Goedel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements Export

@misc{citeulike:845318, abstract = {A Goedel Machine is designed to solve arbitrary computational problems, such as maximizing the expected future reward of a robot in a possibly stochastic and reactive environment. Its initial software includes an axiomatic description of (1) the Goedel machine's hardware, (2) known aspects of the environment, (3) goals and rewards to be achieved, (4) costs of actions and computations, (5) the initial software itself (no circularity involved here). It also includes a possibly sub-optimal initial problem-solving policy and a proof searcher searching the space of computable proof techniques - that is, programs whose outputs are proofs. Unlike previous approaches, the self-referential Goedel machine will rewrite any part of its software (including axioms and proof searcher) as soon as it has found a proof that this will improve its future performance, given typically limited computational resources (its optimality notion is not restricted to the concept of asymptotic optimality). Self-rewrites represent globally optimal self-improvements (no local minima!), since provably none of all the alternative rewrites and proofs (that could be found in the future by continuing the proof search) is worth waiting for. To initialize the proof searcher we may use the recent Optimal Ordered Problem Solver.}, archivePrefix = {arXiv}, author = {Schmidhuber, Juergen}, citeulike-article-id = {845318}, citeulike-linkout-0 = {http://arxiv.org/abs/cs/0309048v2}, citeulike-linkout-1 = {http://arxiv.org/pdf/cs/0309048v2}, day = {27}, eprint = {cs/0309048v2}, keywords = {adaptability, aixi, all-logical-theories, atp, autonomic-computing, bias, biops, chu-space, computability-logic, computational-epistemology, computational-learning, computational-mathematics, computational-science, domain-theory, dynamics, efficiency, formal-development, formal-system, generalized-kolmogorov-complexity, goedel-machine, gtm, iff, inference, inference-dynamics, interactivity, lattice-of-theories, least-time, machine-learning, metaevolution, metalogic-programming, metalogics, metascience, model-selection, oops, optimality, paraconsistency, problem-solving, proof-search, reinforcement-learning, resource-allocation, robotics, searching, self-adaptation, self-managing, self-reference, semweb, speed-prior, survivability, temporal-logic, universal-search, universality, utility}, month = {Dec}, posted-at = {2006-09-15 14:13:48}, priority = {2}, title = {Goedel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements}, url = {http://arxiv.org/abs/cs/0309048v2}, year = {2004} }

TY - ELEC ID - citeulike:845318 L3 - citeulike-article-id:845318 N2 - A Goedel Machine is designed to solve arbitrary computational problems, such as maximizing the expected future reward of a robot in a possibly stochastic and reactive environment. Its initial software includes an axiomatic description of (1) the Goedel machine's hardware, (2) known aspects of the environment, (3) goals and rewards to be achieved, (4) costs of actions and computations, (5) the initial software itself (no circularity involved here). It also includes a possibly sub-optimal initial problem-solving policy and a proof searcher searching the space of computable proof techniques - that is, programs whose outputs are proofs. Unlike previous approaches, the self-referential Goedel machine will rewrite any part of its software (including axioms and proof searcher) as soon as it has found a proof that this will improve its future performance, given typically limited computational resources (its optimality notion is not restricted to the concept of asymptotic optimality). Self-rewrites represent globally optimal self-improvements (no local minima!), since provably none of all the alternative rewrites and proofs (that could be found in the future by continuing the proof search) is worth waiting for. To initialize the proof searcher we may use the recent Optimal Ordered Problem Solver. TI - Goedel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements KW - adaptability KW - aixi KW - all-logical-theories KW - atp KW - autonomic-computing KW - bias KW - biops KW - chu-space KW - computability-logic KW - computational-epistemology KW - computational-learning KW - computational-mathematics KW - computational-science KW - domain-theory KW - dynamics KW - efficiency KW - formal-development KW - formal-system KW - generalized-kolmogorov-complexity KW - goedel-machine KW - gtm KW - iff KW - inference KW - inference-dynamics KW - interactivity KW - lattice-of-theories KW - least-time KW - machine-learning KW - metaevolution KW - metalogic-programming KW - metalogics KW - metascience KW - model-selection KW - oops KW - optimality KW - paraconsistency KW - problem-solving KW - proof-search KW - reinforcement-learning KW - resource-allocation KW - robotics KW - searching KW - self-adaptation KW - self-managing KW - self-reference KW - semweb KW - speed-prior KW - survivability KW - temporal-logic KW - universal-search KW - universality KW - utility AU - Schmidhuber, Juergen PY - 2004/Dec/27/ UR - http://arxiv.org/abs/cs/0309048v2 ER -