Re: 1000 humans in a grain of salt

Anders Sandberg (
25 Jan 1998 00:58:47 +0100

Henri Kluytmans <> writes:

> Anders Sandberg wrote:
> >I think this is wrong, since the genetic information is just the basic
> >rules to grow the brain, the exact structure is laid down in an
> >emergent fashion most likely very dependent on the environment; brain
> >growth creates information.
> I agree. The problem, ofcourse, is that inputs arrive even during
> the embryonic growth of the brain. However (theoretically) even
> these imputs could be simulated.

My guess is that these inputs are fairly easy to deal with compared to
the inputs that appear later; at this point they are mainly chemical
and temperature signals. Things get more complex later.

> >That makes 10 million axons of sensory information. The
> >normal activity for neurons lie beneath 1 KHz, so an upper bound on
> >the input would be 10 billion bits/s.
> When uncompressed...

The big question is of course the coding. If it is just frequency
coding, then we will have a huge compression ratio. Phase coding would
complicate things a bit. But assuming only spikes are relevant, then
one could code a signal as (spike at axon N at time T), replacing
around a thousand values per axon per second with tens to a hundred of
values (depending on activity), creating an compression ratio of

(but never trust this kind of computer science estimate for the CNS; I
have already started to think that nature has gone out of its way to
surprise and complicate things for us scientists)

> >My personal estimate for the human capacity is around a bit per
> >synapse (based on assumptions from neural networks), there are around
> >10^14 synapses, making a capacity of around 10^14 bits.
> And what about the information required to store which connection
> the synapses belong to. I.e which other neuron each dendrite is
> connected to.

Good point, I overlooked that. That would take log2(10^11) =~36 bits
per synapse. Interesting, it seems that there may be more information
in the connectivity than in the exact synaptic values.

> I assumed the neural network of the brain can be
> simulated by a deterministic system. (i.e. that a selfconsiousnes
> intelligence can be created by a deterministic system)
> This has ofcourse not been proven but seems quite likely.
> Some persons claim that some internal randomness should be
> inherent for generating selfconsiousness intelligence.
> However others claim that the random noise input from the
> environment is sufficent.

Just having internal randomness is not enough, since it can be
emulated by a deterministic system with a look-up table. It seems
obvious that one random look-up table would do just as well as another
with the same statistical properties, so in that case it could be
generated using pseudorandom numbers.

> That the human consiousness could be transferred to a
> deterministic system is unacceptable to some people
> because it kind of reduces themselves to machines...?

Yes, many people seem to dislike the idea immensely. I see it the
other way around: instead of seeing us as mere machines, the "mere
machines" can in fact be equal to us!

Anders Sandberg                                      Towards Ascension!                  
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y