Xref: utzoo comp.ai:8288 sci.bio:4209 sci.psychology:3937 alt.cyberpunk:5442
Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!hsdndev!wuarchive!julius.cs.uiuc.edu!apple!portal!cup.portal.com!mmm
From: mmm@cup.portal.com (Mark Robert Thorson)
Newsgroups: comp.ai,sci.bio,sci.psychology,alt.cyberpunk
Subject: Re: The Bandwidth of the Brain
Message-ID: <37273@cup.portal.com>
Date: 27 Dec 90 02:03:11 GMT
References: <37034@cup.portal.com>
Organization: The Portal System (TM)
Lines: 69

Jim Lai says:

> Not so.  If it is a pipeline of "agents", you only know the maximum speed of
> any individual filter along the pipeline.  (A chain is only as strong as its
> weakest link.)  Each agent can encode data minimally for compactness in
> theory, but that doesn't mean that the brain is therefore optimal and actually
> only uses 50 bits per second bandwidth.  There's nothing wrong with a model
> that assumes the sending gigabits per second down the pipeline.  In the agent
> paradigm, each agent can feed output to many agents.  A tree ten levels deep
> can still be quite wide, and be particularly nasty if when cycles are added.
>  
> My gripe with experiments that claim to measure the bandwidth of the brain
> is the validity of their measurement criteria, which may have already
> assumed a fair amount of preprocessing to have taken place.

This posting and a later one from Fred Sena hit on the weakest point of
the argument I presented in my original posting.  If we are to talk
intelligently about the bandwidth of the brain, we need to recognize the
distinction between bits and bauds.  Bits are information, bauds are some
sort of lower-level phenomenon which may (but not necessarily does) carry bits.

For example, let's say I developed a program to enhance the reliability of
serial communications by using triple redundancy.  This program takes an
input message such as "hi, mark" and converts it to "hhhiii,,,   mmmaaarrrkkk"
for sending over the communication channel.  On the receive end, another program
performs the reverse process, using the redundancy to correct for any errors
that occurred during the transmission of a single character.

Now what would you say the "bandwidth" of this transmission mechanism is?
It certainly requires triple the bandwidth over the communication channel,
but is triple the amount of information being sent?  I think not, the bauds
have been tripled but the bits have stayed the same.

Likewise in the brain we see enormous neural structures used to perform
low bandwidth functions like reading and listening.  Do these structures
have some incredibly high internal bandwidth not evidenced in either the
input or the output?  Again -- for the same reason -- I say no.  The amount
of _information_ has stayed constant, even if it temporarily fanned out
into some highly decoded (i.e. redundant) representation.

I will admit there is bandwidth which is not visible at either the
input or output ends.  For example, a single move by a player playing chess
is an input which results in a single responding move by the opponent.
Each move is an event with very low information content, but a great deal
of internal processing takes place during the ten minutes the master
chessplayer spends deciding on his next move.  But this bandwidth is not
so exceptionally high when compared to other intense human activities like
reading, writing, or speaking.  The chessplayer examines each possibility,
one at a time, at a very human rate.  It would not slow down the chessplayer
very much to tell you what he is thinking as he thinks it.  There is no
mega-bandwidth simultaneous perception of the entire chess position resulting
in a responding move in a single clock cycle.  Instead, he thinks, "I can
move here.  No.  Here.  No.  Here.  Hmm, that's interesting -- then he moves
there and I go here and he goes either there or there."

It's also true that there is some unconscious pre-processing which takes
place without thinking.  For example, the chessplayer excludes moves involving
a trapped piece (such as a rook boxed into a corner) from his consideration
unless a reasonable scenario involves removing the obstacles to moving that
piece.  Likewise, while reading you take in the words but ignore the specks
of wood pulp embedded in the paper, unless those specks make a letter 
difficult to recognize.

Should this pre-processing be counted in the bandwidth?  Once again, for the
same reason as before, I say no.  To say otherwise is like saying that when I
read text at 1200 baud I'm actually reading faster than 1200 baud because I'm
managing to ignore the dust on the face of my CRT.  It is like saying my 9600
baud modem is faster than 9600 baud because it manages to ignore random clicks
and pops on the phone line.