Professional Documents
Culture Documents
Bruno MARCHAL
I.R.I.D.I.A. Université Libre de Bruxelles
50 av. F. Roosevelt. CP194/6. B-1050 Brussels, Belgium
φ φ .. .(x n ) (u) = φ t(e,x1 ,...,xn ) *** In 25, I argue that the ultimate (limit) story
.φ (x ) ..
e 1 of a solipsist, in Platonist Mechanist
philosophy, is given by S4Grz, i.e. S4 system +
If the branching is produced by the synthesis of a the Grzegorczyk formula :
set of erroneous or approximate hypothesis by
inductive inference, then although the nets is „ ( „ (p-> „ p)->p)->p
presented constructively (outside AND) any
"correct branch" cannot be algorithmically thanks to an elucidation of the relationship
determinable (inside OR) (see 25, 26). between G* and S4Grz provided by Boolos (see
2, 17).
formalized version of „ which exists for inconsistency by admitting that itself is a
self-appropriate machine by hypothesis machine.
H. That is not a proof of consistency but In what logic could the last two
a (not ending) experience of consistency. paragraphs be formalized ? The solution
In the limit the machine is able to emulate is : accept " „ p–>p", and drop the
its own (self)referentially correct) verbal necessitation rule (or drop the idea that
communication. So we will have : axioms are theorems). Think about an
(◊T–>- „ ◊T); so by (first order arithmetical theorem prover having a
abductive rule to anticipate his own
abduction) and (- „ ◊T) the machine (maybe non stopping) behaviour.
will have ◊T confirming the feeling Because (like the UIIM) it will prove
described by -" „ "⊥ . Let us read „ P ◊ T –>- „ ◊ T and because „ ⊥ will never
as "I give a convincing communication stop, it will in the limit synthesize the
of", ◊P as "I am able to imagine (or
consider, or dream about) P", and by metaprogramming rule -„ ⊥
"The system feels (or knows) that". (experienced) and ◊T (abduction and
Then we see that in the limit the system experience), but it will never be able to
feels or knows that it is able to imagine infer „ ( „ p–>p). There will never be
(or consider) truth but feels or knows that
it is not able to give any convincing identification between „ and , and the
communication of that fact. If we admit limit result will be G + „ p–>p without
defining consciousness as an internal the necessitation rule, i.e. G*.(both G
feeling of consistency (or an equivalent and G* are finite decidable theory) The
manner : an internal feeling of having the arithmetical theorem prover + memory +
ability to imagine truth) then the approach abductive rules concludes that it is
here explains why solipsism is irrefutable consistent, but because there is a use of
(although Platonistically false). The the internal experience (abduction, or
approach shows also that such machines in case of UIIM) it cannot infer that such
will eventually have a richer inner set of a conclusion is a communicable one.
beliefs than what they will be able to What it can do is to develop a theory of
communicate convincingly between mind by enriching S4(-) (S4 with a
themselves. This suggests some weakening of the necessitation rule) with
epistemic interpretation on the non-union new axioms. Reinhardt (see 34) proves in
phenomenon. Such machines are also S4(-) + a strong mechanist axiom, that
able to refute any attempt to identify them there are absolute truths which are not
(selves) with a (universal) hypothesis. provable. Philosophy of mind is at the
G* level, and Philosophy of mind cannot
be self-referentially correct. Note the
9. PHILOSOPHY OF MIND duality between incorrigibility and
incommunicability : the first one implies
It is not possible, for a sufficiently rich that if you are conscious nobody can
system, to be consistent, self-referentially "prove" you are wrong, the second one
correct and to obey the reflection principle implies that if you are conscious you
„ p–>p. Now when I propose ◊ for cannot "prove" it to somebody (see 25).
imagination, I am making something
dangerous because ◊T is the same as
„ ⊥–>⊥. Of course ◊⊥ is not the same 10. S OME LIGHT ON THE
PARADOXES
as „ ◊ T but the system obeys p–>p,
so it seems that, at least internally, we Naïve Mechanism is correct with respect
will have ◊T . So how could a machine to the less naïve approach I gave relative
talk about internal logic (feeling) and to the subject theory I develop. The level
external logic (convincing at which a translator works is the level
communications) without falling into which emulates the iterations e–>e'–> …
which can all be seen as a "translation"
(or duplication + annihilation). There is does vehicule an experience which has
no way to give a convincing been vehiculated by causal relationship
communication about the very existence (so there are no new experiences). The
of such a level, although it is consistent to paradoxical situation is reduced to a
admit that there is such a one. postponed duplication like paradox.
