Modeling creativity in artificial intelligence: possibilities and limits

Abstract

The article is dedicated to exploring the possibilities and limitations of developing creative artificial intelligence, particularly the ability of machines to determine the level of creativity in the objects they produce. To assess creativity, a three-step model of novelty is proposed, including ontological, subjective, and semantic levels. Three features of creative ideas or artifacts are identified: novelty, unexpectedness, and value. The article describes the model of a competitive generative network called "CAN: Creative Adversarial Networks," which creates new artistic styles and evaluates their novelty. The possibilities and limitations of modeling humor in creative artificial intelligence are discussed. The article analyzes examples of successful work by neural networks that generate jokes and write scripts, showing that the limitation of such systems is the machine's lack of a sense of context, space, and time. Additionally, a crucial condition for successfully writing jokes is the ability to laugh at them; humans can consciously choose the topic and format of humor, while machines lack their own goal-setting. It is shown that the technologies developed to date can be generalized as "weak creative artificial intelligence" since they can create new objects but are not capable of goal-setting and reflection. However, the possibilities of artificial intelligence are constantly expanding, changing our understanding of the limits of modeling natural intelligence.