July 18, 2023, #46
In an article in the Chronicle of Higher Education on 7 July 2023, Tom Bartlett writes about “‘A Study Found That AI Could Ace MIT. Three MIT Students Beg to Differ.”¹ It seems that “15 authors, including several MIT professors”¹ wrote a study that “found that ChatGPT, the popular AI chatbot, could complete the Massachusetts Institute of Technology’s undergraduate curriculum in mathematics, computer science, and electrical engineering with 100-percent accuracy.”¹ The students found “‘glaring problems’ that amounted to, in their opinion, allowing ChatGPT to cheat its way through MIT classes.”¹
The students who collaborated on the critique were Neil Deshmukh, Raunak Chowdhuri, and David Koplow. The more they looked at the paper, the more they had doubts about the methodology and ultimately about the data. “The study used what’s known as few-shot prompting, a technique that’s commonly employed when training large language models like ChatGPT to perform a task.”¹ But the “examples were so similar to the answers themselves that it was, they wrote, ‘like a student who was fed the answers to a test right before taking it.’”¹ When the authors posted their critique, the positive response to their comments surprised them.
It seems that some of the paper’s authors had not expected the paper to be posted as a pre-print. Blame seems to fall on “Iddo Drori, an associate professor of the practice of computer science at Boston University.”¹ “The problems went beyond methodology. Solar-Lezama says that permissions to use course materials hadn’t been obtained from MIT instructors even though, he adds, Drori assured him that they had been.”¹ This may not be the only problematic example. A paper Drori “published last year in the Proceedings of the National Academy of Sciences”¹ made similar exaggerated claims. “Ernest Davis, a professor of computer science at New York University…” provides a balanced perspective: finding the right formulas is one thing, but “deep understanding may well take considerably longer”¹.
This story shows how easily the hype about AI can lead to unreliable conclusions. It is also an example of how easily the search for fame can seduce even top professors into becoming involved with unreliable scholarship.
1: Tom Bartlett, ‘A Study Found That AI Could Ace MIT. Three MIT Students Beg to Differ.’, The Chronicle of Higher Education, 7 July 2023, https://www.chronicle.com/article/a-study-found-that-ai-could-ace-mit-three-mit-students-beg-to-differ.
Comments