by Constantine Andoniou
ABSTRACT
The rise of large language models (LLMs) such as ChatGPT, Claude, and Gemini has reshaped writing, learning, and authorship in higher education. Detection platforms like Turnitin now classify texts as human or AIgenerated, yet these classifications are grounded in surface-level probability metrics rather than epistemic indicators of thought. This study investigates the linguistic and cognitive foundations of AI text detection and introduces the Human–Synthetic Discursivity Model (HSDM) as an interpretive alternative to binary detection. Drawing on a corpus of sixty documents analyzed through perplexity, burstiness, lexical entropy, and reflexive density, the study compares synthetic, synthetic-humanized, and authentically human discourse. The findings demonstrate that synthetic writing is governed by predictive saturation, equilibrium, and semantic closure, while human discourse exhibits cognitive elasticity and recursive reasoning. The HSDM reframes authenticity as intentional discursivity rather than statistical irregularity and argues for a shift from AI detection toward epistemic discernment in academic writing.
