Importation of methods from statistical physics into machine learning has led to rapid advancement in algorithms for efficient learning of good representations for complex problems. This paper explores the potential for cross-fertilization in the other direction. The Stapp ontology for quantum dynamics (Stapp, 1997, 1999) can be coupled with the prequential theory of probability (Dawid, 1999) to yield a unified ontology for science that is a fully adequate foundation for scientific understanding across the physical, biological, and social sciences. Conscious experience and learning play a central and fundamental role in this ontology, as distinct from their epiphenomenal role in the classical ontology. Adding an evolutionary economic theory such as that of Nau (1999) provides a natural role for conscious decision making and free will in the scientific worldview. The ontology suggested here is fully consistent with our current knowledge of the workings of the physical universe (Stapp, 1997, 1999). Moreover, it fills an acknowledged gap in currently popular ontologies for quantum theory and at the same time fills a complementary gap in current theories of neurobiology, psychology and artificial intelligence.
Classical mechanics is a dynamically complete theory with no role for conscious thought and efficacious deliberate action. As an engineering approximation it was extremely successful at enabling sophisticated technological artifacts to be fashioned out of non-conscious material. However, it fails at precisely the place at which learning and artificial intelligence require an adequate theory. The attempt to found artificial intelligence and computational psychology on a deterministic computional dynamics with no role for consciousness and efficacious action has led to counterproductive thrashing over issues such as the "zombie" question (Sutherland, 1995) and the Chinese room (Searle 1980). At the same time, physicists have been struggling just as counterproductively with a theory of the physical universe in which predictable dynamic evolution is inexplicably punctuated with periodic "collapses" that occur without apparent rhyme or reason. Coupling quantum dynamics with Bayesian probability and decision theory, particularly in its more recent evolutionary formulations (cf., Dawid, 1999), provides an ontologically complete non-dualistic theory that seamlessly fills both gaps.
According to the orthdox Copenhagen interpretation promoted by Bohr (1934), quantum theory replaces a classical theory referring to the material world with a new theory that explicitly refers to the experience of observers. A quantum system's physical state is not directly observable. Its role in the theory is to organize, explain and enable accurate sequential prediction of the experience of observers. No attempt to banish the observer from quantum theory has as yet produced satisfactory results. Quantum systems evolve according to three distinct dynamic processes. In the absence of observations, the state of a quantum system evolves deterministically according to the Shrödinger "wave equation." In order to test the theory, a scientist must observe the system and compare it with prediction. This occurs according to a process not specified by the theory, in which an observer prepares an experimental set-up that poses a question to the system. Finally, the system provides an answer to the question according to the statistical rules specified by the theory. Quantum theory has nothing to say about when observations occur and what questions are posed to the system. This creates an unsatisfactory gap in the theory, because future evolution of the system depends on whether a question is asked and if so, what the question was. von Neumann (1932) and Wigner (1967) suggested filling this explanatory gap by bringing the bodies and brains of the observers into the quantum state, and allowing an interaction between the informational structure represented by the quantum state and the informational structure of conscious experience. Stapp (1997, 1999) argues that such an interaction can fill a fundamental lacuna in the formulation of quantum theory, considered as a theory of reality, and that it allows consciousness to become efficacious without disturbing any of the precepts or rules of quantum theory. The Stapp model for the role of conscious deliberation in quantum theory can be represented as an influence diagram. Although Stapp does not formally specify a value component in his model, he speculates that a brain might compare multiple macroscopically distinct possibilities in parallel and select according to psychological criteria. Value nodes in an influence diagram provide a natural way to model the psychological valuation of each possibility.
Stapp suggests that the brain encodes a "body-world schema" that is the brain's representation of the body and its environment. More faithful representations evolve as information is gained via the measurement process. Prequential learning (Dawid, 1999) provides a natural framework, consistent with the probabilistic language of quantum physics, for dynamic evolution of body-world schemas given environmental feedback. Nau's no-arbitrage theory of rationality (1999) may provide a basis for a prequential economic theory in which behavior consistent with decision theoretic rationality emerges.
The ontology presented here provides a unified account of the physical and mental aspects of the scientific description of the world. As such, it shows promise for a post-classical theory of computing founded on explicitly non-deterministic and irreversible quantum systems. We speculate that Bayesian network fragments (Laskey and Mahoney, 1997) may lead to a useful "interaction (not programming!) language" for quantum computers. Building blocks such as the "noisy logic gates" currently popular in the graphical model community might be exploited to develop an "assembly language" for such systems. Current research in mixed deterministic / probabilistic belief propagation might prove useful for quantum computing, given the engineering difficulties that plague attempts to maintain decoherence in quantum computing systems with more than a few qbits (Chang and Yamamoto, 1997; Gershenfeld and Chuang, 1998).
In closing, it is worth considering the sociological implications of the ontology proposed here, especially if it proves to be fruitful in generating engineering advances. If all quantum systems are in some sense conscious, then building a quantum computer would amount to creating an engineered proto-consciousness. Because all physical systems are quantum systems, this is true even of digital computers, although it would appear that the degree to which they exhibit the property we call consciousness at the human level is extremely limited. However, we may soon succeed in building adaptively intelligent quantum computers. It is worth pausing to give serious thought to how we wish to go about doing this. An important first step is to found both our theory of quantum computing and our economic and social theories of collective decision making on a scientific ontology that has an explicit place for free will and responsible choice. The ontology proposed here does this quite naturally and is fully consistent with known science.
Acknowledgements
The author extends grateful acknowledgement to Henry Stapp for many helpful discussions on the ideas presented here. Thanks for helpful discussions are extended to Menas Kafatos, Paul Lehner, Tod Levitt, Randall Morck, Harold Morowitz, Jim Myers, Fred Alan Wolf, and many others too numerous to list here.
References
Bohr, N. (1934), Atomic Theory and the Description of Nature. Cambridge: Cambridge University Press.
Chang, I.L. and Yamamoto, Y. (1997) The Persistent Quantum Bit. Physics Reviews A.
Dawid, A.P. and Vovk, V.G. (1999), Prequential Probability: Principles and Properties, Bernoulli, 5: 125-162.
Gershenfeld, N. and Chuang, I. (1998) Quantum Computing with Molecules, Scientific American, June 1998.
Laskey, K.B. and Mahoney, S.M. (1997), Network Fragments: Representing Knowledge for Constructing Probabilistic Models.Proceedings of the Thirteenth Conference on Uncertainty in Artificial Intelligence, San Mateo, CA: Morgan Kaufmann.
Nau, Robert F. (1999), Arbitrage, Incomplete Models, And Interactive Rationality, working paper, Fuqua School of Business, Duke University. Searle, John R. (1980), Minds, Brains, and Programs, Behavioral and Brain Sciences 3: 417-457. (Lockwood and SEL, QP360 .B425)
Stapp, H. (1999) Attention, Intention, and Will in Quantum Physics, in J. Conscious Studies 6, 140-164, and in The Volitional Brain, eds. B. Libet, A Freeman, and K. Sutherland, Imprint Academic
Stapp, H. (1997), Science of Consciousness and the Hard Problem. The Journal of Mind and Behavior 18, 171-193.
Sutherland, K. (1995), Conversations with Zombies. Journal of Consciousness Studies 2(4), 312-372.
Wigner , E. (1967), Remarks on the Mind-Body Problem. in Symmetries and Reflections, Indiana University Press, Bloomington, 171-184.
von Neumann, J. (1932), Mathematical Foundations of Quantum
Mechanics (Ch. VI), Princeton, NJ., Princeton University Press.
