"Dreaming" Computers

Arthur T. Murray uj797 at victoria.tc.ca
Tue Jul 18 21:35:20 EST 2000


Chris North, chrisnorth at mac.com, wrote on Tue, 18 Jul 2000:

> Hello, 

> I have but an elementary understanding of Neural Networks,
> and this may well have been discussed on the group before.
> I apologise if this is the case. A quick question :

> During my research into this subject, I've been
> reading about the work of "Imagination Engines"
> ( http://www.imagination-engines.com )
> and their research into "dreaming" neural-networks
> to create new (inventive?) ideas from existing data.
ATM:
Bjorn Wesen in message <8l2s9n$qeq$1 at merkurius.lu.se>
has written about the down-to-earth neural-net aspects
of imaginative dreaming computers; I would like to add
that the dreaming itself already seems feasible within
http://www.geocities.com/mentifex/mind4th.html AI at Home.

The http://www.geocities.com/mentifex/theory5.html theory
behind Mind.Forth AI encompasses dreams as the involuntary
associative recall of not yet "passivated" memory engrams,
combined with newly generated responses to the engrams.

The primitive Mind.Forth code already includes a marker
to preserve the information of whether an engram came
in from the outside or was deposited during "re-entry"
(as described so well in the books of Gerald Edelman).

http://www.geocities.com/mentifex/acm.html DIY AI
includes step-by-step AI coding instructions into
which anyone may insert simple code for robot dreams.
Just close off the external avenues to the SENSORIUM
by analogy with a "reticular activating formation"
in http://www.geocities.com/mentifex/sleep.html
and think up an algorithm of letting "unsettled"
memories conjure up demons of dream and nightmare.

> One of the examples they give is "teaching" a network
> the rules of "hard" material compounds, putting the
> network into a pseudo-random dream state, and utilising
> a second neural network to watch for useful results.
> Apparently, a prototype of this system was able to
> "discover" within a few seconds new material compounds
> which had taken human beings years to realise. If this
> is true then the potential, it seems to me, is incredible.

> I'm being incredibly naive, right? :)

> Any thoughts welcome!






More information about the Neur-sci mailing list