There are fears that OpenAI’s ChatGPT can help students cheat on tests, but how well would a chatbot do when asked to write a higher-level exam?
In a recently published study, University of Minnesota law professors asked ChatGPT to prepare answers for final exams from four courses of their faculty. The AI passed all four exams, but with an average grade of C+.
In another recent study Wharton School of Business professor Christian Terwiesch found that ChatGPT passed the business management exam with a grade of B to B-.
So, it hardly makes sense to use this technology to impress teachers.
Research groups found the AI inconsistent, to say the least. The team at the University of Minnesota noted that ChatGPT does a good job of basic legal rules and summarizing doctrine, but gets confused when trying to identify issues that are relevant to a particular case.
Christian Terwiesch said the generator did an “amazing” job on simple operations management and process analysis questions, but couldn’t handle more complex questions. It even made math mistakes at the 6th grade level. However, the bot skillfully changes its answers when it receives human prompts.
ChatGPT may not pass an exam or essay, but it can generate approximate answers and refine them.
The researchers warn that schools should limit the use of the technology to prevent ChatGPT-based cheating. Students should still learn fundamental skills and not rely on a bot for help, according to the University of Minnesota.
At the same time, researchers believe that ChatGPT can take a decent place in the classroom. Professors could teach students how to rely on AI in the workplace, or even use it to write and grade exams. Technology could save teachers time that could be spent on students, such as more student meetings and new teaching material.