Computational Creativity - A General Theory of Creativity

A General Theory of Creativity

A general model of creativity was presented by S. L. Thaler through a series of international patents, beginning in 1997 with U.S. Patent 5,659,666. Based upon theoretical studies of traumatized neural networks, this extensive intellectual property suite taught the application of all forms of noise and damage to a trained neural network so as to induce the formation of novel or confabulatory patterns that could potentially qualify as ideas and/or plans of action. The major advance over previous neural generative approaches, aimed largely at optimization, was that the novel patterns produced represented whole new ideational classes created by the disruptions to the attractor landscape of multilayer networks. As such, they represented more than interpolative patterns characteristic of prior energy minimization techniques. In effect, the network was not only driven by synaptic fluctuations, but it attempted to carry out internal pattern completion, upon such disturbances. Oftentimes the failure of this completion process in reconstructing a stored memory resulted in the formation of original concepts or action plans. This simple generative mechanism, coined "virtual input effect" could operate in the absence of any network inputs as a mixture of memories and ideational patterns nucleated upon synaptic noise internal to the net, allowing it to perform "eyes-shut" discovery and invention. Supplied contextual inputs, the system could creatively interpret any applied input patterns.

Supervisory algorithms, preferably unperturbed neural nets, could serve as objective functions, that could (1) absorb complex scientific or mathematical models, (2) emulate subjective tastes, or (3) develop their own preferences through totally unsupervised learning schemes. In the first case, this fundamental neural architecture, called a "Creativity Machine," functioned as a scientist and in the second, subjective mode, an artist (see, for instance,). In the third case, the Creativity Machine Paradigm became a laboratory to follow the evolution of artifcial consciousness.

In operation, such critics called alert associative centers or "AAC," could control the synaptic perturbations injected into the generative network so as to increase the novelty of emergent patterns, or if "satisfied" lower synaptic perturbation level so as to automatically explore variations upon the more successful schemes. In this sense, the tandem arrangement of networks manifest an attentional consciousness in which initial interest by the AAC resulted in the manipulation of emergent patterns via what would be called "curiosity" by top-down psychologists. Similarly, an environmental pattern applied to the inputs of the perturbed net, could be imaginatively varied via the assessments of the critic net until it arrived a self-consistent interpretation of a perceived object or scenario. If the perturbed net was made recurrent, then the AAC could experiment with the noise-driven unfolding of complex processes as in logistical planning, control sequences, and musical composition. In effect, this patent represented the process wherein different portion of the cortex enter into a dialog with one another to contemplate solutions to various challenges presented to the brain.

In 1997, the inventor, S. L. Thaler, claimed that both empirical and theoretical studies had demonstrated that Creativity Machines output ideational patterns with the same rhythm or prosody as that of human cognition, when the cycle time of computational neurons was slowed from the nanosecond time scale of a digital computer to the 300 millisecond time frames characteristic of neurobiology. The hesitancy accompanying more challenging creative cognition seemed, by comparison with Creativity Machine function, associated with large perturbations applied to the synapses of isolated neurons. On the other hand, similar comparisons between the ideational streams of these neural architectures and human cognition suggested that the smooth, linear flow observed in less challenging tasks (e.g., rote memory recall) appeared to be the product of many smaller perturbative influences uniformly spread over the network as a whole. From these studies emerged an equation linking fractal dimension of cognitive turnover with the number of neural “clock cycles” required to perform cognitive tasks. Furthermore, the chemical diffusion of rogue neurotransmitters, ultimately driven by heat, was implicated as a major driver of all cognition, creative or not.

This canonical neural architecture has been applied to myriad fields of endeavor and has been responsible for the design of a multitude products and services over the last 20 years. It has also been utilized in military applications ranging from materials discovery to the conceptualization of new weaponry to the control of whole constellations of communications satellites. The success of this approach stems from the discovery of a narrow noise regime wherein mild constraint softening occurs leading to novel patterns that are plausible and coherent in terms of the learned knowledge domain, thus vastly shrinking the search space the watching perceptrons must comb through for solution patterns. Using adaptive anomaly detectors as computational critics, the system may manifest the equivalent of "amazement" at its own self-generated concepts thereafter self-absorbing such curiosities.

