Computational Approaches to Creativity Through Goal-Directed Cross-Domain Analogy

随着人类对计算工具的依赖增加，不仅需要支持常规任务，还需要支持创造性活动。这项研究的目的是增加我们对创造性认知的计算性特征的理解，并将这种理解转化为支持人类创造过程的人工制品。我们将关注跨域类比的使用和提取，这是一种与创造性认知有关的能力。拟议的研究将产生一类新的方法，通过跨域类比展示计算创造力，并涉及两个想法的新颖组合。一种是类比映射和提取的过程是以目标为导向的，这对它们的操作有很强的限制。另一种是，问题陈述和解决方案由抽象模式索引，这些抽象模式支持检索相关结构，即使它们具有最小的表面相似性。总而言之，这些假设应该支持对跨域类比的健壮检索和映射，这反过来应该导致帮助人类创造力的新软件工具。虽然这个项目的主要关注点是科学创造力，但同样的机制和技术将在各种目标导向的环境中被证明是有用的，这些环境涉及计算机中介的创造性工作，包括设计的许多方面。除了发表科学论文外，用于类比绘图和检索的软件将提供给更广泛的科学界。从长远来看，根据抽象模式为结构编制索引的技术在帮助设计过程和改进对万维网内容的搜索方面具有相当大的潜力。

As human reliance on computational tools increases, there is a critical need to support not only routine tasks but also creative activities. The aims of this research are to increase our understanding of the computational character of creative cognition and to transform this understanding into artifacts that support the creative processes in humans. We will focus on the use and retrieval of cross-domain analogies, an ability that has been implicated in creative cognition.The proposed research will produce a new class of methods that exhibit computational creativity through cross-domain analogy and that involve a novel combination of two ideas. One is that the processes of analogical mapping and retrieval are goal directed, which places strong constraints on their operation. The other is that problem statements and solutions are indexed by abstract schemas that support the retrieval of relevant structures even when they have minimal surface similarity. Together, these assumptions should support robust retrieval and mapping of cross-domain analogies, which in turn should lead to novel software tools that aid human creativity. Although the primary focus of this project is on scientific creativity, the same mechanisms and techniques will prove useful in a variety of goal-directed settings that involve computer-mediated creative work, including many aspects of design. In addition to publishing scientific papers, the software for analogical mapping and retrieval will be available to the broader scientific community. In the longer term, the techniques for indexing structures in terms of abstract schemas have considerable potential for aiding the design process and for improving search for content on the World Wide Web.