Computational Creativity Theory

点击翻译按钮获取中文摘要

Computational Creativity is the study of how to build software which takes on some of the creative responsibility in arts and science projects. We are at a stage where software can generate pictures, melodies, jokes and poems, can invent new words and discover new and interesting mathematical theorems, and regularly helps scientists to make important discoveries. This kind software can be used autonomously, or in collaboration with creative people. It is also used in cognitive modelling projects, to shed light on aspects of human and animal creativity. In the last decade, Computational Creativity has come of age, as evidenced by special issues of publications such as the Minds and Machines journal and the AI magazine, and the first International Joint Conference on Computational Creativity, which replaced 10 years of successful workshops at major AI conferences. The proposed Leadership Fellow, Simon Colton, is a recognised expert in Computational Creativity, and has been working in the field since 1996. He is unique in having been involved in successful applications of creative software to four different domains, namely mathematical invention, video game design, graphic design and the visual arts. His mathematical theory formation software, HR, has produced theorems and concepts published in the mathematical literature; his visual art software, The Painting Fool, has produced pictures that have been exhibited and attracted much public attention; and research being done in the Computational Creativity group that he leads at Imperial College is helping video games companies to design the next generation of adaptive, personalised games.A number of authors, such as Boden, Wiggins and Ritchie, have introduced formalisms which help us to be more precise about the creativity of software. However, there is no agreed upon theory which can describe the behaviour of software with sufficient acuity, coverage and formality that enables accurate comparison of implementations. In short, we have no generic way of saying that software B is more creative than software A. This has held back our field, because with no concrete and formal measures of the creativity of the software we build, it has been hard to put forward falsifiable scientific hypotheses that one approach is more creative than another, hence it has been difficult to progress, and to show progress. With this Fellowship, we propose to change this situation, by developing Computational Creativity Theory (CCT). This will comprise a series of models, each of which contains some conceptual definitions and some calculations involving those definitions which can be used to compare and contrast the creativity of software. The foundational models will make more precise the notion of a creative act and the impact they can have, and the more acute models will cover aspects of creative behaviour including intentionality, interpretation, imagination, appreciation and affect. To model computer creativity sufficiently well, we generalise past the merely generative and past usual AI notions of value, into new areas where software is expected to invent its own aesthetic and utilitarian measures, and frame its creations by describing its motivations, intentions, methods and innovations and by putting its work into historical and cultural contexts.The proposed programme of research has the development of CCT at its heart. This is informed by a series of practical projects involving applications to creative language, music, visual arts, mathematics and games, and covering modes of creativity including realtime generation, assistive technologies and creative collaborations. By building and disseminating CCT, we will help to bring Computational Creativity research into a new era, where formal notions of creativity underpin software systems which really enrich our cultural lives.