Who is the Creator: Subjectivity Game in AIGC Narrative Generation and Adversarial Prompting

Abstract

The narrative generation of generative artificial intelligence is based on the probability prediction of large-scale data training, which naturally tends to output the narrative mode of "statistical averaging", resulting in content homogenization. In this paper, the concept of "adversarial prompt" is proposed, which is different from the dialogue mechanism of conventional prompt engineering. Through the alternating use of three types of prompts: value presupposition, reflective questioning and structural constraints, AI is forced to deviate from the generation inertia and shift to the preset critical direction. The study takes Lu Xun's Madman's Diary as the ideological engine, refines its core critical elements and encodes them into AIGC generation parameters that can be guided, and analyzes the creation of AI narrative short film The Memory Inspector as a case. According to the research, the essence of adversarial cues is not to tame artificial intelligence (AI), but to expose the creator's own thinking inertia. The creation of AIGC era is no longer just the cooperation between people and tools, but also the game between people and themselves.