Professor stops evaluating student work because AI can do it faster
AIBerkeley Professor Stops Grading, Delegates to Claude
The decision marks the first documented case of a faculty member retiring a core academic responsibility in favor of algorithmic assessment.
Dr. James Hendricks, a computer science professor at UC Berkeley, announced Tuesday that he would cease manually evaluating student assignments and instead route all submitted work to Claude, a large language model operated by Anthropic, according to a message sent to his department chair and obtained by Nextish News. Hendricks, who has taught the introductory algorithms course for twelve years, reported that the shift would reduce his grading time from approximately eight hours per week to roughly ninety seconds per assignment, allowing him to redirect his labor toward activities he described as more fulfilling.
The change affects roughly two hundred and thirty students across three sections of the course. Hendricks configured Claude to return numerical scores and written feedback within minutes of submission, using a rubric he uploaded as a text file. The model has processed four thousand and seventy-three assignments since the arrangement began last month.
The software is faster and more consistent, and it does not develop opinions about whether my lectures are any good.
"The software is faster and more consistent, and it does not develop opinions about whether my lectures are any good," Hendricks said in an email to a colleague who had expressed concern about the arrangement. He added that students had reported receiving feedback more promptly than under the previous system, and that grade appeals had declined sharply, perhaps because no student wanted to argue with a machine.
The approach has begun to ripple through the department. Two other faculty members have requested information about implementing similar systems, and a graduate teaching assistant asked whether she could delegate her grading responsibilities as well, which the department declined to authorize. A spokesperson for UC Berkeley's provost said the university was currently developing a policy to address the use of automated systems in academic assessment but declined to say whether that policy would permit or restrict the practice.
At press time, Hendricks reported that he had spent the hours previously devoted to grading on email and administrative tasks, and was now considering whether to stop attending office hours, since students seemed satisfied with the feedback they received from the model and rarely stopped by to discuss their work or their understanding of the material.
He was also exploring whether the same system could evaluate final projects, write letters of recommendation, and submit grades to the registrar, according to a follow-up message sent to his teaching assistants on Friday afternoon.