Detecting AI Created Essays

Like many people within higher education, I’ve been watching the release of OpenAI’s ChatGPT tool with trepidation. The ChatGPT software is a marvel of software development. A few weeks ago, I prompted ChatGPT to write a one page essay about the life of George Washington, only to see it immediately write a paper that would receive a passing grade in most classes. While ChatGPT has been reported to sometimes give inaccurate statements, there were none in this essay. The writing, though simple in style, would be perfectly acceptable in most classes. The ease with which this could be used to cheat was my immediate concern. How would this affect student learning? Would there still be a point in assigning essays for students if they were merely going to use AI to write for them?

Those questions are still open to debate, but at least options to combat ChatGPT are starting to emerge. This morning I came across an NPR story that mentions a tool for detecting AI generated writing, it’s called GPTZero. GPTZero attempts to detect portions of a paper that are AI generated and render a score for how much of the document is AI written, which seems similar to how Turnitin delivers a score for content that is copied. While I haven’t tested GPTZero yet, I’ve already shared it with a few professors. As we get closer to finals I’m curious if we start to see papers appearing that faculty suspect are AI created. If so, this might be our best way to spot them.