Did An AI Write This?

While AI has taken on many roles over the years, one of its most controversial applications involves writing. For example, the generative AI ChatGPT is capable of producing believable-sounding fake news articles, which has sparked fears that it could be used to manipulate elections and other events.

But did an ai write this whether or not something is written by AI can be difficult, even for experts. Moreover, there’s no foolproof way to identify if text is AI-written, especially when it comes to irony and subtlety, two things that humans excel at.

The AI Writing Check: Ensuring Accuracy and Authenticity in AI-Generated Content

That’s why many teachers and other professionals are eager to have a tool they can use to confirm whether a particular text is the product of an algorithm. Last month, developers at OpenAI bowed to pressure and released a classifier that can determine if text was generated by GPT-2, an AI model that’s been used for things like generating marketing copy or rewriting the screenplay for a Friends episode.

Its accuracy is low, however, estimating that it misses 74% of AI-generated text. And while it’s an important step forward, it won’t be useful if the tool isn’t widely available and easy to use.

That’s why Darren Hick, an assistant professor of philosophy at Furman University in South Carolina, has teamed up with Tian to develop a free app called GPTZero. It was launched on Jan. 2 and already has more than 30,000 users. The app uses two indicators to distinguish between human and AI-generated text: perplexity and burstiness. Perplexity measures how complex the text is; if it’s hard for GPTZero to understand, it’s more likely to be written by humans. Burstiness compares variations in sentence length, with shorter sentences more often being generated by AI and longer ones more commonly written by humans.