Commentary: Can you teach yourself to detect AI writing? Maybe

Before generative AI, written text was generally attributed to a human author. However, AI is reshaping writing, altering its structure and language patterns. As a result, spotting AI-generated text is becoming increasingly challenging.

One potential giveaway is the word "quietly," which appears more frequently in AI-generated prose than in human writing. Other words, such as "genuinely," "delve," and "meticulous," also show up more often in AI-generated text. These words tend to coexist with polished sentence structures and a penchant for the "not X, but Y" construction and the rule of three.

Sentence length may also be a clue. When faced with a sudden, dramatic event, humans tend to write shorter, more straightforward responses. In contrast, AI can generate lengthy, polished text almost instantaneously, even in situations that would typically elicit a brief, informal reply.

However, these indicators are not foolproof, as AI detection software is far from perfect. A 2023 Stanford study found that many popular AI detectors incorrectly flagged essays written by non-native English speakers as AI-generated. Furthermore, AI-generated content is becoming more prevalent, leading to less variation in writing style across the internet. To combat this, the European Union is now requiring all AI-generated content to be labeled or watermarked.

Ultimately, identifying AI-generated text requires a nuanced approach, considering factors such as the author's trustworthiness, knowledge, humor, and unique writing style. While AI detection software can be helpful, it should not be relied upon as the sole method for determining authenticity.