Bridging the Campus Divide With ‘Dangerous Ideas’ and AI Debate Moderators

Simon Cullen wants students to argue with each other. But smarter.

“If people can’t talk and argue, how on Earth are they expected to form robust opinions?” Cullen asked in a recent interview. Yet, he said, research shows that students fear the judgment of their peers for sharing unpopular opinions about controversial topics. “This fear of engaging seriously with a range of ideas is a disaster for liberal arts education,” he said.

Cullen, as an assistant professor in Carnegie Mellon University’s philosophy department, wants but doesn’t yet have tenure. Despite lacking tenure’s protections, he has focused on teaching about controversial issues at a time when some are reluctant to. He even titled his course Dangerous Ideas.

Most Popular Stories

Most Popular

To help students sharpen their ideas, Cullen, who’s also an artificial intelligence and education fellow at the university, has required them to argue with an AI chat bot called Robocrates that he helped create. Now, he and a postdoctoral scholar, Nicholas DiBella, are experimenting with an AI program that digitally matches students with those they disagree with on issues such as abortion.

Cullen said he doesn’t know when the program will be widely available, because the required computing power is expensive. But next month, he and DiBella will offer faculty members and administrators outside of Carnegie Mellon the chance to use the program for the first time. He said it could enter perhaps 10 or 20 classrooms and, with more funding, the reach this fall could be greater.

Students can type out arguments and discussions online as an AI moderator observes, suggesting better or new lines of argumentation, or just more effective phrasing if students want it. The program also stops certain insults from reaching the other student.

Funding for the new program, Sway, and research on its effectiveness partly comes from the U.S. intelligence community, Cullen said. Cullen said only anonymized data will be shared publicly, and not even the researchers get to see actual conversation transcripts—other than for those who give signed permission to share transcripts as part of research studies. DiBella said nothing written into Sway will be sent to the intelligence community.

After an academic year when protests over Israel’s bombing of Gaza rocked multiple campuses, a presidential election looms this fall that now includes one candidate who just survived an assassination attempt. The need for faculty members to discuss the controversial issues that divide America with their students, and for students to learn how to constructively debate those topics with one another, may only grow.

Cullen said a few faculty members have told him they’re concerned that discussing these topics with students would impact their own mental health. Further, he said, educators worry about being anonymously reported to university administrators—a concern he shares.

But he said he can count on one hand the number of his students who’ve both complained that his class discussions were difficult and found the challenge wasn’t worth it. He said the feedback he’s gotten has been overwhelmingly positive. “I regularly cry when reading those reflections,” he said. “It’s certainly the most rewarding thing I’ve ever done in my career and possibly my life.”

The fruitful debates Cullen is now seeking to spread to other campuses, using AI, came out of his class that introduced students to arguments they hadn’t heard before and forced them to challenge and question one another’s beliefs, along with their own.

Dangerous Ideas

In 2021, Cullen launched a class called Dangerous Ideas in Science and Society (DISS). He taught the course through last fall, when the university said it enrolled 198 students—with more than 250 on the wait list. The course involved lectures, student-versus-student in-person debates and an early use of AI, Cullen said.

The dangerous ideas? Cullen said he, guest speakers and students discussed the morality of abortion, infanticide, euthanasia, gun rights, human genetic enhancement, humans’ obligations to other animals, along with campus speech culture itself. The topics included this ancient question, as Cullen phrased it: “If God exists, why is there so much suffering, and if he doesn’t, what’s the point of anything?”

“If you wish to avoid the possibility of being offended, you should reconsider taking this class,” Cullen told his students in an introductory video, recorded for them to watch before their first day. But Cullen also told them he didn’t mean “it’s OK to deliberately or carelessly offend people.”

When hearing others’ views, Cullen told them they “should look for the kindest, most compassionate, most intelligent and most charitable interpretation of what they’ve said.” He calls it the principle of charity, or giving the benefit of the doubt.

