Re: Beyond Freedom and Dignity: A Philosophical Review

Allan C Cybulskie <allan_c_cybulskie@xxxxxxxx> wrote:
On Jun 1, 7:05 pm, c...@xxxxxxxx (Curt Welch) wrote:

The fact that an agent is controlled by his environment does not remove
it's ability to choose.

In order to make this claim, you have to define what you mean by
"controlled" and what you mean by "choice".

Both are easy enough to do.

A learning machine is "controlled" by it's environment in the sense that by
changing the environment, the future behavior of the robot will be changed.

It's also controlled equally by its nature - its internal structure.
What it learns from the environment, and what motivates it, is determined
by its internal structure. (none of this is lost to a Behaviorists).

A reinforcement learning machine must include the critic which is the
foundation of what the machine will see as "good" and "bad". It's the
definition of what the machine's purpose and goals are.

For example, I could build a reinforcement learning robot which a critic
that rewarded the robot based on how much light was falling on it's light
sensory. More light could mean more reward. This robot, because of how the
citric worked, and because it was a reinforcement learning machine, would
have a goal of seeking out light, and avoiding darkness. How clever it was
at finding complex behaviors to reach these goals would just be a function
of good of a reinforcement learning machine it was.

Now, we put this beast into some environment, and it will learn how to
react to the environment, in order to maximize its reward - it will learn
what behaviors work best, in that environment, to stay in the light, and
avoid the darkness.

The external behavior we see such a machine produce, will be under direct
control of the environment. We could create an environment with a red
button and a black button. Hitting the red button turned on the light for
a few seconds, and hitting the black button turned off the light. Our
machine would learn to keep hitting the red button and never hit the black
button.

Put it in a different environment, where the roles of the red and black
buttons are reversed, the the machine would learn just the opposite - it
would hit the black button all the time and avoid the red button.

In these two situation, we didn't change the robot at all. We only changed
the environment. But as a result, the robot's behavior was completely
different. Within the limits of what the robot was able to learn, we could
create an environment to make the robot do whatever we wanted it to do. If
we wanted it to go over to the typewriter and type out a Shakespeare play,
we could create an environment which would train it to do just that
(assuming the machine was able to learn something that complex).

This is what is meant by the machine being under the control of the
environment. But you will notice, that it's also valid to say it's under
control of the critic hardware. Our robot will like light because of the
critic hardware, and no environment we place it in will make it hate light.
So, we can't change its prime motivation by putting it in a different
environment (unless we consider the critic part of the environment - which
they do actually do in reinforcement learning but which doesn't really
apply to humans where we don't have the power to rewire their brain to
change the prime motivations).

For a machine which is not a learning machine, but which simply had many
hard-coded behaviors built into it (instincts), changing the environment
would not change the system. If it was programmed to press red buttons,
then changing what the red button does in the environment, would not change
the behavior of the robot. It will keep pressing red buttons no matter
what you put in the environment. That's an example of a machine which is
not under the control of the environment.

I don't think these ideas change the normal notion of control and I
wouldn't expect any of these to confuse you.

The definition of choice is where you seem to have the most problems.
The learning robot is free to make any decision it wants about what to do
next. This is what we call choice. It is free to select different
behaviors depending on the current context. And because it's a learning
machine, it's also free to change what it will do in the future, when faced
with the same choice again.

Now, you might respond that it's not "free" at all since it's actions are
determined by it's physical structure. But how do you think this is any
different than the type of freedom you have as a human? Do you think your
body is not forced to do what it does, by the laws of nature like gravity?
Do you think that your behavior is not determined by the laws of nature
acting on the physical structure of your body?

The only way to escape this conclusion (that I can see) is to believe that
our actions are actually controlled by something outside the laws of nature
(like a soul).

It's not really important however if someone were to believe something like
that. In the end, we can look back at what has happened, and just assume
that it all happened exactly like it did, because it was predetermined by
some laws of nature. And that idea, whether it was true or not, doesn't
change anything about how we live our lives. Calling what we do "free
will" tells us nothing about why we do what we do, and it doesn't tell us
whether it was predetermined or not.

All we really know, is that whether it's predetermined or not, the cause of
human behavior is too complex to allow any human to fully predict the
actions of another human, or to predict their own actions.

Only then can we see if
these contradict the normal notions of choice and control used by
people arguing for free will.

And likewise, you have to make it clear what you think choice means.
Currently I don't understand what you think it is, because if you
understood it like I do, you wouldn't be having the problems you are having
understanding these issues.

