A new tool was recently created at Harvard that can predict whether a given body of text is AI-generated. It operates based on an algorithm that’s studied millions of words that are both human and bot test-generated. It can distinguish between the two by looking for specific clues such as grammar.
It also uses two other indicators to determine if the text is likely written by a robot. The first is a measurement of how difficult the text is to understand. The second is a measure of the variety of sentence structures used in the text. Bots tend to use sentences that are much more uniform in structure.
The Art of Bot Testing: Strategies to Perfect Your Conversational AI
While it is not foolproof, the app can be a useful tool for determining if an article was written by a robot or not. Tian says that he designed the app to combat what he sees as increasing instances of students turning in AI-generated papers as their own.
One of the biggest giveaways that an article was written by a bot is repetition. Bots are great at reciting facts but they lack the analytical skills necessary to extrapolate meaning from those facts. For example, if a blog post contains several different variations of the same point, it is likely that the content was written by an AI. This is not the case with human writers who can use their imagination and intuition to create unique insights that are not found in other writings on the same topic.