From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Fri Oct 26 2001 - 17:57:53 MDT
J. R. posted a note about a review of a recent book by Pierre Baldi:
An included quote:
> Baldi adds a number of intriguing back-of-the-envelope calculations to
> this discussion, including an estimate of how much information is
> contained in a lifetime of experience, or brain inputs: By his estimation,
> the amount is 2.2x10^18 bits (about 2.7x10^17 bytes), which at current rates
> of memory storage advances will be within the memory capacity of a typical
> personal computer in 27 years. Based on this, Baldi then investigates the
> information size of what he calls the external self, a complete genomic
> sequence together with the recording of all inputs and outputs of an
> individuals brain over a lifetime, which he estimates to be still on the
> order of 10^18 bits, within reach of modern computer technology. The
> internal self, the amount of information required to capture the relevant
> structure of your brain that defines your identity, is largerat least 10^27
> bits but might still be within reach of emerging technologies.
People should take these numbers with a *very* large grain of salt
for they *are* back-of-the-envelope calculations.
Based on [1], human long term memory capacity is no more than of the
order of 10^9 bits (a few hundred megabytes). This is for "recall"
memory, which may be somewhat different from "recognition" memory.
It is very doubtful that humans store a "lifetime" of experience
(e.g. ~10^18 bits). It is highly likely that chunking and compression
occurs. I'm also very doubtful of the 10^27 bits of structural information.
At ~10^10 neurons and 10^4 synapses per neuron that only gets you to 10^14
connections. I'm very doubtful that there is 10^13 bits of "significant"
information per synapse. I'd be interested in what Anders thinks.
Robert
[1] Landauer, T. K., "How Much Do People Remember? Some Estimates of the Quantity
of Learned Information in Long-term Memory", Cognitive Science 10:477-493 (1986).
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:11:41 MST