HOW DO I SPOT A TEXT GENERATED BY AI

3 February 2024

By Anisa Pinatih

Why do we need human editors if we’ve already had ChatGPT, Trinka.ai, and other AI-powered tools doing all the work? For starters, to uphold ethical considerations and minimize the potential for bias in AI systems.

That’s a question I often receive as a human editor in the age of artificial intelligence and my simple answer. I have the more elaborate version, and it starts with how I spot a text written by or written with the assistance of AI.

But just to be clear, I’m not writing this to glamorize the role of human editors but to advocate the middle path, that is to accept the rise of AI as a fact and find a way to use it as leverage and focus on the more important task: to push academic publication forward.

Besides, we're still at the beginning of the hype cycle, and both our expectations and fears are inflated. Sooner or later, we will reach a plateau and see the technologies mature, becoming an integral part of publication practices, widespread and well-understood.

Until then, there will be questions and I feel obliged to answer them. So let’s begin with how to spot the AI watermark because if I can spot it, then so can the journal editorial teams.

1. It’s unusually “clean”

If you think that very clean drafts with almost no mistakes must have been written by a human. Think again because human-written texts are often riddled with errors and incredibly variable in style and diction. By contrast, language models never make grammatical errors or typos. The text is almost perfect lexically and grammatically, although it doesn’t necessarily convey deep meaning.

2. It’s overly formal and polite

ChatGPT often uses some unusual words and generates overly polite and formal text. For example, “Within the epistemological confines of this academic inquiry, it is imperative to embark upon an exhaustive survey of extant literature, marshaling a panoply of scholarly works spanning a diverse array of disciplines.” Like, who talks like that? Even in the context of scholarly publication, it’s a bit too much, isn’t it?

3. It’s too generic and lacks substance

Another indicator of text written by language models is a lack of detail. For example, it won’t be able to synthesize an argument on why, say, the gender gap in Indonesian politics prevails despite all the policy measures. It will only use general information about the gender gap instead of describing specific cases to illustrate what and why it happens in the specific context of Indonesia.

4. It lacks citations

Research has shown that language models cannot mimic humans in terms of subtlety in argumentation, figurative language, and anecdotes. But what’s striking in academic text is the lack of citations, especially in a long argument. Human-generated argumentation is based on research, hence backed up by references, while machine-generated text doesn't have this feature.

5. It has too many repetitions

Although AI-generated text is loaded with fancy vocabulary, the sentence structure tends to be simplistic or repetitive. In fact, fancy vocabulary is often used over and over again in a way that does not occur in human-generated text.

6. It sounds like a marketing campaign

Stylistically speaking, considering all the ingredients above, text generated by AI sounds more like a marketing campaign than a decent argumentation. It is way too optimistic, with a lot of big words but no in-depth discussion. Academic text generated by ChatGPT sounds more like an advocacy piece filled with jargon than an evidence-based description.

Nonetheless, AI can still be helpful in reducing authors’ workload. For example, in concluding this piece, I asked ChatGPT, “Do we still need human editors?” It gave me an extended answer with bullet points and the following conclusion:

“While technology continues to advance, the role of human editors remains significant. It's worth noting that the integration of technology can enhance the efficiency of the editing process, allowing human editors to focus on higher-level tasks that require emotional intelligence, creativity, and critical thinking.”

It's actually pretty neat. And I saved some time by copy-pasting the above paragraph than actually writing it myself.