The Birth of Roles is a writing experiment processed with machine learning language model. By using either “He” or “She” as initial trigger to complete the rest part of paragraphs constantly, a huge sentence pool is created. Gradually, the characteristics of the He and She emerge, and their story grows loosely. The artist wrote a story in circular structure based on it, of the two protagonists who are on an endless adventure trip, trying to figure out what is inside their relationship, and how the whole system they exist in works, which however, are going nowhere. Sometimes they also talk about themselves and their past.

The language model employ GPT-2, which was one of the world’s biggest and best language models by 2019, trained by huge amount of corpus from Wikipedia, news, novels etc. Despite of the bias of language itself (it’s trained in English), the result somehow shows general pattern of human thinking, as well as stereotypes and biases in different aspects and layers, which, however, drives the birth of roles.

《角色的诞生》是一次写作实验 。“ 他 (He) ”或“ 她 (She) ”作为自动书写中每一 段落的初始触发词被输入程序，以获得基础的数据库。渐渐地，“他”和“她”的角色特征开始出现，故事也以松散的方式逐渐形成。艺术家根据这些素材写作了一段循环叙事结构的故事。故事讲述了两位主人公在永无尽头的历险中试图认识彼此之间的关系，以及探索他们所存在的系统究竟是如何工作的，期间穿插着他们的回忆和对话。
程序的语言模型采用了GPT-2，它由大量来自维基百科、新闻、小说等的英语语料训练而成。在不考虑语言/地区的偏见和语料的有限性的基础上，其生成结果显示了人类思维的一般模式，以及刻板印象、偏见和意识形态，然而正是这些因素促进了角色的诞生。