[From Bill Powers (941105.0650 MST)]
Bruce Abbott (941104.1500 EST) --
As our field observer of the aliens, you are showing an acute ability to
grasp their thought processes, including what you call being "very good
at verbal rationalizations." One method of verbal rationalization is to
present a program that does one thing while describing it as if it does
something different that is in line with one's theory. The aliens'
program contains a goal-directed mistake which makes all the difference
between working and not working. Observe:
procedure ReinforceOrPunish;
begin
If dNut > 0 then { nutrients rising: reinforcement }
begin
If NutSave > 0 then { S+ present when last tumbled }
begin
if pTumbleGivenSplus > pMin then
pTumbleGivenSplus := pTumbleGivenSplus - LearnRate;
end
else { S- present when last tumbled }
begin
if pTumbleGivenSminus {> pMin} < pMax then
pTumbleGivenSminus := pTumbleGivenSminus {-}+ LearnRate;
end
end
else { nutrients not rising: punishment }
begin
If NutSave > 0 then { S+ present when last tumbled }
begin
if pTumbleGivenSplus < pMax then
pTumbleGivenSplus := pTumbleGivenSplus + LearnRate;
end
else { S- present when last tumbled }
begin
if pTumbleGivenSminus {< pMax} > pMin then
pTumbleGivenSminus := pTumbleGivenSminus {+}- LearnRate;
end;
end;
end;
Stripped down, the original of this procedure says in part
if dNut > 0 then
if NutSave > 0 then
increase delay
else
increase delay
Notice that no matter what Nutsave is, the delay will be increased if
dNut is positive, just as in the control model. The next section says
that no matter what NutSave is, the delay will be decreased if dNut is
negative or zero. So the state of NutSave makes absolutely no difference
in whether the delay will be increased or decreased. This is simply a
slightly-disguised version of the original control program.
You may notice that I have commented out the incorrect comparisons and
signs and entered the correct ones that fit the aliens' theory, in which
the sign of NutSave makes a difference. If you drop this modified
procedure into your program, you will see a random walk. At least I see
one on the ship's computer.
···
-------------------------------
I have consulted with the aliens, and they refuse to accept your
analysis. They say that the response being targeted for "reinforcement"
or "punishment" is Tumbling, and that Tumbling takes place in a
Stimulus Context defined by the sign of dNut, which is, after all, the
only thing that e. coli can sense. A positive dNut is S+ and a negative
dNut is S-. They further note that the Consequences of tumbling in the
presence of these Discriminative Stimuli serve to alter the probability
of tumbling (i.e., reinforce or punish the behavior). These
Consequences which follow a given Tumble are a positive dNut
(reinforcement) or a negative dNut (punishment). Consequences of
Tumbling-in- the-presence-of-S+ change the probability of Tumbling in
the presence of S+. Consequences of Tumbling-in-the-presence-of-S-
change the future probability of Tumbling in the presence of S-. This
is what is called in the alien's language the "three-term contingency."
As Spock would say, "Fascinating." [I misremembered how to spell
Spock's name, or maybe the convention changed during one of our time
warps].
These aliens have slippery ways of getting around holes in their
arguments. Notice the construct "the response being targeted for
'reinforcement'". The use of the passive voice removes any necessity of
explaining who or what does this targeting, and by what means a
reinforcement is made to have an affect on one aspect of behavior rather
than another.
Note also the phenomenon of explanation by classification. A positive
dNut is an S+, and a negative dNut is an S-. Merely by virtue of having
been classified, members of these classes "serve to alter the
probability of tumbling". The question of how class membership can
"serve" to do this is not answered, nor is there any explanation of how
a probability can be influenced by class membership, nor of how the
_correct_ alteration in probability appropriate to the classification is
made to occur.
Note further the phenomenon of adjusting the premises to suit the
observation. A positive dNut would be actually be classified as an S-,
because it must reduce the probability of Tumbling for dNut to be
maintained positive on the average. But what if it is observed that the
real animal behaves to keep dNut negative on the average? This
observation would have effects that cascade back through the
explanation: now the positive dNut would be classified as an S+ because
it increases the probability of tumbling, and Nut itself would change
from being an attractant to being a repellent. The classifications are
completely _ad hoc_ and falsification of the explanation is impossible.
Even further, note the prevalence of magical thinking in the
explanation. There is no physical connection between dNut and Tumbling.
Instead, it is merely Tumbling in the _presence_ of the class S+ that
increases the probability of future Tumbling in the presence of S+. This
is the essence of magic, where events occur because of the mere _fact_
of other events: the demon appears simply because the wizard waves his
wand. To be more charitible, this is pure empiricism, in which one tries
to learn the rules of nature without any grasp of the mechanisms.
When one begins to learn the nature of the mechanisms, rules that once
appeared quite reliable often turn out to be only plausible
misinterpretations. This is the case in the relationship between
disturbances and actions in a control system. The appearance is that the
disturbance is somehow sensed directly by the organism, and that its
sensory effects propagate to the output and cause the motor actions. In
our own culture, this interpretation was given a strong start by
Descartes. The existence of an unsuspected controlled quantity between
the disturbance and the organism, which is discovered only through
looking into the mechanisms in detail, changes this appearance entirely
and destroys what had seemed a simple cause-effect relationship.
The aliens will discover, some day, that the appearance of reinforcers
as causes of behavior is such an illusion. But they are not ready for
this discovery yet. They have not produced their Galileo, much less
their Copernicus, their Kepler, or their Newton. Their world is run by
nebulous abstractions, by rules that apply, or seem to apply, for no
reason. They have more kinship with our own ancient Egyptians and Greeks
and our early theologians than with our modern physical sciences. They
have not yet learned how to make predictions that don't leave loopholes
to allow for any eventuality. Perhaps their experiences with modeling
will reveal to them the difference between a true explanation and a
rationalization. We shall have to wait for them to make that discovery
for themselves.
------------------------------------
Picard: Computer, end program "Ancient Alien Civilizations".
----------------------------------------------------------------------
Best,
Bill P.