Congrats; Astro; Alife

[From Bill Powers (920805.1000)]

I add my congratulations to Joel Judd (and his wife, who should enjoy a
country where she can use her FIRST language). A new center from which PCT
can grow, and a competent gardener to manage the process.

···

------------------------------------------------------------------------
Avery Andrews (920805)--

Great! I don't object to using position as the intrinsic variable for
reorganization; it's as good a way of exploring reorganization as any.

What then of the gain adjustment k1? Here things get a bit tricky. >The

first version of astro used a linear gain function, so that vrel* = >-xk.
But this works rather badly, because the accelerations it calls >for are
too high when x is large, too small when x is small (if you >move at the
controlled-for velocity, the resulting accelerations will >be a linear
function of distance from mother).

Something's wrong here, or miscommunicated. If vrel* = xrel* - xrel, it's
true that velocity will be a linear function of position error, but
acceleration should vary in a way that brakes the motion to a stop just as
xrel becomes equal to xrel*. If you start from a standstill, the initial
acceleration should be toward Mother, and then reverse as the distance
begins to decrease. If all this happens too slowly, you probably need to
adjust k1 and k2 (in my second diagram). I'm confused when you talk about
"moving at the controlled-for velocity." The controlled-for velocity should
be large when the position error is large, and decrease as the position
error decreases. Increasing the velocity reference signal should result in
a positive acceleration; decreasing it should result in a negative
acceleration. The velocity should change right along with the velocity
reference signal.

Reorganization:

If things are getting worse, you don't want another delta in the same
direction, and there is only one other direction to go ...

That's fine for the one-dimensional case, but then reorganization isn't
blind and you can't extrapolate it to the general case. It's OK if a random
choice moves you in the wrong direction, because it will be followed
immediately by another reorganization -- the error will get larger. There's
only a 1 in 16 chance of making 4 wrong moves in a row, and the parameter
won't change much because the steps are small. Once the direction is right,
you'll get many iterations before another reorganization is called for, and
will quickly make up the lost ground.

Want me to try this to compare with your approach?

I don't quite picture where you're putting this square-root thing into the
control diagram.

-----------------------------------------------------------------------
Eric Harnden (920805) --

I'll let Bruce Nevin reply for himself, but I'd like to ask you a couple of
questions.

What do Alife people consider a living system to be? That is, is there any
basic principle of behavioral organization that underlies their attempts to
explain and imitate the behavior of living systems?

Do Alife people assume that once a cognitive command to act has been
emitted, the physical actions that will realize it simply follow from the
command? You've been around CSGnet long enough to know that we see behavior
-- that is actions -- as a variable means of accomplishing repeatable
outcomes in a disturbance-prone environment; that it is perception of the
outcome that is under control, not the action that helps to bring it about.
Do Alife theories that relate thought to action take this closed-loop
phenomenon into account?

Do Alife people see the process of computing behavior as assessing the
environment, computing the actions that will achieve a given result in that
environment, and then executing the actions? This is just another way of
putting the preceding question, but it's more specific to some of the
behavioral models that seem to come out of Alife.

I agree with you that we PCTers are not too careful about making
inflammatory remarks. You have a right to bitch about them, even if only
from the standpoint of demanding equal respect. If everyone will be
reasonable and respect the will of others, we can examine differences in
our approaches and work them out. I don't mean to rule out sharp remarks
and retorts -- it's just that we'll get further faster if we keep in mind
that being friends is better than being enemies. You can say pretty much
anything you please to me as long as I can be sure you don't hate me. I
hope you feel the same way.

As you say, Alife researchers have goals and are heading toward them. This
all comes down to what's meant by designing artificial life, as opposed to
just designing a system that will accomplish something. Maybe you could say
something about that.
------------------------------------------------------------------------
Best to all,

Bill P.