Feng CAO
Centre for English Language and Communication (CELC)
Cao, F. (2023). Does AI-generated writing differ from human writing in style? A literature survey [Lightning talk]. In Higher Education Campus Conference (HECC) 2023, 7 December, National University of Singapore. https://blog.nus.edu.sg/hecc2023proceedings/does-ai-generated-writing-differ-from-human-writing-in-style-a-literature-survey/
SUB-THEME
AI and Education
KEYWORDS
AI-generated writing, human writing, ChatGPT, style, linguistic features
CATEGORY
Lightning Talks
ABSTRACT
Artificial intelligence (AI) has witnessed significant advancements recently, leading to the emergence of AI-generated writing. This new form of writing has sparked interest and debate, raising questions about how it differs from traditional human writing. One popular AI tool which has been attracting much attention since 2022 is ChatGPT, which has been used to create texts in many domains. In this preliminary survey of literature, I aim to review studies which compare the writing generated by ChatGPT with human writing to explore the rhetorical and linguistic differences in style.
This literature survey focuses on the most widely used databases: Google Scholar, Scopus, and Web of Science. An initial search in these databases using key terms such as “AI-generated writing”, “human writing”, and “ChatGPT” returned over 400 items relevant to the topic. I skimmed through the titles and abstracts, and sometimes the full texts to assess their relevance to the research question. Irrelevant items and duplicates were excluded, and only the most pertinent sources were further analysed.
The preliminary analysis showed that the AI-generated writing differed from human writing in a number of genres and disciplines, for example, medical abstracts and case reports, business correspondence, restaurant reviews, and academic essays. Regarding content creation, for example, the literature review shows that AI is capable of generating highly readable medical abstracts and case reports which are almost indistinguishable from human writing. However, a few key limitations, such as inaccuracies in content and fictitious citations, were also reported by expert reviewers.
In terms of tone and voice, the analysis reveals that human writing differs from AI-generated writing by evoking emotions and resonates with readers on a personal level. Human writers bring their life experiences, cultural background, and empathy into their work, enabling them to convey complex emotions, capture nuances, and engage readers’ emotions. AI-generated writing, however, typically lacks the emotional depth and intuition present in human writing.
In terms of linguistic features, the literature indicates that AI-generated writing tends to employ longer sentences than human writing, but the latter is likely to employ more diverse vocabulary and expressions. In addition, AI-generated writing contains a more formal register whereas human writing is more likely to use informal register such as the frequent use of personal pronouns.
In short, this survey of the literature provides an initial overview of some key differences between AI-generated writing and human writing. While AI models like ChatGPT have made remarkable advances in mimicking human writing, they still lack the distinct characteristics that make human writing unique and emotionally resonant. Understanding these differences is vital for harnessing the potential of AI-generated writing while mitigating potential risks and challenges. In the field of language education, a better understanding of the differences between AI- and human writing may help teachers and novice writers to better utilise AI tools for developing academic writing skills and publishing. At the same time, by addressing ethical concerns and nurturing human creativity alongside AI capabilities, teachers and learners can navigate the evolving landscape of AI-generated writing, and leverage it to enhance human expression and communication in a responsible and inclusive manner.
