[FOM] The Lucas-Penrose Thesis (long)
A.P. Hazen
a.hazen at philosophy.unimelb.edu.au
Wed Oct 4 04:08:49 EDT 2006
The current string on "the" Lucas-Penrose argument has generated more
responses than I expected! Several people have posted the
"orthodox" solution (going back to Gödel), that if there is a machine
that simulates our, human, mathematical ability, we will never know
of this machine THAT it simulates our ability. (And so we will never
be in a position to spring the anti-mechanist trap by recognizing the
consistency of its output.) A defender of the Lucas-Penrose argument
has objected that this appeal to inevitable ignorance is
unscientific, that it puts the defender of mechanism in the
position of the scientist in the Harris cartoon whose black-board
full of equations has "Here a miracle occurs" in the middle! So
perhaps it would help to give some more detail on this point. I
believe that this "inevitable ignorance of our own identity as
machines" (A) is plausible in itself, (B) is well-nigh demonstrable
from an a priori plausible bunch of mechanist assumptions, and (C)
when further analyzed helps show the relevance of Jeroslow's notion
of an "experimental logic" to the Lucas-Penrose argument. I
apologize in advance for the length of this post: (A) contains little
logic, and the perhaps more interesting (B) and (C) are largely
independent of it.
(A) INTRINSIC PLAUSIBILITY. Why should we EXPECT that we
should be able to learn "what machine we are"? Perhaps the most
attractive ground is optimism about neuroscience: can't we anticipate
that the full "wiring diagram" of the human brain will one day be
mapped out? (Gödel himself didn't share this optimism: he told Wang
he thought it likely that empirical research would show that the
brain wasn't complex enough to be a basis for the mind! But his view
seems to have become a minority one.)
But this seems inadequate.
Never mind the uniqueness presuppositions of the definite articles in
"the full wiring diagram of the human brain." (Though if we allow
the brain to evolve we perhaps approach the idea of Gödel's late
objection to Turing's Thesis, "Collected Works" vol. II, p. 306....)
It's still inadequate. For some purposes the conception of the
mathematical subject pursuing its thoughts solipsistically is not too
bad a distortion, but real human mathematics is a cooperative
enterprise, so at the very least we'd need, not just the wiring
diagram of a brain, but a theory of the interactions of
(arbitrarily?) large numbers of brains. And, leaving aside
contentious questions of the status of "computer proofs" (e.g. that
of the Four Colour Theorem), human mathematicians have long used
pencil and paper as aids in the calculational side of their work: so
we'd need a theory of groups of brains with "external memory"
devices. And real mathematics has often taken inspiration from
problems in physics, so maybe identifying the "machine" which
simulates human mathematics would require a knowledge of all the
physical phenomena that might INSPIRE groups of brains to come up
with new approaches. ... In short, it seems to me that it may be
far from easy to attain even EMPIRICAL confidence that any particular
machine simulates ALL human mathematics! But the Lucas-Penrose
argument doesn't apply to, e.g., the mathematical thought of X as of
date D: there is no reason to think that such a limited mathematical
theory is consistent (remember Martin Davis's Honor Roll of
logicians who have proposed inconsistent foundational theories!).
The Lucas-Penrose argument has to be about the idealized limit, human
mathematics as it would develope if pursued forever.
(B) NECESSITY. Suppose-- as the Lucas-Penrose argument
supposes the mechanist must suppose-- that there is a formal,
axiomatic, theory, HM, whose theorems express all and only the
mathematical **truths** that have been or ever will be discovered by
humans, or ever would be if human mathematics continued indefinitely
into the future.
The restriction to **truths** may
seem extrinsic to the problem, but we don't WANT a theory that
captures the mistakes mathematicians make: such a theory wouldn't
even be consistent! Perhaps the requirement of truth can be replaced
by a weaker one, but the Lucas-Penrose argument, turning as it does
on the claim that humans can know THEIR mathematics to be consistent,
needs SOMETHING like it.
What would HM
be like? Perhaps it would amount to an extension of ZFC by some
(recursive) set of strong axioms of infinity: axioms even Harvey
Friedman hasn't considered yet! HM would BE consistent, but would we
KNOW that it was consistent? By Gödel's Theorem, we wouldn't have a
mathematical proof that HM was consistent: all our mathematical
proofs are, by definition, proofs of theorems of HM. Our confidence
that HM was consistent would be at best empirical, and based on the
consensus of expert opinion. (And, since it would include the most
extreme and speculative true axioms ever proposed, I doubt the
consensus of the expert would be clear or unanimous!)
Would we know of it that it was the
systematization of human mathematics? Again, our assurance would be
at best empirical and conjectural. The key move in the Lucas-Penrose
argument is that we can "recognize" the consistency of human
mathematics, but the ground for this recognition is the principle
that we are smart enough that if we ever encountered an inconsistency
in our mathematics we would re-think and abandon some axiom(s), so
our **real** mathematics would be a consistent subsystem of the
inconsistent one we had previously TAKEN to be our mathematics.
So any doubt about the consistency of HM -- and I have just argued
that there would be SOME doubt of this -- would yield a doubt that
HM was **really** equivalent to HUMAN MATHEMATICS.
(C) FURTHER ANALYSIS. What could neuroscience discover?
(Ignoring the complications under (A).) The assumption of the
Lucas-Penrose argument is that IF "mechanism" is true, then science
can discover what Turing Machine simulates Human Mathematics.
Suppose this is so, that there is a Turing machine, THM, that
adequately models Human Mathematics. How does it do this? Maybe by
printing out, in correct chronological order, everything that will
ever be published in a Human mathematics journal. Suppose
neuroscience discovers that this is so (as -- to paraphrase Post--
an empirical law about human mathematical thinking).
(((Note that the agreement is supposed to be
lawlike: the abstract machine, HTM, doesn't stop at the point when
bird flu or global warming or a new dark age or someone selling
nuclear technology to a TRULY mad dictator stops Human mathematical
development in its tracks. It prints -- or would print if it were
ever physically instantiated -- all that WOULD be published by human
mathematicians if they went on forever.)))
But
note that mathematics journals don't JUST contain theorems: they
contain RETRACTIONS as well!
Leaving my final question: what would be the relation of the
Turing machine, THM, to the axiomatic theory HM described above in
(B)? Would it be (or would it be convertible into) a machine that
enumerates the theorems of HM? Conceivably, YES. Maybe we are lucky
and on the right track and a time will come (or would come if
mathematics endured long enough) after which no further retractions
will ever be necessary! But also conceivably, NO. Maybe there
will be ever new additions to Martin Davis's honor roll: logicians
who advance mathematics by proposing new and stronger axioms, some of
which will later -- perhaps MUCH later -- be discovered to be
inconsistent. THM, it seems to me, would amount to an EXPERIMENTAL
LOGIC, in the sense of Jeroslow's 1976 "Journal of Philosophical
Logic" paper: thre is no particular reason, it seems to me, to assume
that it would be (or would ever be known to be) equivalent to an
axiomatic formal system. And this, it seems to me, is the most
fundamental problem with the Lucas-Penrose argument. It assumes that
"Mechanism" is committed to a certain relationship between the human
machine and the human mathematical theory: that the machine
enumerates the theorems of the theory. But it isn't. THM might
exist and not enumerate the theorems of HM. THM might exist and have
a Delta-0-2 but not Sigma-0-1 set of "permanent" theorems, in which
case the theory HM would not exist.
---
Allen Hazen
Philosophy Department
University of Melbourne
More information about the FOM
mailing list