Document Actions

Lecture: Generative AI-Authored Immigration Narratives: Algorithmic Sentimentality and Fictionality

When

May 28, 2024 from 06:00 to 07:30 (Europe/Berlin / UTC200)

Where

GCSC (KFR)

Contact Name

Add event to calendar

iCal

Researchers in the fields of machine learning and artificial intelligence have sought to leverage digital technologies to model story structures for several decades now (see Meehan 1977; Dehn 1981; Gervás 2016). The current generation of large language models (LLMs) such as GPT-4 and PaLM can be prompted to produce sufficiently complex narratives. As a result, technology companies are pushing forward a sweeping cultural narrative wherein AI is the ‘future of storytelling.’ In contrast, cultural narratives circulating in creative sectors about LLMs, especially since the public release of OpenAI's ChatGPT, remain concerned with plagiarism, increased circulation of fake news, and the devaluing of creative labor. Nonetheless, artists are experimenting with generative AI, and their creative experiments offer cursory insight into features of the data and parameters training the current generation of LLMs.

This presentation approaches AI-generated fictions as indexing how broad swathes of the online population thinks and feels—an archive of cultural narratives—and instrumentalizes narrative affordances of LLMs, trained on online public discourse, to grasp the “affective economies” (Ahmed 2004), or emotional collectives, forming around critical issues. I discuss findings of my ongoing project that assembles immigration-centric fictions generated through human-AI interactions with the aim of uncovering the underpinnings behind the training of transformer models, including but not limited to studying the data corpus and embeddings from open source LLMs. My method involves creatively engineering prompts that generate relevant immigration narratives from the transformer-based models (GPT), close and distant reading the generated results, as well as theoretically framing and understanding the findings by combining narratology and scholarship on natural language processing.

// Prof. Torsa Ghosal (California State University, Sacramento)