Re: Representationalism rescues reinforcement learning

Allan C Cybulskie wrote:
On May 27, 3:20 pm, Wolf <ElLoboVi...@xxxxxxxxxx> wrote:
[...]
Elizabeth Loftus's experiments (replicated by many other researchers)
have shown that _any_ testimony based on personal experience is
unreliable. That is difficult to explain if memory is a matter of
calling up in the mind's eye some record of what was seen, for example.

Not if memory is rebuilt from pointers, which seems likely in light of
those problems.

"Rebuilding from pointers" is precisely what conditioned behaviour is. Why do I claim that? Because a conditioned behaviour is the system's response to one or more environmental inputs (the context). A behaviour such as swinging at the golf ball is "rebuilt from pointers" every time you do it. If that weren't so, you could not a) adapt the swing to suit a hole you've never played before; and b) adapt the swing to the inevitable differences in lie as you play the same course time after time; c) etc.

Consider an act of "spontaneous recall." There is no such thing. It's notorious that a whiff of freshly baked bread or stale sweat can trigger the memory of grandma's kitchen or a minor league hockey game. IMO, all recall is a response to some stimulus.

IOW, I see no difference in principle between "remembering what I saw" and reporting on it, and "remembering what I did", and doing it again.

But it is easy to explain if remembering is conceived as a behaviour
like any other, and therefore under stimulus control. In that case, each
act of remembering is shaped by the context in which the remembering
occurs. Interrogators have, nor surprisingly, long relied on this fact.
That's why confessions are suspect in and of themselves, and courts go
to some lengths to assure themselves that the confession in question is
not tainted by unacceptable practices.

The problem is that this seems far less than informative, since the
problem is going to be identifying the contexts that matter.
Obviously erroroneous memory is a problem for behaviourist accounts as
well ... since sometimes it's incredibly accurate and sometimes, well,
LESS than incredibly accurate [grin].

The problem is indeed that of identifying the contexts that matter. Without them you haven't a hope of deciding which acts of remembering are good evidence for past events and which are not. I just don't think that talking about representations, accurate or not, is of any help. It glosses over that very difficulty that you mention. It does not help Joe et al figure out how to build a machine that remembers, as distinct from a machine that replays a recording.

Erroneous memory is no problem for behaviourists. Loftus has confirmed what interrogators and survey makers have known for a long time: that the way you ask the questions determines the accuracy or otherwise of the recall. Eg, "Did you see someone in the hall?" will get a "yes/no/not sure." "Whom did you see in the hall?" will get a "X/Y/Z/couldn't tell" surprisingly often even when the correct recall would have been "Nobody." A question is obviously a controlling stimulus.

It's also been shown that people can come to recall something that never happened in a very short time of questioning. IOW, the recall of what didn't happen has been conditioned. If recalling what did not happen is conditioned behaviour, then so must be recall of what did happen - which is why recall is unreliable.

Bottom line: the very fact that erroneous memories occur indicates that memory is not "representation." You can if you wish identify the neural nets that are activated in some act of recall with a "representation" of the content, but IMO that's stretching it. The pattern of the net cannot be transformed part for part into an image the way the pattern of some segment of RAM can be transformed part for part into an image on a screen.

--

Wolf

"Don't believe everything you think." (Maxine)
.

Relevant Pages

  • Re: Unconscious Processes
    ... subconsciousness is determined to be the cause of the feeling. ... recalled memory. ... And to calm the mind to allow this recall is called ...
    (comp.ai.philosophy)
  • So ..what have you been up to DaveU - Reintro?
    ... In 2000 I flew my bike across to the USA and toured 10 states and did around ... Mind you .....I recall most of the guys seem to recall my girlfriend of the ... time, Michelle, more than they did me so it's not just my memory at fault!. ... Anyway Michelle and I have parted company, amicably ...
    (rec.motorcycles.harley)
  • Re: neuropsychology and decision making
    ... into their unconscious memory store to wrestle it out unless they use ... By the time they recall that extra option, ... The number of possible choices are limited --potential choices set by ...
    (talk.origins)
  • Re: Short Term Memory Retention - Huperzine A
    ... why don't you keep your holistic crap to your ... I don't know if I broached the subject of short term memory, ... I don't recall why but it was important. ... certificates, might have been diplomas. ...
    (alt.support.mult-sclerosis)
  • Re: RosAsm is a broken pile of crap
    ... >>> but not the binary search way. ... i recall having had discussions with him about ... access would vaste an unacceptable lot of Memory. ... > Of course the actual handling of messages can matter a lot, ...
    (alt.lang.asm)