Concerning our machines we know from
the outside that a level exists (the back-up
level!). To choose a level for ourselves 11. T HE UNIVERSAL DOVETAILER
depends on empirical consideration. The PARADOX
"probability = 1" argument is at the truth
level, that is the G* level (or an S4(-) When a set is such that there is an ICF
level) so that in the postponed duplication which generates it, the set is said to be
paradox the copy is right when he said "I recursively enumerable (RE). The
was wrong" and the "original" was false cartesian product of finite number of RE
in his induction . Above all the copy is sets is RE. The traditional name given to
right when he realizes he will not be able the algorithm which generates such
to communicate his error to the original. products is the dovetailer. Having a
The experience was just constructed in formal definition of a universal ICF, you
such a way. During a lapse of time "to be can write a program which generates by
the other", although a true possibility, is dovetailing all the finite approximations
actually unbelievable. An UIIM is able to of all the executions of all the ICF
infer that in the iterated duplication including the presence of any oracles (see
without annihilation, the one who always 45). I call such a program an universal
quits the transmitter will in the limit lost dovetailer (UD). Oracles play the role of
his mechanist faith for his experience will possible environments. The platonist
be stochastically impossible (see 24). hypothesis (used elsewhere) entails that
When copies are destroyed, the argument there is no need for an actual emulation of
showing that the probability of being the UD (for actuality is an inner
executed is 1/2, although the probability experience).
of staying alive in the long run is 1 is also Remember that with the theory presented
correct. It can be shown, (for exemple in above the existence of a subjective
S4(-) + mechanist axioms, where the experience corresponds to the existence
UIIM is described by a self-referential net of chains of appropriate self-reference
as in footnote *** above) that the relatively to an environment : e->e'-> . . .
subjective (solipsistic) experience of the Let us call such a sequence a sequence of
copy can be considered equivalent to a state of mind. For each state of mind the
personal forgotten dream : UD is able to process a non denombrable
(in the limit) different environment
including the most unexpected dreams.
Most of them will be inconsistent and will
not play any role in the limit (like
forgotten dreams). But there is no reason,
The filmed bidimensional computer a priori, that a great number of them can
paradox shows, from a UIIM point of be locally consistent although
view, that any subjective experience is contradicting the majority of our
unique and internal. The subjective inductions. The probability aspect of the
experience is defined (and redefined) at translator-like paradoxes was easy to
each state of the process and resumes the work with because the set of possibilities
entire chain of self-appropriated was finite. Unfortunately the Platonist
references, building something like Mechanist solution proposed here entails
personal history (an internal construct of that we must take all consistent
time) relatively to a universal extensions on the actual environment into
environment. The paradox arises from a account. So there is a need to find a
confusion between internal time and measure on the set of sequences of states
external time. The succession of of mind capable of justifying the
instantaneous state captured by the film
normality of our daily inductive beliefs. that the reduction principle can be derived
Curiously enough such a problem has from the continuous evolution S of the
been partly solved in the context of a whole state of the system in H, provided
realist mechanist attempt to interpret the measurement is done and interpreted
quantum mechanical facts by Hugh by machines. Does it help to interpret Ψ ?
Everett and some others (see 12). In a sense any universal machine can do
that. And any UD does it. But from the
inside point of view there is a big price at
12. T URING M ACHINES IN A the ontological (Platonist) level. It has
Q UANTUM W ORLD been said that such work is a beautiful
theory nobody can believe (see 15) : if the
In quantum mechanics the state of a whole environment is described by Ψ, the
system is described by a mathematical splitting of possibilities entails the
object Ψ belonging to a mathematical splitting of the environment including the
space H. The evolution of Ψ through time observers, and measurement just tells the
is given by a differential equation S. That observer in which environment he is.
evolution is continuous and deterministic. Quantum indeterminism is just a
How to interprete Ψ ? That is a hard and particular case of mechanical
highly debated question. Nevertheless indeterminism. Propositions about it
almost everyone agrees on the way to use occur at the G* level and cannot be
Ψ . It happens that when we make a convincingly communicable. For
measurement of a quantity described itself instance, a common refutation of
Everett's work is that it does not explain
by a set of some states Ψ i , Ψ reduces why we are in this environment and not
abruptly in a state Ψ i with a probability another. But that remark is, from a
computable from Ψi and Ψ. That is called mechanist point of view, equivalent to the
the reduction principle. Ψ seems to remark made by the person, who after
describe a set of interacting possibilities splitting himself between Washington and
evolving continuously until we make Moscow, pretends in Washington (resp
some measurement, in which case, one Moscow) that Mechanism is incomplete
possibility occurs. When and how does because it does not explain why he finds
the reduction occur ? Some have put it at himself in Washington (resp Moscow).
the microscopic level (see 10), others The double-edge nature of the Gödelian
have put it somewhere between the argument against Mechanism (see 47)
microscopic and the macroscopic level extends itself for the Quantum arguments.
