John MacDonald: I asked an AI bot how it felt about marking exams

Canterbury Mornings with John MacDonald

With the Government wanting artificial intelligence to be used to mark high school exams, I asked a ChatGPT bot how it felt about that.

And here’s what it said: "If the Government were to ask me to mark high school exams, I would feel (if I could feel) cautious optimism, with a big dose of responsible hesitation."

The bot said, on the upside, it could process thousands of scripts quickly, never get tired, and not suffer from the end-of-day brain fog that human markers do. It also wouldn't mark one student generously and another harshly on a bad day. And it could apply marking schedules with perfect consistency.

The bot also said there would be downsides. It said exam answers aren’t always clean-cut. A brilliant, unconventional insight might not fit the marking template, but a good teacher sees its value. That’s harder for AI to interpret correctly without "massive nuance training".

It said AI can reflect biases in its training data. Even small disparities in language use, cultural references, or phrasing could disadvantage students if the system isn’t carefully designed and constantly audited.

The ChatGPT bot also told me students, parents, and teachers would rightly ask how something was marked, and public confidence could be eroded.

It said, on balance, it would be in favour of "hybrid marking”, with AI doing pre-marking and humans handling the grey areas and double-checking. So AI would be a support tool.

In short, the AI bot said: "I’m capable. But I shouldn’t be trusted alone. Exam marking is too important to hand over fully to a machine - at least not yet."

See omnystudio.com/listener for privacy information.