Abstract
We present the first class of mathematically rigorous, general, fully self-referential, self-improving, optimal reinforcement learning systems. Such a system rewrites any part of its own code as soon as it has found a proof that the rewrite is useful, where the problem-dependent utility function and the hardware and the entire initial code are described by axioms encoded in an initial proof searcher which is also part of the initial code. The searcher systematically and efficiently tests computable proof techniques (programs whose outputs are proofs) until it finds a provably useful, computable self-rewrite. We show that such a self-rewrite is globally optimal—no local maxima!—since the code first had to prove that it is not useful to continue the proof search for alternative self-rewrites. Unlike previous non-self-referential methods based on hardwired proof searchers, ours not only boasts an optimal order of complexity but can optimally reduce any slowdowns hidden by the O()-notation, provided the utility of such speed-ups is provable at all.
An erratum to this chapter can be found at http://dx.doi.org/10.1007/11550907_163 .
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
Blum, M.: On effective procedures for speeding up algorithms. Journal of the ACM 18(2), 290–305 (1971)
Fitting, M.C.: First-Order Logic and Automated Theorem Proving. In: Graduate Texts in Computer Science, 2nd edn. Springer, Berlin (1996)
Gödel, K.: Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I. Monatshefte für Mathematik und Physik 38, 173–198 (1931)
Hutter, M.: The fastest and shortest algorithm for all well-defined problems. International Journal of Foundations of Computer Science 13(3), 431–443 (2002); On J. Schmidhuber’s SNF grant 20-61847
Hutter, M.: Universal Artificial Intelligence: Sequential Decisions based on Algorithmic Probability, pp. 20–26. Springer, Berlin (2004); On J. Schmidhuber’s SNF grant 20-61847
Kaelbling, L.P., Littman, M.L., Moore, A.W.: Reinforcement learning: a survey. Journal of AI research 4, 237–285 (1996)
Levin, L.A.: Randomness conservation inequalities: Information and independence in mathematical theories. Information and Control 61, 15–37 (1984)
Schmidhuber, J.: The Speed Prior: a new simplicity measure yielding near-optimal computable predictions. In: Kivinen, J., Sloan, R.H. (eds.) COLT 2002. LNCS (LNAI), vol. 2375, pp. 216–228. Springer, Heidelberg (2002)
Schmidhuber, J.: Optimal ordered problem solver. Machine Learning 54, 211–254 (2004)
Schmidhuber, J.: Gödel machines: fully self-referential optimal universal problem solvers. In: Goertzel, B., Pennachin, C. (eds.) Artificial General Intelligence. Springer, Heidelberg (2005) (in press)
Schmidhuber, J.: Gödel machines: Towards a technical justification of consciousness. In: Kudenko, D., Kazakov, D., Alonso, E. (eds.) AAMAS 2004. LNCS (LNAI), vol. 3394, pp. 1–23. Springer, Heidelberg (2005)
Solomonoff, R.J.: Complexity-based induction systems. IEEE Transactions on Information Theory IT-24(5), 422–432 (1978)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2005 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Schmidhuber, J. (2005). Completely Self-referential Optimal Reinforcement Learners. In: Duch, W., Kacprzyk, J., Oja, E., Zadrożny, S. (eds) Artificial Neural Networks: Formal Models and Their Applications – ICANN 2005. ICANN 2005. Lecture Notes in Computer Science, vol 3697. Springer, Berlin, Heidelberg. https://doi.org/10.1007/11550907_36
Download citation
DOI: https://doi.org/10.1007/11550907_36
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-28755-1
Online ISBN: 978-3-540-28756-8
eBook Packages: Computer ScienceComputer Science (R0)