From: Emmanuel Charpentier (emmanuel_charpentier@yahoo.com)
Date: Mon Aug 24 1998 - 03:08:58 MDT
---Anders Sandberg <asa@nada.kth.se> wrote:
>
> Emmanuel Charpentier <emmanuel_charpentier@yahoo.com> writes:
>
> > Another solution would simply be to have only one neural net (the
> > original :) but many thought processes. The configuration/topology
of
> > the neural net is mainly the long term memory, the firing of neurons
> > and synapses is mainly the working memory, so you just need to
> > duplicate the later one... no? What you obtain is layers of persona.
> > You can have 1000 thought processes and just one neural net
> > configuration.
>
> Nice idea, although I don't believe in it. It seems that the
> distinction between topological long-term changes and short term
> activities is rather blurred; most likely there are processes on all
> scales, making the short-term effects (calcium levels, dynamical
> synapses etc) necessary for longer-term effects (LTP, synaptic
> movements) and permanent storage. So the thought processes would have
> trouble with their long-term memory: each would suffer a kind of
> encoding impairment that would likely be rather distressing.
>
> > Of course, it could still leads to serious troubles: how do you
> > share important new informations? How do you change the
configuration
> > of the neural net (I guess sleep is a key)?
>
> Sleep is likely the key of long-term memory consolidation, but
> obviously we learn during the day before going to sleep too. Maybe, if
> we are lucky (but don't count on it, this is something I'm almost
> certain doesn't work) you would only need to deal with the medial
> temporal lobe memory system. Then each process will have its own copy
> of the MTL and during sleep the exoself could run consolidation
> between the MTL of process 1 and the shared cortex, then consolidation
> for process 2, and so on. But don't count on it, most likely low-level
> learning is going on across the cortex as a part of normal thinking.
Some solution might be to simply have each process modify the net
right as it does now. There is then no dilemna between short and long
term. Of course, it leads to something else, each process might modify
the same structure at once... I know we are pretty used to wild
things, and we are pretty much making up most of our memories from
bits of recalls here and then, and yet, it certainly is dangerous.
Two processes have only a small chance to modify the same things in
the same day (for a big net, not for a kid), but there might be ways
to reduce this likelyhood: lock parts of the brain until consolidation?
> > What do you do with your body, because you still have only one?
>
> Why only one? If you can do this with the neural nets, making extra
> bodies should be fairly simple (for those who want bodies. I would
> like to exist partially as processes on the net)
Or still have only one body, but try to have many thought processes
in the same brain... important is to make sure that input and output
neurones are in only one state, and we don't conflict too much (well,
more than usual) in our nerve and hormonal impulses...
> >Can a neural net layer function
> > normally without a body?
>
> Can a human function normally without a body? Most likely no, although
> the definition of normal may change. We are input-output driven, and
> without inputs we tend to fall asleep or get sensory deprivation
> problems, and without body feedback emotions become blunted. Most
> likely even completely virtual people will need simulated bodies, at
> least until they can be modified enough to handle new senses and forms
> of action.
It might be the only difference between humans, robots and cyborgs
for a long time: interfaces with the environment (and itself).
Do you know how the net incorporates new neurons and grow
synapses, axons, dendrites: is it chemically and topologically driven?
A function like: grow toward that chemical source at a rate
proportional to its importance?
Manu.
_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:49:30 MST