Universities are reportedly failing to capture answers generated by OpenAI’s popular AI chatbot ChatGPT in their exams, often resulting in students receiving higher marks than their classmates who submit real answers.
A staggering 94% of university exam submissions made via ChatGPT for psychology degrees went undetected as being AI-generated, a study found. report by New Scientist.
Professor Peter Scarfe of the University of Reading and his colleagues used ChatGPT to create answers to 63 test questions for the school’s psychology department, then submitted the AI-generated answers as 33 fake students for an exam that students were allowed to take from home.
Exam graders were not informed that among the real students were 33 fake students who had submitted AI-generated work, and Scarfe and his colleagues found that only 6% of the AI-generated submissions were correctly flagged.
“On average, the AI’s answers received higher marks than real student submissions,” Scarfe says. “Current AI tends to struggle with more abstract reasoning and synthesising information.”
They also found that AI-generated answers were 83.4 percent likely to receive a higher score than students who submitted real assignments.
The researchers claim that their experiment is the largest and most extensive study to date.
Although the study only looked at studies relating to psychology degrees at the University of Reading, Scarfe says the findings should be of concern to the wider academic community.
“There is no reason to think that similar problems wouldn’t arise in other subjects,” he said.
“The results are exactly what I would have expected: we’ve shown that generative AI can generate plausible answers to simple, constrained text questions,” said Thomas Lancaster, a senior educational research fellow at Imperial College London.
Lancaster also noted that unproctored exams are always open to cheating.
“It’s highly unlikely that a time-constrained grader of short-answer questions would bring up a case of AI cheating on a whim,” Lancaster added. “This institution can’t be the only one where this is happening.”
Meanwhile, Scarfe believes it will be nearly impossible to capture AI-generated submissions at their source. “I think the whole industry needs to wake up to the fact that we need to incorporate AI into the assessments that we give to students,” he says.
As reported by Breitbart News, the data shows ChatGPT is primarily used to cheat on homework, and its use dropped significantly over the summer, only to rise again as Americans returned to school.
Last year, students in an elite academic program at a Florida high school were accused of cheating by using ChatGPT to write essays.
you can Follow Alana Mastrangelo Facebook And X is Follow,Such Instagram.
