Advertisement
Advertisement
Artificial intelligence
Get more with myNEWS
A personalised news feed of stories that matter to you
Learn more
When a chatbot AI passed an MBA exam, the professor who set the questions noted that ChatGPT made some basic mathematical errors, showing that AIs are ready to take over the world yet. Photo: Gabby Jones/Bloomberg

ChatGPT scores ‘B to B-’ on MBA exam, but professor who set bot the test isn’t worried about AIs taking over

  • US professor Christian Terwiesch tested AI program ChatGPT with questions from a final MBA course exam he had written, and it passed with a ‘B to B minus’
  • However, ChatGPT made some errors in simple maths, showing that AIs aren’t quite ready to take over the world from humans

Wharton professor Christian Terwiesch was sitting with his grown children around the dinner table when the subject of artificial intelligence came up. Both of his kids had been experimenting with the nascent technology in their respective fields: “one of them is interested in design … and the other one is interested in computer science”.

Eventually his son prompted ChatGPT to explain a sorting algorithm “using terms from Homer Simpson”.

“It was funny,” Terwiesch said, “it was, from a computer science perspective, correct, and it was so easy as a user interface”.

ChatGPT users can input any prompt and the tool will respond nearly instantly with an answer, written so clearly it would be easy to believe a human had written it. ChatGPT can write essays, legal briefs, and even full songs that mimic an artist’s writing style.

Christian Terwiesch, the Andrew M. Heller Professor at the Wharton School of the University of Pennsylvania. Photo: University of Pennsylvania

Terwiesch wanted to test it himself. He fed ChatGPT questions from the final exam for his operations management class – a subject he literally wrote the book on – to gauge whether the tool could pass an MBA-level course.

And it did. Terwiesch rated the bot’s “performance as a B to B-”, but noted in his research summary that some answers contained “surprising mistakes in relatively simple calculations at the level of sixth-grade maths”.

He said those mathematical shortcomings prove that we’re far away from using the current technology to replace trained professionals.

“Imagine a medical professional making a decision of what dosage of a drug to give … If you would automate it as an investment adviser, this thing would be total garbage,” Terwiesch said. In these cases, professionals need to “stay away from this technology”.

Terwiesch, who teaches in Wharton’s MBA programme, encourages fellow educators to consider “opportunities where we can think about improving our learning process” through using AI tools in the classroom.

That could include prepping testing materials – Terwiesch asked it to write new exam questions with some success, according to his research – and other ways to lighten educators’ workloads across every grade level, including elementary and public schools.

We should not believe that this is the end of human thinking and machines [starting to] take over
Christian Terwiesch, Wharton business school professor

“We all see and admire how all these teachers come to work and work their butts off in a very difficult work environment, and so I think we have to find ways to use technology to help them as opposed to just kind of worrying about the cheating point,” Terwiesch said, referring to instances where students have already used ChatGPT to cheat on tests, a concern for many critics.

When it comes to things like taking a driving licence test or a CPA exam, Terwiesch supports an outright ban, because the purpose of those exams is to certify the test taker is qualified to perform particular skills. But the purpose of teaching, and of learning, is to “engage with the material”, not simply recite it.

Terwiesch remains optimistic that tech tools can be used for good in the classroom. “Our job as educators is to use the technology to engage [students] differently,” he said.

An example he gave was using the tool to act in the place of French philosophers, giving students the opportunity to interview them in “real time”. (ChatGPT does, indeed, speak French.)

“How can we reimagine teaching French in a world where we have French pen pals available by the dozens?” he asked.

ChatGPT’s knowledge isn’t limitless, Terwiesch warned. That’s mainly because it is built on “what it has seen in the past”, meaning the technology should be viewed more as a support for developments in fields where it may be useful.

“We should not believe that this is the end of human thinking and machines [starting to] take over,” Terwiesch said.

Post