[ExI] The digital nature of brains (was: digital simulations)

[Eric Messick]

Stathis writes:
>I don't see any problem in principle with the human being the whole
>system but not understanding what he is doing.

In principle, no problem.  I was addressing a problem of scale.

Any program capable of passing the Turing Test is going to have to be
pretty complicated.  Humans don't do a very good job at simulating
computers, and are really bad at it if you don't even allow them a
piece of paper to scratch some notes on.  As a result, a human just
sitting and thinking about being a computer running a program which
passes the Turing Test is going to run that program very slowly.

Note that this is a problem with either the original Chinese room, or
the Chinese room subsumed into a man.

If you need to simulate at the neuron level (which is a conservative
estimate of what might be required), then the program has to keep
track of 100 billion neurons and their thousands of interconnections
each.  How many people are capable of remembering that much raw data
without any outside assistance?

None.

Basing an intuitive argument on something which is so far out of the
bounds of the possible is dishonest.  If Searle actually believes it,
he's profoundly deluding himself.

As I said, this same objection applies to the original Chinese room.
Even if we postulate an incredibly simple program, like Eliza for
example, a human simulating a computer running such a program would
take years to have a simple conversation.  The necessary computing
speed is off by at least a factor of a million.

I know that the argument is supposed to be a philosophical one about
the locus of understanding, but human intuition only works for things
at human time and space scales.

Make things a million times smaller and you start running into quantum
mechanics issues.  Make things a million times faster and you're
getting into the realm of relativity.  Neither of these follow human
intuition.

Let's look at what it would take to make a Chinese room run a neural
level simulation in real time.

We could probably get away with a base simulation rate of about 1000
Hz.  For each millisecond we'd need to perform at least one
calculation per synapse.  Figure 100 synapses per neuron, so that's
1E5 calculations per neuron per second.  If a human could perform
about 1 calculation per second, that's 1E5 humans per neuron.
Multiply by 1E11 neurons and we need about 1E16 humans to perform the
calculations.

There are currently about 1E10 humans, so we're talking about 10,000
times the population of the planet to simulate a single human.

Even if my numbers are *way* off, the endeavor is clearly ridiculous.
Intuition about how such a system might behave is profoundly silly.
Claiming that such a system cannot have a particular property is again
either dishonest or delusional.

-eric