The Principles of Creative Robotics

摘要

A creative robot is a robotic implementation of the Comprehensive Turing test for computer simulation of creative activity. The article explores the conceptual and technical principles of creative robotics. Conceptual principles include Lovelace's argument (1843), Turing's counterargument (1950), Lovelace's test (2000), Lovelace's test 2.0 (2014) and Lovelace's test 3.0 (2015). According to A. Lovelace's argument a computer can create a highly artistic work, but it cannot create on its own: the machine only passively executes the instructions of the developerprogrammer. A. Turing, on the contrary, believes that the machine can create and, therefore, think. Turing's counter-argument falls into two parts: the symbolic objection and the connectionist objection. The first is unconvincing: the developer is trying to deceive the observer (the Turing Interrogator) into the indistinguishability of creativity and its software imitation. The originality of a product is determined by an unexpected program flow, such as a machine failure. The connectionist (neural network) answer is more plausible: in general terms, the behavior of a robot is predictable, but particular machine learning products can seem original. Lovelace's test denies Turing's symbolic response by postulating axioms that deceive the observer and prevent program failure. The uniqueness of the artifact is limited by the fact that the products are known to the developer: the set of products can be potentially infinite, nevertheless, it is declaratively or procedurally specified, and therefore, it is actually finite and predictable. The Lovelace 2.0 test rejects Turing's connectionist answer as it applies to deep learning neural networks. It is difficult for the Interrogator to ascribe the creativity of an artifact outside the scope of his creative competence. Axiomatics is being tightened by the specialization of subject areas and the introduction of the role of evaluator of product creativity. In the article it is proposed the Lovelace 3.0 test. It complements the previous version with the axiom of the evaluator detecting meaningful artifact production. The machine is incapable of producing meaning, so it cannot truly create. The new test cannot be passed, but the relative successes of creative robotics are possible. This requires technical principles for integrating symbolic and connectionist approaches. The Korsakov-Turing machine is proposed as a formal definition of the algorithm for the functioning of a creative robot: a stack of punched cards of the Korsakov connectionist machine (1832) characterizes the connections between the sub-symbolic components of the tape and the table of the Turing symbolic machine (1936).