ChatGPT and the Distinguishability of Human-Written Speech
Author(s)
Palmer, Sadie Elise
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Powerful large language models (LLMs) like OpenAI’s ChatGPT have caused a disturbance in the writing world, bringing about issues both in the classroom and in everyday life by blurring the line between human and machine. Due to the lack of disclosure around the use of text-generating AI in written texts, humans have been left to distinguish whether texts they encounter have been written by their fellow humans or are the work of an LLM that has trained on large amounts of human data. In this thesis, I explore this new task of humans and the ethical questions that follow. Can humans reliably distinguish between human-written and AI-generated text? If so, is there an innate “human” quality about texts that an AI cannot replicate? And how do feelings towards AI indicate its future uses in our society, particularly when it comes to the ever-increasing collaboration between AI and humans? I addressed these questions using a mixed-methods approach where 16 participants engaged in an authorship activity preceded and followed by interview questions where they described how they went about determining the authorship of the 10 presented texts.
Sponsor
Date
2024-04-29
Extent
Resource Type
Text
Resource Subtype
Thesis