Today, the Fourth Industrial Revolution is accelerating with the development of ICT and AI, and people are enjoying benefits such as “biological human enhancement” through various advanced technologies. However, there are also concerns that the advancement of science and technology may shrink the realm of human creativity. This study explores the possibilities and limitations of AI's ability to create literary art. Since the first AI-authored literary attempt in 1984, AI's ability to create literature has improved by leaps and bounds with the introduction of deep learning. GPT-3, in particular, has learned far more text and 175 billion parameters than its predecessors, resulting in a remarkable linguistic ability. GPT-3 can produce text that is virtually indistinguishable from human writing, and some researchers describe it as creative, witty, deep, and beautiful. However, there are questions about GPT-3's capabilities. GPT-3 is limited to generating text based on existing works, does not form a coherent narrative, and requires guidance from a human user. It has also been criticized that GPT-3 plays a “probability game,” choosing words based on statistical probabilities, which is inherently different from human creation. Nevertheless, GPT-3's creations often contain compelling and eye-catching parts. In the creation of literary works, novelty is divided into novelty in the “story” and novelty in the “telling” of the story. A story requires a “fictional world” that includes characters and events, and to create a finished story is to create a fictional world. The narrative technique or style is telling, which refers to the author's distinctive style. It is still useful to examine whether AI can reach each stage. While AI may be able to imitate some forms, the question remains whether its output will have lasting and unique formal characteristics. The technology for creating AI poetry involves entering keywords based on a language model, extracting.
Read full abstract