(see 21 for details), still others have put it For those who knows the Schroedinger's
between mind and the whole physical cat paradox, Mechanism + Everett enable
system including the brain. This last us to infer the "subjective experience" of
solution has given rise to a lot of rival the cat. Roughly speaking, it is the
(almost all dualist) approaches in the following : "Well, nothing very special,
philosophy of mind (see 27). I know except the presence of more and more
only one meeting between Mechanists physicists with more and more astonished
and Quantum Dualists (see 23). There is eyes" (resee part 9). The goal of Everett
still the possibility that no reduction was to provide an interpretation of
occurs at all. The solution of the filmed Quantum mechanics coherent with
two-dimensional computer paradox given cosmology (see 4, 43). Some of these
above is a generalization of that idea. approaches give an equation for the
Everett shows by using explicitly the history of the universe in which there is
hypothesis that the observer is a machine no more explicit reference to time. Here
and that the whole system observer + also time is internal and relative.
measuring apparatus + object obeys the (Information-theoretic extension of
differential equation S, that the result of Gödel's theorem (see 8) gives hope of
measurement will still reflect the finding analogous internal semantics for
reduction in the memory of the machine. thermodynamical processes or chaotic
Put in another way, Mechanism entails dynamics (see 30)). Everett provides not
only a proof that machines do not record 12 DE WITT B.S. & GRAHAM N., The
the split but also, that if machines make Many-Worlds Interpretation of
successive measurements they will, in the Quantum Mechanics, Princeton
limit, verify the usual quantum statistics. Series in Physics, Princeton Univ. Press,
That work was corrected and refined by 1973.
Graham (see 12) and independently by 13 ENDERTON H.B., A Mathematical
Intoduction to Logic, Academic
Hartle (see 18). It is the equivalent of Press, 1972.
Graham or Hartle's work which is still 14 EVERETT H., The theory of the
needed, concerning the recursion Universal Wave Function, In De Witt
theoretic theory of identity presented B.S. & Graham N.pp. 3-140, 1973.(see
here, to solve the UD paradox. 12).
15 GARDNER M., Time Travel
& Other Mathematical
Bibliography Bewilderments, W.H. Freeman and
Company, N.Y., 1987.
1 BOOLOS G., The Unprovability o f 16 GOLD E.M., Language Identification in
Consistency, an Essay in Modal the Limit, Information & Control 10, pp.
Logic, Cambridge University Press, 447-474, 1967.
1979. 17 GRZEGORCZYK A., Some relational
2 BOOLOS G., Provability, Truth, and systems and the associated topological
Modal Logic, Journal of Philosophical spaces, Fundamenta Mathematicae, LX
Logic, 9, pp. 1-7, 1980. pp. 223-231, 1967.
3 BLUM L. & BLUM M., Toward a 18 HARTLE J.B., Quantum Mechanics of
Mathematical Theory of Inductive Individual Systems, American Journal of
Inference, Information and Control 28, Physics, vol. 36, n° 8, pp. 704-712,
pp. 125-155, 1975. 1968.
4 BROUT R. et ENGLERT F., 19 HOFSTADTER D., DENNETT D.C.,
Cosmologie quantique. in Aux confins (eds) The Mind's I, Basic Books, inc.
de l'univers, Schneider, pp.269-289, Pub., N.Y., 1981.
1975. 20 HUME D., Treatise of Human
5 CASE J., A Note on Degrees of Self- Nature, 1739.
Describing Turing Machines, Journal of 21 JAMMER M., The Philosophy o f
the A.C.M., vol. 18, n° 3, 1971, pp 329- Quantum Mechanics, J. Wiley &
338. Sons, N.Y., 1974.
6 CASE J., Periodicity in Generations of 22 KLEENE S.C., Introduction to
Automata, Mathematical Systems Metamathematics, P. Van Nortrand
Theory. Vol. 8, n° 1. Springer Verlag, Comp. Inc., 1952.
