[Martin Taylor 970411 15:15]
Rick Marken (970411.1100 PDT)]
And Eco had a a good enough answer to his enquiry right in
front of his typewriter. Eco asked (according to you):> Is truth something unknowable but to be approached by painstaking
> investigation based on observation?The "good enough" answer to this question is "yes". The answer to
all Eco's other enquiries about truth are clearly "no". What am I
missing?
You have to read, and to contemplate the message of, the book. I cannot
get you to think outside the cultural environment of the late 20th
century in quite the way Umberto Eco can (at least for me).
Actually, Eco didn't _ask_ the question in this way. What he does is to
set up situations in which one is influenced to ask onesself such questions
(not just that one or the ones I listed). The questions intertwine, the
answers seeming sometimes to reinforce one another, and sometimes to
contradict.
Of Eco's books, I have read only The Name of the Rose, Foucault's Pendulum,
and The Island of the Day Before. The three are very different, but each,
in its way, pursues the same enquiry. If you don't care to think about it,
facile answers are easy to make, and those answers will be worthless, even
if they are the same answers you might make after having contemplated the
multifaceted issues involved.
As for my gentle hints;-), I don't think I really need spell it out, if
you try to "go up a level." But I suppose I will, since you prefer it so.
Do you think I have the goal of shielding people
(common or otherwise) from certain writings? If I am in a conflict
I certainly don't notice it.
No, I don't supose you do. And I don't know how to perform the Test in this
instance. My suggestion of conflict comes only from your answer "No" to the
question, which seems inconsistent with the methods you use to suggest to Hans
Blom, Bruce Abbott, Jeff Vancouver and others (including me) exactly how
our understanding is flawed. Basically that method is "If you can't say
or see it my way, don't say it at all."
If you hadn't answered the question "No", I would not have suggested the
possible existence of a conflict.
Personally, I would also answer the question "No", but with hesitation and
some qualms. I'm not at all sure that a little knowledge isn't a dangerous
thing. It seems to depend rather on who might be shielded (or not) and
whether what they might do would be damaging. A little knowledge about
evolution has had and continues to have quite damaging social consequences.
It's considered a "good thing" for evolutionary reasons that people should
be allowed to starve on city streets. Would we be better off if those in
positions of influence were somehow required to take a full course in
evolutionary theory rather than basing their cruel policies on tag phrases
like "the survival of the fittest?"
As I said, I _would_ answer the question with a "No." But I wouldn't find
it an easy question to answer straight off the bat. A "No" is easier when
the context is a scientific discussion than when the consequences of half-
baked ideas kill people (or lead to multi-million dollar lawsuits that
should be laughed out of court but that actually succeed).
If I believed whatever it is that you apparently believe about
the work of Hans, Bruce, Jeff, and me, I would try to cast my views in
such a way that the errors you see could be perceived by the person
concerned (or by myself as an erroneous critic). That's what I'm trying
to do at present in discussions with Hans about the relation between
things that work in continuous time and discrete simulations of those
things. If neither of us get bored with it, in the end either I'll
understand and accept what he's trying to say (and will then try to
interpret to you, probably), or he will accept what I'm trying to
say, and will perhaps change something in his writing (and coding?).
Your messages to Hans mostly take the form of "MCT doesn't work so why
keep going on about it" or less polite equivalent statements. I'd take
that to be rather a persistent attempt to suppress whatever it is that
Hans is trying to get across. And similar methods have succeeded in
persuading me that there's no point in trying to get across my ideas
in a different area of concern, so they are effectively suppressed and
unavailable for _informed_ criticism, as well.
Please note that I'm taking you at your word when you say:
Why worry about causing disturbances to perceptions I'm controlling? If
I'm controlling those perceptions, the disturbance will have
little or no effect on them. Don't you worry.
If you really are in conflict, a disturbance can make you angry. I hope
it doesn't.
···
------------------------------------
Martin Taylor (970411 10:11) --
> Here's how I think e-coli learning works. I'm sure Bill or Rick
> will correct me if I am wrong.Here I am:-)
Your view of e-coli seems, I if I understand you correctly, to be the way
I used to think about it before Bill P set me straight. Perhaps I'm still
wrong about it, but let's see.
> (4) Procedure:
> (a) Evaluate the criterion at the current location.
> (b) Make a small change by moving the location of the point in a
> random direction.
> (c) If the criterion gets worse, return to (b).
> (d) If the criterion gets better, make another change in the same
> direction againStep (c) is note quite right. The _rate_ at which random changes in
weights (that determine that value of the "criterion" -- really the
controlled perceptual variable) occur is determined by the size of the
error (the difference between "criterion" and reference setting).
I think this refers to random weight settings, not to e-coli, doesn't it?
That's a different facet of reinforcement. Even in your e-coli demonstration,
there aren't any changes at all when the moves are in an improving direction,
no matter how high the error (how far it is from the target). In random
weight setting, the changes lead to a static condition that may be better
according to the criterion, or may be worse. There's no sense of "direction."
With random weight setting, what you say is as I understand it, too.
Step
(d) is not quite right either. If the criterion gets "better" (error is
reduced) the _rate_ at which random changes in the weights occur
is reduced. When (if) the error is zero, the changes stop completely or
keep occuring at some VERY low rate.
I'm puzzled, then, as to how the e-coli keeps going on in a straight line
so long as things are getting better.
When we did our numerical analyses of the effect of dimensionality on the
e-coli procedure, random changes of direction occurred only, and always,
when a "step" caused a worsening of the criterion. No change of direction
occurred for the next step if the criterion got better. That was what we
did as a consequence of detailed discussions with Bill P. All the data
we posted was based on this procedure.
What you say about rates is what I thought applied to the random changes
in the sign-setting or relinking aspects of reorganization, not to the
Hebbian changes exemplified by what I understand to be e-coli.
----------------------
As a separate issue, I can imagine a random weight-setting process applied
at a second level to the version of e-coli I described. The criterion for
this second level would be the rate of improvement at the first level. The
faster the improvement, the less frequent any change in direction. That
might make it even more efficient. Interesting to try.
----------------------
I remain, sir, most respectfully yours in puzzlement.
Martin