The Creativity Machine, known in the patent literature as "Device for the Autonomous Generation of Useful Information" or "DAGUI," stemmed from Thaler's earlier research into traumatized artificial neural networks, wherein long-term and transient destruction of connection weights between neurons led to the generation of novel concepts. that could be deemed useful or interesting to attendant neural networks. (see also the Computational Psychology section of Wikipedia entry for Near-Death Experience ). From these controversial studies of fantasy generation via dying neural networks, Thaler saw that idea-generation originated from fluctuations in synaptic connection strengths from their trained-in values. He also noted that with progressively larger weight perturbations, Δw, neural nets transitioned from the generation of rote memories to potentially useful confabulations. This process of novel pattern formation by a synaptically perturbed network (coined an "imagitron") accompanied by pattern recognition by another net (e.g., a perceptron) accounted for a wide range of seminal cognition, which Thaler then generalized to parallel cascades of networks carrying out juxtapositional invention, and both deductive and inductive logic.

More recent patent variations on the Creativity Machine in 1998, enabled it to learn from its successes and failures through reinforcement training within both its generative and monitoring nets. Equipping this system with both actuator and sensor suites it is able to implement its conceptual patterns within the environment and to glean the effect of such notions. Through such bootstrapping sessions this new form of self-learning Creativity Machine called a "DABUI" or "Device for the Autonomous Bootstrapping of Information" may rapidly enrich its inventive capacities. Learning from audience reaction, a whole album of music was synthesized in a matter of hours, with no prior musical mentorship or programming. Working with the US Air Force, such self-adapting Creativity Machines served as the brains of robotic swarms that cleverly gained entrance to underground facilities and collectively explored and mapped them altering their sensor suite to suit their immediate environment. Later in 2006, NASA spacecraft simulators learned to perform autonomous rendezvous and docking using brain-like vision systems configured via a Creativity Machine. In this exercise, the system spontaneously generated the required navigational fields and then optimized its traversal through them via Creativity Machine Paradigm.

Taking creative control of themselves, Creativity Machines may self-organize themselves into extensive brain-like structures, if need be incorporating hundreds of billions of computational neurons. Such self-constructed neural architectures have served as the basis of advanced machine vision systems for both the automotive industries and homeland security. Furthermore, these compound Creativity Machines were adopted by a U.S. intelligence agency in the summer of 2001 in counter-terrorism operations, performing sense-making functions wherein raw textual content was creatively interpreted to assign semantic meaning thereto. Similarly, compound Creativity Machines have routinely performed juxtapositional invention in order to blend knowledge from disparate conceptual spaces

Thaler argues that the entire gamut of human ingenuity may be couched in terms of the Creativity Machine Paradigm, including latent idea formation, convergent and divergent thought, and more reflexive aspects of cognition that were heretofore not regarded as creative, including consciousness itself, the spontaneous invention of significance to the relentless stream of neural activation patterns within cortex. Thaler also explains the entire range of inner mental life in his philosophical explorations in terms of the levels applied to a Creativity Machine emulating thalamocortical function of the brain.

In summary, Thaler proposes a master theory of creativity that boils down to the noise-induced generation of novel patterns within cortex that other such nets may attach significance to. In so-called “H-Creativity” (see below), the societal collective juggles interpretation of ideas via Creativity Machine Paradigm to discern novelty and usefulness, oftentimes incorporating hidden agendas into the group think. Blind rediscovery, at least from the collective’s perception, likewise via Creativity Machine Paradigm, leads to so-called “P-creativity.” Thaler adds so-called “V-“ or “Visceral” creativity wherein simple acts of perception are considered inventive in nature such as the process of “foveation” wherein a small population of noisy cortical networks directs eyeball movement until other nets detect interesting objects of interest in the environment. In effect, the Creativity Machine paradigm is inventing a place to look, dwelling on interesting portions of the environment when it detects novel or interesting objects or activities therein.

Worthwhile noting here, is that the Creativity Machine paradigm has become the central AI development fueling legal debate over the ownership and validity of intellectual property generated by any form of synthetic intelligence.