The AI-Generated Article Paradox

The world of words is undergoing significant changes due to the rapid advancements in artificial intelligence (AI) and its increasing presence in digital lives. One area where this shift has been pronounced is in writing, with many news outlets, websites, and social media platforms relying on algorithms to produce articles, stories, and books.

This trend raises questions about the way we consume and interact with written content. Is it possible to spot when something has been written by a machine? Are the lines between human and AI-generated writing becoming increasingly blurred?

A recent study conducted by The Economist aimed to answer these questions by comparing the writing of top large language models (LLMs) like OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, and xAI’s Grok with that of human writers. Researchers sought to identify the hallmarks of AI-generated text.

The Quest for Distinctiveness

Identifying AI-generated writing is challenging due to the lack of a clear distinction between machine and human language. Linguist Karolina Rudnicka notes that “writers have idiosyncrasies – Emily Dickinson, for instance, loved em-dashes – and bots may do too.” This means relying on simple word or phrase detection is unlikely to yield accurate results.

The Rise of Polysyllables

The study found that AI-generated texts tend to use a significant number of polysyllabic words. These words are often rare in human writing but frequently employed by LLMs. Researchers have previously noted the increased use of complex vocabulary in AI-generated text.

However, this shift towards more elaborate language raises concerns about the value and authenticity of AI-generated content. George Orwell warned against “pretentious diction”, where writers use complicated words to sound clever. The over-reliance on polysyllables in AI-generated writing risks perpetuating a culture of artificial sophistication.

The Punctuation Paradox

The study’s findings also highlight the evolving nature of AI-generated punctuation. While many assume that LLMs are fond of em-dashes, this is no longer true after recent software updates. Only Claude uses more em-dashes than human writers, with ChatGPT using markedly fewer.

This raises questions about the role of punctuation in distinguishing between human and machine writing. The study suggests looking for texts without much punctuation at all – a characteristic associated with AI-generated content.

The Sentence Structure Conundrum

The researchers observed that AI-generated sentences tend to be long and lacking in variety, often relying on rhetorical devices like “not X but Y” or “not only but also”. While these constructions can make for engaging writing, their overuse by LLMs risks creating a monotonous and formulaic style.