreased in the long run via a different path.

This is a typically subtle and insightful way of saying what I've been
trying to say all along. Bruce's "law of effect" model contains logic that
"strengthens" the "wrong" response (tumbling when going up the gradient)
This doesn't cause a problem because (as Bill notes) a "chance property
of the situation" (what I call an "artifact") makes this logical path
least probable. When this "chance property of the situation" (the
fact that the gradient is more likely to be smaller after a tumble when
going up the gradient) is eliminated (as it is by my "de-artifacting"
code) the law of effect no longer "selects" the appropriate actions.

Bill Powers (941125.1620 MST) to Bruce Abbott --

Have you tried to make your model work in a gradient of a repellant? As
it is stated, it clearly won't work: an increase in a repellant is not
reinforcing. You will have to change some definitions, particularly the
definitions of S+ and S-. Now a positive dNut must be considered an
unfavorable condition, and an increase in dNut from before to after a
tumble must be considered punishing. So the logic has to adjusted on the
basis that we know the organism will see a positive dNut as having
negative _value to the organism_. We must change our concept of the
organism's reference level for dNut.

I can't wait to hear the reponse to this; I've always wondered how the
law of effect knew which effect were good (positive) and which were

Bill Powers (941126.0515 MST) to Bruce Abbott (941121.1830) and
Hans Blom (941123) --

Hans, citing Bruce Abbott (941121.1830 EST))

... it's really irrelevant whether you phrase it as selection
BY consequences or selection OF consequences. ... The important fact is
that the perceived consequences of behavior determine which behaviors
are retained and which are eliminated.

Hans:

Exactly.

Correct me if I'm wrong, but this implies that a particular consequence
of behavior, once it has determined what behavior is to be retained, is
always brought about by the same behavior.

Now I ask both of you: what would be the circumstances in which control
theory would predict that the same consequence will be produced by the
same action every time?

Great question.

If selection by consequences is the same as selection of consequences
then don't blame me when I say "people who believe this are clueless
dunces".

What I say is just selected by consequences; don't blame me;
blame the consequences.

Blamelessly

Rick