[ExI] Brain emulation, regions and AGI [WAS Re: Kelly's future]

[Richard Loosemore]

Kelly Anderson wrote:
> If I understand what you're saying, the cortex is structurally uniform
> (other than perhaps connection patterns in the dendrites, which is
> what I think you may be saying about within-column) but that different
> functions somehow manage to navigate themselves into similar areas
> across individuals. That's extremely interesting.
> 
> The cortex is just one brain area. If you look at the brain,
> http://en.wikipedia.org/wiki/File:Human_brain_left_dissected_midsagittal_view_description_2.JPG
> it seems pretty clear that there are structures or regions that are
> quite distinct. One would have to be crazy to assume that they did not
> have distinct purposes.
> 
> http://en.wikipedia.org/wiki/Brainstem
> http://en.wikipedia.org/wiki/Putamen
> etc.
> 
> So, if I'm understanding right, what you are talking about is just in
> the cortex, right?

Yes.  As far as I can see, the cortex is where most of the higher-level 
action occurs, while the other stuff (admittedly, a lot of other stuff) 
is either specialized support machinery (e.g. hippocampus) or more 
primitive autonomic mechanisms.


> That's extremely interesting. I've heard that the brain has three
> overall structures, the reptile brain (roughly the brain stem), the
> mammalian brain (cortex), and the distinctly human part of the brain
> (neo cortex). Are you talking about the mammalian part, the human part
> or both?

I think the basic functionality is common to the mammals.  Quite what it 
is that we humans have that is extra, I am not sure:  it looks to me 
like a souped-up version of the same basic design.  So when I talk about 
the columns, I mean the cortex generally, rather than the neocortex in 
particular.

It is worth me being balanced here and saying that not everyone accepts 
this type of interpretation.  Even the whole idea of columns is 
sometimes controversial.  For a perspective, see

http://www.pnas.org/content/105/34/12099.full



> And would it not make sense that part of our thought processes take
> place outside the cortex?

Yes, that is true.  But it does depend on what you mean by thought 
process.  Example cases:  when people are overtaken by anger (or other 
overwhelming emotion) they are quite probably being driven by modules 
below the cortex.  Also, the state known to psychologists as "romantic 
love" is pretty much a state of total insanity controlled by some 
structures whose only purpose is to subvert the reasoning faculty and 
make the person becoming obsessed with a single other individual.  And 
so on.

Motor control, also, can be disturbingly autonomous.

>>> Aren't there fully functional computational models of parts of the
>>> brain now? Aren't those models based on bottom up analysis, rather
>>> than top down?
>> There is a model of the cerebellum, but that really is a separate, fairly
>> simple function.
> 
> Oh, I wish I understood more about all this.

For the cerebellar model, you should grab Marr and Nishihara or Marr and 
Poggio (can't remember which it is).  From the point of view of a 
mathematician, it is very elegant.



>> If you are talking about the wiring diagrams that have recently been
>> announced, I believe you will find that all those announcements are kind of
>> sneaky:  what they actually mean by building a computational model is that
>> they have *sampled* the neurons and patterns of wiring in a small area, and
>> then done a *statistically* accurate reconstruction of that area.  I
>> consider that to be a cheat.
> 
> Even if it is a cheat, it might be useful. Time will tell.

I don't know.  Try to imagine that you were trying to understand how a 
complex floating-point math engine worked (silicon-chip level), and 
someone gave you a randomly wired network of transistors that had 
similar statistical activation patterns to the real thing.  To my mind, 
I don't think I could learn anything at all from that.  I think it might 
just confuse me and distract me from trying to understand the true 
functionality.


>> I am less sure whether anyone has done a real circuit diagram or model.
>>  Because all these announcements and press releases tend to be fuzzy on the
>> details, it can be very frustrating to try to find out exactly what level of
>> detail they claim to have done.  To the best of my knowledge, ALL of the
>> current claims about having bottom-up models of parts of the brain are
>> "cheats" in the above sense.
> 
> I am less interested in cheating than in utility... :-)
> 
> Kurzweil talks about areas of the auditory channel that have been
> fully emulated, among others. Do you have any comment along those
> lines?

I don't know about that work, but the general rule is that if it is 
peripheral processing (input or ouput pathways) it is probably good 
science.  So I would not be surprised if they had good models of that 
stuff.  The availability of meaningful signals coming in (or going out) 
means that it is easier to do a trace and understand the next few levels 
of processing of those signals.  The difficulty comes when you get 
deeper in ..... which means that success in projects like deciphering 
the auditory processing layers is not an immediate harbinger of success 
for more abstract stuff.

What seems to happen is that the neuroscience folks get some success 
with the periphery processing, then they try to understand deeper 
structures and, lacking a fundamental grasp of the psychology, they seem 
to revert to the only kind of psychological theory that they can get 
their heads around in a hurry.  Which, alas, is the stimulus-response 
kind of theory .... aka behaviorism or conditioning.  This is sad, 
because there are good reasons why behaviorism was killed off as a 
viable approach to psychology, some fifty years ago.

Pity you're not near upstate NY, because I am writing a brand new course 
on Neural Nets and Cognitive Systems, to be given at Wells College in 
the fall semester.  I'm going to be hitting them with as much state of 
the art as I can, so it should be fun.



Richard Loosemore