This robotics professor has a new strategy for AI era homework
Northeastern professor Hanumant Singh pairs homework with in-person quizzes, counting the lower score of the two as the final grade.
As a Northeastern graduate student studying a subject as complex as robotics, Raj Harshit Srirangam understands the temptation to rely heavily on artificial intelligence to complete assignments.
“A student might just want to get the assignment over with and get a good score,” he said. “But the professor might want the student to have certain takeaways from the course that they’ll be able to use in their professional lives.”
Harshit Srirangam is of the mind that when used responsibly, AI can be a useful tool in helping students accomplish tasks faster and to learn new concepts. But it can certainly be a slippery slope if the technology is used with no guardrails.
So he was glad to participate in a robotics masters course taught by Northeastern University professor Hanumant Singh that allowed him to use AI tools but still encouraged him to think critically.
Singh, a professor of electrical and computer engineering, is deploying a new method in his class to ensure students still are learning course concepts in the AI age. For every homework assignment the students complete in his graduate level robotics courses, they must also take an in-person quiz. The lower grade of the two tasks is the student’s final score. That way they are encouraged to perform well on both to avoid receiving a poor final grade, he said. It means they must actually study and comprehend what they produced for their homework assignment in order to ace the quiz.
“It forced me to add an extra layer of asking myself if I really understand this at a level that I’ll be able to answer any questions if I’m tested on it,” Srirangam noted.
Singh said that he modeled that teaching approach based on the mobile robotics course taught by his colleague Michael Everet, a fellow professor of electrical and computer engineering.
“The interesting part is that the lower of the two grades is going to count,” Singh said. “What’s the implication of that?” You have to do work – whether that’s yourself or with Claude — but just as importantly you have to understand the material because if you don’t understand the material, you’re not going to do well on the quiz.”
For in-class quizzes, Singh uses the game-based learning platform Kahoot. Questions are placed on a television screen and students answer them using their smartphones to write in.
Singh’s approach is part of a rising trend in academics where educators are increasingly incorporating more in-classroom exercises to reduce the opportunities for students to use AI to cheat.
Editor’s Picks
Harshit Srirangam was a student in Singh’s autonomous field robotics course this past academic year, which teaches students how to develop hardware and software systems for autonomous cars and underwater robots.
As part of that course, Srirangam had to complete four homework assignments where they had to make 3D structures out of still images through code. To help him with planning and some of the more tedious aspects of writing code, Srirangam used Microsoft’s Copilot, the technology company’s AI chat service, but did most of the heavy lifting himself.
One particular challenging assignment was reconstructing a three-dimensional statute of a Budda head statue from pictures. The AI was helpful in helping him understand the problem and for experimenting with different coding solutions, he said.
But there were also key areas where the AI made incorrect assumptions that Srirangam knew were wrong because he had read up on the material.
It’s instances like that one, and the fact that he knew he had to be properly versed in the material for an in-class quiz, that helped him learn the subject more completely.
Shalini Agrawal, who is also pursuing her masters in robotics and took an autonomous field robotics course, said Singh’s approach to AI mirrors how she used AI in her co-op at Orpheus Ocean, where she worked on improving the company’s underwater robots.
“It makes sense to write basic parts of the code and then give it to Claude to finalize it because it saves a lot of time,” she said,while acknowledging that it’s important to understand the code well in cases where the chatbot makes mistakes.
