Second Chances in CS50
David Malan ’99 teaches CS50, perennially one of the most popular undergraduate courses at Harvard: Introduction to Computer Science attracted more than 750 students last fall. A doppelganger of the course is taught at Yale, and suitably tailored versions are used at Harvard’s law and business schools as well. (To get a sense of what it’s like to be in a front row seat for one of the lectures in Sanders Theatre, watch this 180-degree virtual reality (VR) recording):
The end-of-semester CS50 fair last fall, at which students presented their final projects, was a hip, raucous affair, with pulsing music. This is not a typical computer-science class.
CS50 is distinctive in another sense—more cases of academic dishonesty are reported from this course than from any other at Harvard. Last fall, 5 percent of the class was referred to the University’s honor council, a board of students, faculty members, and administrators that evaluates violations of academic integrity codified in the College’s honor code. In an interview, Malan, the Gordon McKay professor of the practice of computer science, attributed the high rate of referrals to his use of a computer program to detect cheating on the course’s challenging problem sets, which take the average student 12 or more hours a week to complete. He said the level of plagiarism is consistent with that detected at peer institutions, such as Princeton, Berkeley, MIT, and Stanford, based on his conversations with colleagues, and has been steady within CS50 over time as well, with minor fluctuations.
Nothing Malan has done to date has had much of an impact, he said, on the number of students who copy others’ work: not a discussion of the subject by the dean of undergraduate education at term’s start in 2017, nor coverage in the student newspaper; not the automatically generated reminder every time students electronically submit their weekly problem sets; not even the fact that students must write, as one of their course assignments, a software program to detect plagiarism. “Remarkably,” Malan emphasized, “no matter how forthright we are, there is a percentage of learners for whom it doesn’t resonate or sink in.”
In mid-March, Malan was scheduled to present a paper outlining the statistical data on plagiarism in CS50—and his efforts to teach academic honesty—at a conference of the Association for Computing Machinery’s Special Interest Group on Computer Science Education, in Portland, Oregon. The conference was canceled because of COVID-19, but the paper, “Teaching Academic Honesty in CS50” is available online. In it, Malan outlines the interventions that he has tried, including two that have been at least somewhat helpful to a small number of students: the course’s “regret clause,” and interventional conversations.
The regret clause originated with the teaching staff’s recognition that there was a pattern in the cases of dishonesty: they were often “the result of late-night panic, a combination of little sleep, much stress, and one or more deadlines looming,” according to the paper. As a result, course leaders added in 2014 the option to self-report the dishonesty within 72 hours of submitting the work, with the understanding that while the assignment in question might be graded zero, the matter would not be referred for further disciplinary action.
In each year since then, between 1 percent and 3 percent of the class have invoked the regret clause, and in most cases the work has been zeroed. Only repeat offenders are referred to the University’s honor council.
In 2017, Malan introduced “interventional conversations,” a strategy for heading off behavior that might lead a student to cross the line when collaborating with others. Course guidance on collaboration is that students may show their code to someone else for help, but may not look at someone else’s code. “We aspire to make these more teachable moments, as opposed to punitive moments,” he explains. When two submissions are very similar but there is no clear-cut evidence that the work was not done independently, Malan calls the two students in and asks them to explain their workflow in approaching the problem set. “We make clear to them that ‘you skated awfully close to the line that we draw on the syllabus, so let’s get you reoriented to that so that it doesn’t happen again,’” he says. Those students, he reports, have “not appeared again on our radar,” so the conversations may be working.
And yet none of these interventions have decreased the frequency of cases referred for disciplinary action. “That came as a surprise,” said Malan. “We’d hypothesized that [the regret clause and interventional conversations] would put some downward pressure on the number of cases, when in fact they’ve remained roughly the same or gone up.” Nevertheless, he feels strongly that such interventions have merit: “We’ve identified a demographic of students with whom we hadn’t previously connected who were teetering or had crossed a line, but didn’t necessarily know how to handle themselves. This allowed us to chat with them about issues of mental health, or academic struggles, or personal issues, so that we could then connect them with the right resources on campus. In years prior, those students would have been completely under the radar, and that seems the worse outcome.”
Malan wrote the paper in the hope that other educators at Harvard and elsewhere would consider adopting policies similar to those two, which “seem to have had the most positive educational impacts for students,” he said. Given the clear and detailed procedures for punishment in cases of academic dishonesty, he added, “I think it is reasonable” to balance that with “well-defined steps” that students can take to preempt punishment by “taking ownership of their actions earlier on.”