인컨텍스트 러닝을 통한 ChatGPT의 시적 표현 해석 가능성 연구

Abstract

In 2023, as artificial intelligence(AI) moves beyond the realm of imitation and into the realm of creation, Korean literature is also entering the era of “creative AI,” with Slitscope’s SIA releasing a book of poetry, “시를 쓰는 이유(Why I Write Poetry).” However, the time and cost required to reach such a level is quite burdensome for the general public to attempt.
 Therefore, this study focuses on in-context learning of conversational A.I. that has already been trained through deep learning to explore the possibility of learning about A.I.’s allegory. In particular, by using ChatGPT as a research object, we can gain research advantages in terms of accessibility and functionality over previous studies on the model.
 In order to study the learning potential of ChatGPT in an in-context learning environment, this study consists of a structure of (1) pre-learning verification, (2) output request based on pre-learning, (3) related information input, and (4) verification re-execution of (2) after learning, and examines the possibility of conversational AI learning about poetic alliteration.
 The poetry texts that are input as learning information in the above process are allegorical poetry texts extracted from Yoon Gon-gang’s 『動物詩集』(Hansung Book Co., 1938) because it was judged to have value as the first Korean poetry collection with a single animal theme, which was approached and interpreted from the perspective of satirical and allegorical poetry at the time and even now.
 This study determines the learning potential of literary rhetoric for ChatGPT through the education and output process based on Yoon Gon-kang’s poems, and considers its utilization in various aspects such as education, creation, and education in the future. This will be an exploration of possibilities to solidify the position of literature in the era of artificial intelligence that will become a reality in the future.