<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
Eugen Leitl wrote:
<blockquote cite="mid:20110629085740.GN26837@leitl.org" type="cite">
<pre wrap="">On Tue, Jun 28, 2011 at 02:49:18PM -0400, Richard Loosemore wrote:
</pre>
<blockquote type="cite">
<pre wrap="">Well, all I can say that is "definite" is that the above statement is
definitely garbage. ;-)
</pre>
</blockquote>
<pre wrap=""><!---->
It would help if you realize that there are juste a few GBytes
of memory with some 10 GByte/s bandwidth in a current PC.
This only applies to streaming, random access is much, much
worse. Graphics accelerators only add an order of magnitude.
</pre>
<blockquote type="cite">
<pre wrap="">Until you understand what is involved in the mechanisms of intelligence,
</pre>
</blockquote>
<pre wrap=""><!---->
I undestand enough of neuroscience to know that it currently
takes the whole Earth's hardware to match what the space
between your ears does.
</pre>
</blockquote>
<br>
No, that is not true, because everything depends on the relevant
FUNCTIONAL level at which the emulation has to take place (a point that
I made, IIRC, on this list in the last few months).<br>
<br>
Are you duplicating the structure of the human brain at the quark
level, because you reckon the quark level to be the relevant functional
level? Why not?<br>
<br>
You assumed ...... what? ..... that it had to be duplicated with
accuracy down to the neuron level? Synapse level? Molecular level?
On what basis would each of these choices be made (i.e., what forces us
to assume that proper functionality requires that level of detail)?<br>
<br>
The point is that there are choices here. My own work tends to
indicate that something around the cortical column level of
functionality would be sufficient for most of the processing. Now, if
that were true then an AGI could very well be built using something
within an order of magnitude of a current PC.<br>
<br>
I don't think anyone pays sufficient atention to this little issue.
Does anyone reading this post truly realize that the implication is
that if cortical columns are the relevant functional level, and if
someone figured out what the functionality is, that means someone could
build an AGI that would fit in (probably) a single server rack, and do
so by the end of the year...? <br>
<br>
<br>
<br>
Richard Loosemore<br>
<br>
</body>
</html>