It is no use proving a limited form of
control and choice that will not do the job that the free will
advocates want it to do and then simply claiming that because you use
the word "control" and "choice" that they aren't incompatible. And as
Neil pointed out, doing exactly that is lamentably a common problem in
free will discussions.

Well, what magic powers do you intend to bestow on a human by saying they
have free will that isn't bestowed on a robot with sufficiently complex
machinery to duplicate human choice making?

If you REALLY want to get into this, look at my point 3) in the
original post and see which version you want to sign on to, or provide
another alternative to my descriptions.

I've not read the book so I can't comment on anything relating to what
Skinner might, or might not, have been trying to say - except to just guess
what I suspect he might have been trying to say based on your words. I can
comment on your logic however.

Skinner I'm sure believes that the controlled are just as much determined
by their environment as the controllers. To assume otherwise would be
absurd.

You comment in that first post:

This is an issue that creeps up a lot in the philosophy of free will:
people with deterministic views smuggling free choices into the
picture to show how deterministic views can still be okay or at least
less scary.

I think this is the basis of all your confusion. You don't understand how
what you call "free will" is completely compatible with a deterministic
view of behavior. I don't see a conflict, so I'm having problems
understand why you think there is one. You are going to have to explain to
me why my definition of choice from above is not what you are calling free
will.

>But this point, which should be trivial to
understand, seems to go right over your head because I suspect you have
a view of "choice" is that is inconsistent with with Skinner is talking
about.

This has NOTHING to do with that. Skinner IS inconsistent; he assigns
responsibility to controllers but not to controlled.

Again, I can't comment on the book because I have not read what he wrote.
But I do know that Skinner is not as stupid as you imply he is. He would
not fail to see the obvious symmetry. I'm sure he understood it completely
and if he was here to read your comments, I would suspect him to look at
you dumbfounded wondering what you are thinking just as I look on
dumbfounded trying to grasp what you are thinking and where you are getting
confused.

If he focused on the "controllers" I have to assume it was because he was
trying to get the people in charge (aka government, top management -
whoever) to understand what he understands, and to change their behavior.
He was trying to control the controllers by writing the book, in the same
way he was trying to get them, to do their job - by changing the
environment of the people being controlled. He changed the environment of
the controllers, by putting his book into their environment, and in turn,
he hopped that the change to the environment would result in the
controllers changing.

That has to be
rectified or explained, no matter WHAT he means by choice and control
(and I gave him two different meanings for choice and control in the
original post, and there are likely to be others).

I find it hard to believe that Skinner was actually using the words you are
using here - the idea of assigning responsibility to the controllers but
not to the controlled. I suspect that what he was doing was simply trying
to manipulate the controllers to make them manipulate the controlled
because he felt that was the best way to make society better. I'm sure he
well understood that the controlled would also be manipulating the
controllers indirectly though tools like voting, or writing books, etc.

Every agent that ends up changing the environment, will be indirectly
having an effect that which is controlled by the environment.

To understand Skinner, you have to first learn to think like a
materialist. From what you have written, I think that's not possible
for you to do.

Well, I take this more seriously if I thought you knew what you were
talking about. But you define anyone who disagrees with you as a
dualist.

You talk as if I'm just name calling. I'm not. I'm being excessively
logical. I call anyone a dualist who is not being consistent with a
materialistic view - whether they understand they are being inconsistent or
not.

I personally prefer to take the word of materialists who
don't do that, and most of those agree that there are problems here
that need to be dealt with. Most of them don't think that there IS a
problem there,

What do you mean by "here"? Skinner's book, or some issue of free will
vs determinism?

but that an explanation is indeed required for these
things. You do not. Who do you think I should give more credibility
to.

This isn't an issue where you need to trust anyone. All the facts are here
for you to look at. Figure it out for yourself and stop waiting for
someone else to show you the answer. I don't trust one person more than
another here. I look at what they have to say and I figure out for myself
which ideas are valid and which are inconsistent with known facts.

You shouldn't be worrying about whether I am right or whether someone else
is right. You should be able to look at the facts that everyone shows you,
and find the truth for yourself. You should only be using me as a source
of stuff to think about - not as a source of the right answers.

What I suggest to you, is that there is no incompatibility between the
ideas of free will, and determinism. Now, the ball is in your court.
Solve the problem for yourself. I've already solved it for myself.

I would also suggest that nothing Skinner wrote was inconsistent (at least
not in the ways you seem to believe it is). If you think it is, I would
say you have not yet figured out the puzzle for yourself. I'm very sure
that Skinner did figure it out for himself.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.