On 13 Aug 2014, at 21:35, meekerdb wrote:
On 8/13/2014 6:26 AM, Bruno Marchal wrote:
On 12 Aug 2014, at 11:24, Bruno Marchal wrote:
Hi,
I think it is better I let you discuss a little bit.
Yes Russell made a nice introduction to this problematic.
Below, I just put a <comment>, which you might try to guess from
my preview post (notably to Brent) on this issue.
On 12 Aug 2014, at 02:48, meekerdb wrote:
On 8/11/2014 4:03 PM, LizR wrote:
I have never got this idea of "counterfactual correctness". It
seems to be that the argument goes ...
Assume computational process A is conscious
Take process B, which replays A - B passes through the same
machine states as A, but it doesn't work them out, it's driven
by a recording of A - B isn't conscious because it isn't
counterfactually correct.
I can't see how this works. (Except insofar as if we assume
consciousness doesn't supervene on material processes, then
neither A nor B is conscious, they are just somehow attached to
conscious experiences generated elsewhere, maybe by a UD.)
It doesn't work, because it ignores the fact that consciousness
is about something. It can only exist in the context of thoughts
(machine states and processes) referring to a "world"; being part
of a representational and predictive model. Without the
counterfactuals, it's just a sequence of states and not a model
of anything. But in order that it be a model it must interact or
have interacted in the past in order that the model be causally
connected to the world. It is this connection that gives meaning
to the model. Because Bruno is a logician he tends to think of
consciousness as performing deductive proofs, executing a proof
in the sense that every computer program is a proof. He models
belief as proof. But this overlooks where the meaning of the
program comes from.
<can you guess my future comment here?>
You might say that I model mind and belief ([]p), and the whole
working of a computer, by proof in arithmetic, but with comp, it
has to be valid at the correct susbstitution level for the case of
correct machines.
But for consciousness, I model it by knowledge (modal logic S4),
which I obtained from the Theaetetus' method applied on belief/
provability ([]p & p). Then I insist that this is not modeled by
*anything* you can define exclusively with 3p terms.
Insisting sounds like an attempt to cut off debate.
I insist because I see that people forget this very often, or perhaps
have not yet realize the impact of the fact that incompleteness makes
[]p and ([]p & p) obeying quite different logic, and have a quite
different nature ([]p is definable, []p & p is not: we can only
describe it for each individual arithmetical sentence p).
You model it by true belief and then you say belief is an attitude,
which is implicitly a conscious thought or feeling.
Only when you are conscious of believing something. The belief itself
might be purely representational and not conscious in general. You
have to distinguish "the machine believes p = []p", with the fact that
such a belief can be known by the entity which is [][]p & []p, that [k]
[]p, wit [k]p = []p & p. For Löbian entity we have indeed thvalid
inference: []p / [k][]p.
Indeed that (meta) definition, in the comp-arithmetical frame makes
knowledge being a non propositional attitude.
Yet it depends on the proposition being believed being true.
Yes. Why should that be a problem?
I am even led (may be influenced by salvia) to the idea that
consciousness is more in the "& p" than in []p. The proof aspect of
the brain machinery would be a filter of consciousness and memory,
and consciousness itself is more in the complementary of what is
provable. The meaning comes from truth, not proof.
Which leads to my position, which is that epistemology precedes
ontology.
Well, it looks like the comp assumption leads to this, with ontology
being the physical observable. But I am astonished that you say this.
If epistemology precedes ontology, physicalism is directly false. It
might help you axiomatize a little bit your position. I can understand
that epistemology precedes the physical ontology, but with comp,
epistemology is defined by the knowledge of the machine, ([]p & p),
and to define "[]p" you need to assume elementary arithmetic,
universal numbers, etc.
How do we determine what is true and worthy of belief.
We can't, by incompleteness. But "true" is still well defined, and can
be defined in set theory. p is true if it is satisfied in the
mathematical structure (N, +, *) studied in high school and used in
math and physics all the time.
I think we can never be certain,
That's a theorem in comp.
so it is a question of degrees of belief and weight of evidence.
But those don't fit in your scheme which depends on necessity and
truth.
Only because I define the "probability one", but it is not a certainty.
People that want to deny computers can be conscious point out
that the meaning comes from the programmer. But it doesn't have
to. If the computer has goals and can learn and act within the
world then its internal modeling and decision processes get
meaning through their potential for actions.
OK.
This is why I don't agree with the conclusion drawn from step 8.
I think the requirement to counterfactually correct implies that
a whole world, a physics, needs to be simulated too, or else the
Movie Graph or Klara need to be able to interact with the world
to supply the meaning to their program.
Hmm... If you want the full counterfactualness of a *universal*
machine, you need the UD*, and thus the full sigma_1 complete
reality, which is tiny compared to the full arithmetical truth, but
still infinitely bigger than known physical universe.
I don't understand what "full counterfactualness" means.
It means having the correct output for all input possible. Imagine a
car which works fine as long as the car does not speed beyond 140 km/
h, but breaks the motor above. In that case we have a car with a
counterfactual correctness restricted below that limit speed. It is
not full counterfactualness.
If for handling the counterfactualness you need (like Russell seems to
say) the presence of the counterfactual computations, then the
universal machine, to be counterfactually correct, needs the whole UD*.
A human being or any physical system reacts to the world in one way
or another. What was asked was for was counterfactual correctness,
i.e. the the MG reacts the same as would the conscious being
emulated - which might be no change at all.
I agree with you. The counterfactualness needs "If we change the input
then the output will change in the relevant way". But I am not sure
that we need the actual (physical) counterfactual behavior. I might
differ from Russell here.
Then with a computer, the interaction with the environment are
finite, and can be re-entered in the computer machinery locally,
and the MGA can be resume again.
They are finite over a finite time span, limited by the speed of
light. But the number is very large for even the shortest time
period that could be called "conscious", and my point is that this
is a physical context. It can't be just any random counterfactuals,
but only those consistent with "meanings" instantiated in the brain.
But you need to define and explain how that "physical" plays that
role, without making it violating digital emulability of your brain.
Then the MGA shows that this will invoke magical attribute to some
primitively physical matter, making hard to still say "yes" to a
doctor following the computationalist reason to do so. You might as
well say that a white male God is needed to make that primitive matter
real.
But if the Movie Graph computer is a counterfactually correct
simulation of a person within a simulated world, there's no
longer a "reversal".
Why?
Because it doesn't make the physical derivative from the
pyschological, but rather makes them interdependent and both
derivative from computations.
OK. Then we agree it seems to me.
Bruno
Brent
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
http://iridia.ulb.ac.be/~marchal/
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.