Tracing the Humanisation of AI: A Computational Analysis of Agentic Language in News Media and Twitter Discourse on ChatGPT

Abstract

This thesis examines how artificial intelligence (AI), particularly ChatGPT, is anthropomorphised through agentic language in news media and Twitter discourse. It investigates the prevalence, development, and thematic patterns of agentic framing, language that attributes human-like autonomy, cognition, or emotion to AI systems, across these platforms. A mixed-methods computational approach was employed, analysing 3.4 million tweets and 23,000 news articles collected between 2020 and 2024. These were subjected to syntactic dependency parsing, coreference resolution, sentiment analysis, and topic modelling. The findings reveal that social media, especially Twitter, shows markedly higher levels of anthropomorphic framing compared to news media, often ascribing human-like actions, intentions, and emotions to AI entities. Core agentic verbs such as know, say, and think, along with gendered pronouns, predominantly male, were frequently used in subject positions, reinforcing the portrayal of AI as an active agent. News media demonstrated a more restrained and event-driven anthropomorphisation, often influenced by narrative context and editorial standards. The study introduces a novel cross-platform framework for detecting agentic language and offers practical insights into how linguistic framing can influence AI perception, adoption, and governance. By emphasising the implications of anthropomorphic discourse, this research advocates for more critical and transparent communication regarding emerging AI technologies.