Re: motor emulation, modularity, feedback, prediction, and dreams

"feedbackdroids" <feedbackdroids@xxxxxxxxx> wrote:
> Curt Welch wrote:

> > I don't yet know how useful my approach is, but I think it shows that
> > there are ways to solve this internal model problem that puts no real
> > performance problem on the reaction speed of the system. The main
> > point being, is that the model ends up being structured in ways that
> > don't look anything like the world models we like to talk about. And
> > the big reason for that, is that the best way to model the world, is to
> > not model the world, but to instead, model how the agent needs to react
> > to the world. It's harder for us to think in those terms, but not at
> > all hard, for a learning system, to be set up to learn everything in
> > those terms.
> >
> >
>
> "... the best way to model the world, is to not model the world, but to
> instead, model how the agent needs to react to the world ...".
>
> I think this is a good point that needs be thought about some more
> However, this still sounds very low-level, and I see it fitting into a
> broader scheme. Right now, there is a large gulf between the Brooksian
> bottom-up reactive concept, and the GOFAI top-down symbolic concept.
> There are many levels, I think, that need to be interposed inbetween.
> Minsky started to talk about this in his 1991 paper regarding combining
> neural nets and symbolic processing, and more recent papers on
> multi-level AI's. I see the motor emulator concept as a useful
> inbetween step.

Yeah, there's a large gap still. My belief however is that it will (and
must be) bridged from the bottom up. When it happens, it's going to
re-write everything we have already tried to do top down - but got wrong
because we didn't do it in a way that was compatible with the foundation
that had not yet been laid. I also believe that while everyone has been
trying to engineer the top, what we will find is that it must be learned,
and not engineered.

What has to be done, is to engineer the learning technolgy, and to optimize
it's performance to what we want it to learn (not to mention assist it by
adding whatever additional infrastructure is needed to simplify and scope
the learning problem for it).

Once we figure out how to do that, we will end up wioth GOFAI-like
processing machines at the high level, but they will be based on a very
different foundation from the GOFAI of years past.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: DTV antennas?
    ... the other engineer had a PE license. ... illiteracy is rampant, as is innumeriacy. ... ways of learning things. ... communications problems. ...
    (sci.electronics.design)
  • Re: DTV antennas?
    ... the other engineer had a PE license. ... illiteracy is rampant, as is innumeriacy. ... learning a completely no concept quite painful. ... documentation", but there was no way I could have found it in there. ...
    (sci.electronics.design)
  • Re: DTV antennas?
    ... the other engineer had a PE license. ... illiteracy is rampant, as is innumeriacy. ... ways of learning things. ... I find myself reading the manual ...
    (sci.electronics.design)
  • Re: IEE or IEEE
    ... For my part, as a Systems Engineer, I have certainly learned a lot about ... Torque Measurements, Power Control, PLC programming, Forth Programming and ... day-by-day learning, whatever way they come to you. ...
    (comp.arch.embedded)
Loading