خلاقیت به کمک کامپیوتر : شبیه سازی فرآیندهای شناختی بر روی یک سیستم چند عامله

For creativity to be computed, it is paramount to understand the cognitive processes involved, which have been elucidated by either surveying creative people or discovering regions of the human brain that activate during creative endeavors. From this scattering, the author proposes a holistic framework to describe them and their interaction. Hence, creativity can be regarded as a meta process which coordinates autonomous cognitive processes such as planning or divergent thinking. To represent the interplay of cognitive processes around creativity, models are developed in the Agent Unified Modeling Language (AUML). Then, the execution of each process is delegated to autonomous agents and a global coordination protocol is devised. The implementation of the MAS is done on the JADE platform. Two modules of the resultant system are exemplified: opus planning and divergent exploration. The coordination protocol is also presented. The domain in which the software system is tested is the creation of musical pieces.

Human beings posses the ability to visualize and materialize innovative constructs, either artistic or scientific, tangible or intangible. This is broadly labeled as creativity. Artificial Intelligence theorists and practitioners question whether it is possible to compute creativity. One line of thought argues that any computer system must be considered creative when it goes beyond doing the bidding of a programmer, that is to say, when the system transforms the search space embedded in the algorithms or when the idea produced by the system was not influenced by a human counterpart (Jennings, 2010). On the other hand, in computer-assisted creativity there is a direct interaction between the processing system and the human creator. The human being changes parameters of certain algorithms until the resultant opus pleases her/him. Nevertheless, to compute creativity it is paramount to understand its mechanics i.e. the cognitive processes involved. To elucidate the cognitive processes around creativity, two approaches have been followed. One favors surveying creative subjects such as musicians, scientists or artisans. The other is based on analyzing the actual activation of brain areas by means of Positron-Emission Tomography (PET), Electro Encephalograms (EEG) and similar scanning techniques. Even though such findings have been reported separately, they lead to regarding creativity as a meta process i.e. a process that triggers and coordinates autonomous cognitive processes. Planning, divergent thinking, accessing to knowledge stored in long-term memory, and selective attention are among them. To compute creativity, the author proposes a model based on cognitive processes and their relations. The resultant system is implemented on the agent-oriented approach because autonomous agents are fit to be responsible for executing each cognitive process. In this sense, agents are enabled possess the necessary intelligence and algorithms that emulate them. Coordination among agents is achieved by a communication protocol. The creative subject makes decisions that concern primarily with the outcome of each autonomous process. Altogether, the multi-agent system and the creative subject interact to complete the meta-process of creativity. This paper is organized as follows. Section 2 contains theoretical notions surrounding creativity. Section 3 presents the cognitive processes involved in creativity. Next, the agent-oriented models are described in Section 4. Then, the actual performance of the software is exemplified in Section 5. Two modules are presented i. e. opus planning and divergent exploration. A discussion comes in Section 6. Finally, conclusions and future work are drawn. Creation of musical pieces is the chosen domain to exemplify the proposals.

Creativity has been characterized as a meta process that coordinates autonomous cognitive processes in order to produce something new and valuable. It has been described the complexity of the interplay involved, from planning, exploring, combining, evaluating and resolving. Conscious attention to what is being created is important to evaluate and decide whether to continue with the original setting. Such interplay has been modeled by using the agent-oriented paradigm. Agents represent a suitable approach because they reproduce autonomous processes, which are coordinated by message interchanging. Our agent-based system possesses attributes that make it suitable to be used for supporting creative tasks. Firstly, PlannerAgent helps manipulating a manageable conceptual space, which originally only consists of a very large system of possibilities. Creative subject concentrates on a limited set of parameters and, more importantly, on the perceived outcome. Thus, the restriction of working memory capacity is overcome. By reproducing either on line or off line the resultant opus, the sensory mechanisms of the brain are stimulated, presumably leading to spontaneous insights not envisioned in the original setting. Explorer Agent emulates divergent thinking by computing nonlinear systems, which allow the formation of complex, non-repetitive patterns. The coordination between Creative Subject, Planner Agent and Explorer Agent was exemplified. Even though the notion of conceptual space is used reiteratively, further analysis is needed to clarify when a conceptual space is the system of possibilities given by valid concepts or whether after applying an exploration process (traversal in Forth et al. (2010)) it is still a conceptual space or an exploratory space from which another set of combination rules or strategies apply. In order to advance in autonomous artificial creativity, future work includes the definition of combination algorithms in order to create secondary interim opus. It is also pivotal to develop evaluation templates suitable to experts and non-experts to capture subjective valuations and technical evaluations. These will lead to learning from the creative experiences that occur when the Creative Subject interacts with the multi-agent system.