Re: poly: Software for superintelligence

From: Nick Bostrom <>
Date: Thu Jan 08 1998 - 15:23:08 PST

carl feynman <> wrote:

> At 11:38 PM 1/7/98 +0000, you wrote:

> In
> another model, which I will call model B, the large-scale structure of the
> cortex is fairly homogenous, but there are lots of possible types each cell
> can be recruited into depending on signals it recieves from its
> environment. The cell types are regulated by mutually stimulating and
> inhibiting sets of regulatory proteins, much like the reguatory cascades
> that decide whether a fetal cell will be a liver cell or a spleen cell.
> Depending on type, a particular cortical cell will have different
> thresholds and electrical behavior, grow synapses onto other cells in
> different patterns, display different recognition proteins on its surface,
> and perhaps even release and respond to different neurotransmitters. In
> order to make this model falsifiable (:-), here's two things neurons can't
> do: grow new long-range (more than 1 mm) connections, and change from one
> cell type to another, except along a one-way branching path.

What this modal B essentially says is that the resulting fine-grained
cortical architecture is a function of both genes and behaviour. I
think there is excellent reason for holding B true, at least
for most of the human neocortex. (In some more primitive regions it
is possible that the environment doesn't play an important part.) The
three arguments that I gave do however apply also in this model B.
They would then indicate that the genetical contribution is not too
difficult to replicate on a computer. So I do not think that my
arguments fail because they attack the straw man model A.

> Let's look at your three points:
> >First, consider the plasticity of the neocortex, especially in
> >infants. It is known that cortical lesions, even sizeable ones, can
> >often be compensated for if they occur at an early age. Other cortical
> >areas take over the functions that would normally have been developed
> >in the destroyed region.
> Model B explains this by saying that recruitment to various cell types
> happens largely during infant development. Cells are recruited from a
> less-differentiated pool into more differentiated states.

And that indicates that it is the sensory input that drives the
differentiation, doesn't it? Since no new neurons are produced after
birth, the distribution of genes that are present in a given cortical
area remains constant throughout infancy and beyond. Nonetheless,
that area can support either visual or somatosensory processing (for
example), depending on sensory contingencies. This shows that what
takes a new-born's cortex and makes it into something that can see or
can feel touch is sensory input; which, of course, can easily be
provided artificially. So people who believe that providing the
initial structure will be extremely difficult because it is so
complicated and determined by a big set of genes will have to
consider that whatever contribution these genes make, it cannot
amount to more than the greatest common denominator between vision,
audition, and somatosensory processing -- and that denominator seems
to contain relatively little information about the world, since these
sensory tasks are so different. -- That, in outline, is the argument;
though one also need to consider different modules of abstract
reasonsing and the fact that transferrability of tasks is not

> >The second consideration that seems to indicate that innate
> >architectural differentiation plays a relatively small part in
> >accounting for the performance of the mature brain is the that as far
> >as we know, the neocortical architecture in humans, and especially in
> >infants, is remarkably homogeneous over different cortical regions and
> >even over species:
> This is explained just fine by model B. In fact, the less-differentiated
> cortex of infants is exactly what the model would predict.

Sure, but it speaks (weakly) against a hugh essential initial
architectural complexity.

> As far as I know, the wiring diagram of any particular
> cortical column is not known to any degree of detail. If, for example, one
> column contained rings of six neurons, each neuron of which inhibited its
> neighbors, and another contained similar rings of five neurons, we would
> not yet be able to tell the difference. However, these rings are
> functionally immensely different: one is a flip-flop and the other is an
> oscillator. I may be wrong about this; perhaps someone who has advanced
> professional Swedish knowledge of neurophysiology can comment.

Here you are: It all depends on the type of neuron and the input
they get from elsewhere. For example, both the five-neuron and the
six-neuron ring could be a flip-flop if all of them are excitatory,
with some constant inhibitory input from elsewhere. About your
general point, it is true that we don't know very much about how
small architectural alterations within the columns would affect the
global properties of the network. My personal view is that many small
changes wouldn't matter much, and part of my reason for this view is
the general noisiness of the CSN makes it impossible to have much
depend on some small detail.

-- Ok, "advanced professional" might be to strech it a bit ;-) (I
have written an MSc in computational neuroscience), but I'm certainly
Swedish all right and that's the main thing.

> >The third consideration is an evolutionary argument.
> Or perhaps you are saying that once we get human-level AI, getting to
> superhuman AI will be *relatively* easy? I'd agree with that.

I do say that but not in the section I posted. This evolutionary
argument would rather indicate that the step from chimp-equivalence
to human-equivalence is relatively easy.

> Here's an argument in favor of complexity, using a source of data you may
> not have considered.

Interesting. The question is what these 13,000 genes achieve. I have
always been amased at the ability of genes to code for such inborn
seemingly high-level properties as perceptual recognition of an eagle
in some animals, or sexual preferences in humans if these are partly
determined visually.

Nick Bostrom
Received on Thu Jan 8 23:16:47 1998

This archive was generated by hypermail 2.1.8 : Tue Mar 07 2006 - 14:45:29 PST