NY, pp. 15-32, 1974. 23 LETOVSKY S., Ecclesiastes : A Report
7 CASE J. & SMITH C., Comparison of from the Battlefields of the Mind-Body
Identification Criteria for Machine Problem, A.I. Magazine, vol. 8, n° 3,
Inductive Inference, Theoretical Computer pp. 63-69, 1987.
Science 25, pp. 193-220, 1983. 24 MARCHAL B., Informatique théorique et
8 CHAITIN G.H., Algorithmic philosophie de l'esprit, Acte du 3ème
Information Theory. Cambridge colloque international Cognition et
Tracts in Theoretical Computer Science 1, Connaissance, pp. 193-227, Toulouse,
Cambridge University Press, 1987. 1988.
9 DALEY R.P., Inductive Inference 25 MARCHAL B., Amoeba, Planaria and
Hierarchies : Probabilistic vs Pluralistic ...Dreaming Machines, submitted to
Strategies, Mathematical Methods of publication, 199?.
Specification and Synthesis of Software 26 MARCHAL B., Des Fondements
Systems, Springer Verlag, Lecture Notes Théoriques pour l'Intelligence Artificielle
in Computer Science 215, 73-82, 1986. et la Philosophie de l'Esprit, Revue
10 DE BROGLIE L., La théorie de l a Internationale de Philosophie, 1, n° 172,
mesure en mécanique ondulatoire. pp 104-117, 1990.
Paris, Gauthier Villars, 1957. 27 MARGENEAU H., The Miracle o f
11 DEWDNEY A.K., The Armchair Existence, Shambhala, New Science
Universe, W.H. Freeman and Company, Library, Boston & London, 1984.
N.Y., 1988. 28 MINSKI M., The Society of Mind,
Simon and Schuster, N.Y., 1985.
29 MYHILL J., Abstract Theory of Self- 47 WEBB J.C., Mechanism, M e n t a l i s m
Reproduction, in Views on General & Metamathematics, an Essay o n
Systems Theory, M.D. Mesarovic, ed. Finitism, D. Reidel Pub. Company,
Wiley NY, 1964, pp 106-118. 1980.
30 Mc CAULEY J.L., Chaotic Dynamical
Systems As Machines Computational
Systems - Natural and Artificial,
Springer, Berlin, 1987.
31 OSHERSON D.N., STOB M.,
WEINSTEIN S., Systems that learn,
The M.I.T. Press, Cambridge, 1986.
32 PERRY J., (ed), Personal Identity,
University of California Press, Berkeley,
1975.
33 PUTMAN H., Reason, Truth and
History, Cambridge University Press,
1981.
34 REINHARDT W.N., Epistemic Theories
and the Interpretation of Gödel's
Incompleteness Theorems, Journal of
Philosophical Logic 15, pp. 427-474,
1986.
35 ROGERS H., Theory of Recursive
Functions and Effective
Computability, McGraw-Hill, 1967.
36 ROSLER O.E., Endophysics, In Casti &
Karlqvist (ed), Real Brains, Artificial
Minds, North Holland, 1987.
37 RUCKER R., Infinity & the Mind,
The Hervester press Ltd, 1982.
38 SHAPIRO S., Intensional
Mathematics, North Holland, 1987.
39 SMITH C.H., The Power of Pluralism for
Automatic Program Synthesis, Journal of
the Association for Computing
Machinery, vol. 29, n° 4, pp. 1144-1165,
1982.
40 SMORYNSKI C., Self-Reference and
Modal Logic, Springer Verlag, 1985.
41 SOLOVAY R., Provability
Interpretations of Modal Logic, Israel
Journal of Mathematics 25, 1976, pp.
287-304.
42 SMULLYAN R.M., Modality and Self-
Reference, In Shapiro, 1985, pp. 191-
211, (see 38).
43 TIPLER F.J., Interpreting the Wave
Function of the Universe, Physics Report
(review of Physics Letters),137, n° 4, pp.
231-275, 1986.
44 TROUILLARD J., L'un et l'âme
selon Proclos, "Les belles lettres",
Paris, 1972.
45 TURING A.M., Systems of Logic Based
on Ordinals, Proceedings of the London
Mathematical Society, ser. 2, vol. 45,
1939.
46 WANG H., From Mathematics t o
Philosophy, Routledge & Kegan Paul,
London, 1974.