Dear Jesse,
Thank you for this very nice remark. I will have to think about it
and read your reference.
On 8/23/2012 8:19 PM, Jesse Mazer wrote:
A quibble with the beginning of Richard's paper. On the first page it
says:
'It is beyond the scope of this paper and admittedly beyond my
understanding to delve into Gödelian logic, which seems to be
self-referential proof by contradiction, except to mention that
Penrose in Shadows of the Mind(1994), as confirmed by David
Chalmers(1995), arrived at a seemingly valid 7 step proof that human
“reasoning powers cannot be captured by any formal system”.'
If you actually read Chalmers' paper at
http://web.archive.org/web/20090204164739/http://psyche.cs.monash.edu.au/v2/psyche-2-09-chalmers.html
he definitely does *not* "confirm" Penrose's argument! He says in the
paper that Penrose has two basic arguments for his conclusions about
consciousness, and at the end of the section titled "the first
argument" he concludes that the first one fails:
"2.16 It is section 3.3 that carries the burden of this strand of
Penrose's argument, but unfortunately it seems to be one of the least
convincing sections in the book. By his assumption that the relevant
class of computational systems are all straightforward axiom-and-rules
system, Penrose is not taking AI seriously, and certainly is not doing
enough to establish his conclusion that physics is uncomputable. I
conclude that none of Penrose's argument up to this point put a dent
in the natural AI position: that our reasoning powers may be captured
by a sound formal system F, where we cannot determine that F is sound."
Then when dealing with Penrose's "second argument", he says that
Penrose draws the wrong conclusions; where Penrose concludes that our
reasoning cannot be the product of any formal system, Chalmers
concludes that the actual issue is that we cannot be 100% sure our
reasoning is "sound" (which I understand to mean we can never be 100%
sure that we have not made a false conclusion about whether all the
propositions we have proved true or false actually have that
truth-value in "true arithmetic"):
"3.12 We can see, then, that the assumption that we know we are sound
leads to a contradiction. One might try to pin the blame on one of the
other assumptions, but all these seem quite straightforward. Indeed,
these include the sort of implicit assumptions that Penrose appeals to
in his arguments all the time. Indeed, one could make the case that
all of premises (1)-(4) are implicitly appealed to in Penrose's main
argument. For the purposes of the argument against Penrose, it does
not really matter which we blame for the contradiction, but I think it
is fairly clear that it is the assumption that the system knows that
it is sound that causes most of the damage. It is this assumption,
then, that should be withdrawn.
"3.13 Penrose has therefore pointed to a false culprit. When the
contradiction is reached, he pins the blame on the assumption that our
reasoning powers are captured by a formal system F. But the argument
above shows that this assumption is inessential in reaching the
contradiction: A similar contradiction, via a not dissimilar sort of
argument, can be reached even in the absence of that assumption. It
follows that the responsibility for the contradiction lies elsewhere
than in the assumption of computability. It is the assumption about
knowledge of soundness that should be withdrawn.
"3.14 Still, Penrose's argument has succeeded in clarifying some
issues. In a sense, it shows where the deepest flaw in Gödelian
arguments lies. One might have thought that the deepest flaw lay in
the unjustified claim that one can see the soundness of certain formal
systems that underlie our own reasoning. But in fact, if the above
analysis is correct, the deepest flaw lies in the assumption that we
know that we are sound. All Gödelian arguments appeal to this premise
somewhere, but in fact the premise generates a contradiction. Perhaps
we are sound, but we cannot know unassailably that we are sound."
So it seems Chalmers would have no problem with the "natural AI"
position he discussed earlier, that our reasoning could be adequately
captured by a computer simulation that did not come to its top-level
conclusions about mathematics via a strict axiom/proof method
involving the mathematical questions themselves, but rather by some
underlying fallible structure like a neural network. The bottom-level
behavior of the simulated neurons themselves would be deducible given
the initial state of the system using the axiom/proof method, but that
doesn't mean the system as a whole might not make errors in
mathematical calculations; see Douglas Hofstadter's discussion of this
issue starting on p. 571 of "Godel Escher Bach", the section titled
"Irrational and Rational Can Coexist on Different Levels", where he
writes:
"Another way to gain perspective on this is to remember that a brain,
too, is a collection of faultlessly functioning element-neurons.
Whenever a neuron's threshold is surpassed by the sum of the incoming
signals, BANG!-it fires. It never happens that a neuron forgets its
arithmetical knowledge-carelessly adding its inputs and getting a
wrong answer. Even when a neuron dies, it continues to function
correctly, in the sense that its components continue to obey the laws
of mathematics and physics. Yet as we all know, neurons are perfectly
capable of supporting high-level behavior that is wrong, on its own
level, in the most amazing ways. Figure 109 is meant to illustrate
such a class of levels: an incorrect belief held in the software of a
mind, supported by the hardware of a faultlessly functioning brain."
Figure 109 depicts the outline of a person's head with "2+2=5"
appearing inside it, but the symbols in "2+2=5" are actually made up
of large collections of smaller mathematical equations, like "7+7=14",
which are all correct. A nice way of illustrating the idea, I think.
I came up with my own thought-experiment to show where Penrose's
argument goes wrong, based on the same conclusion that Chalmers
reached: a community of "realistic" AIs whose simulated brains work
similarly to real human brains would never be able to be 100% certain
that they had not reached a false conclusion about arithmetic, and the
very act of stating confidently in mathematical that they would never
reach a wrong conclusion would ensure that they were endorsing a false
proposition about arithmetic. See my discussion with LauLuna on the
"Penrose and algorithms" thread here:
http://groups.google.com/group/everything-list/browse_thread/thread/c92723e0ef1a480c/429e70be57d2940b?#429e70be57d2940b
Jesse
On Thu, Aug 23, 2012 at 6:38 PM, Stephen P. King
<[email protected] <mailto:[email protected]>> wrote:
Dear Richard,
Your paper <http://vixra.org/pdf/1101.0044v1.pdf> is very
interesting. It reminds me a lot of Stephen Wolfram's cellular
automaton theory. I only have one big problem with it. The 10d
manifold would be a single fixed structure that, while conceivably
capable of running the computations and/or implementing the Peano
arithmetic, has a problem with the role of time in it. You might
have a solution to this problem that I see that I did not deduce
as I read your paper. How do you define time for your model?
-
--
Onward!
Stephen
http://webpages.charter.net/stephenk1/Outlaw/Outlaw.html
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to
[email protected].
For more options, visit this group at
http://groups.google.com/group/everything-list?hl=en.