“If you’re offended by something, whether it’s something a classmate says or something your TA says or something I say or something in one of the readings, your job is to argue against it and to argue for your own view,” Cullen told his students. “And arguing means providing reasons.”

Cullen said one topic, on the acceptance of transgender and transracial identities, has been particularly challenging for him to teach. A transracial person is someone, such as Rachel Dolezal, a former NAACP chapter head and adjunct instructor of Africana studies, who identifies as a different race from what people perceive them to be or from what they previously identified as.

Cullen provided Inside Higher Ed data from a live poll in his class at the start of the lecture, asking students, “Which transitions are possible: Gender, Race, Both, Neither?” Nearly 40 percent of the responding students answered neither, a result that he said his transgender and nonbinary students could see broadcast on the screen at the front of the class. “My noncis[gender] students who’ve written about that [in surveys afterward] have often pointed out that the survey was challenging but they could see why it was important,” Cullen said. (He has also asked students about other topics before and after teaching about them, such as how confident they were in their views on abortion and how they felt about banning civilians from owning guns.)

In his lecture on transgender and transracial identities, Cullen used a controversial 2017 article published by Rebecca Tuvel, an associate professor and chair of the philosophy department at Rhodes College, in which she argued that society should be as accepting of people like Dolezal as they are of Caitlyn Jenner. After Cullen argued Tuvel’s point, his data shows that the number of students saying neither gender nor racial transition was possible dropped under 30 percent, and those saying that both transitions were possible rocketed from below one in 10 respondents to just over 30 percent.

He took different approaches for different topics, including bringing in a Catholic philosopher and pro-life legal activist to discuss his antiabortion views. On Fridays, students wouldn’t have lectures but met instead in groups of 20 to debate each other using a free technology tool he developed, Palaver, that allowed those wishing to speak to “bid” for more or fewer speaking points depending on how much they want to speak on a topic. “It really prevents domination, and that means that lots of viewpoints can emerge in the discussion,” Cullen said.

Cullen advocated for his approaches during a panel at last month’s conference put on by Heterodox Academy, a group that advocates for intellectual diversity in higher education. “If students never encounter arguments for views that they’re inclined to reject, if they’re not even aware of the objections that thoughtful critics have made, then, in some sense, they don’t even understand their own beliefs,” Cullen told listeners. By the end of the semester, he said his students “seem to realize that they had held passionate moral views on topics about which they were almost completely ignorant.”

He didn’t teach this spring and won’t be in the upcoming 12 months, he said. But that’s because he’s working on AI-assisted debate and conversation software that could be deployed across his university, if not higher education more broadly. He’s trying to find funding “to do this at a massive scale, if I can.”

Robocrates and Sway

One way Cullen helped students realize how poor their own arguments were in Dangerous Ideas, he said, was providing them a chat bot to argue with that would take the opposite position, speaking as though they were the authors of assigned readings. He required his students to do “argument visualization,” creating family tree–like structures where they break down assertions and uncover implicit arguments.

As part of filling these out, he said, he required them to argue with the chat bot. “A lot of my students have never had a discussion about any of these topics with anyone who will disagree with them, so they just haven’t developed the skills,” Cullen said. It’s now called Robocrates, and he said it was essentially the first AI project he worked on at the university.

When people find themselves in an enjoyable and productive discussion with someone they previously saw as “the scum of the Earth,” Cullen said, “it really changes them.”

Now, Cullen said the new tool, Sway, will help match students who disagree on a topic. He showed Inside Higher Ed how it helped facilitate a debate on abortion that evolved into a discussion on what constitutes human life and when it begins. For both sides of the debate, it suggests better phrasing and new arguments and sums up arguments for both sides, helping stalled discussions to advance. Students can reject the AI’s suggestions, though it will completely block certain speech, such as slurs and direct threats.

Among the topics Sway can help with in debates, Cullen said, is the Israel-Palestine conflict. He said he wished it had been ready to go for the antiwar protests this past academic year, but he hopes campus administrators will see it as a way to stabilize their campuses and help people understand each other amid possible continued protests in the fall.

At an August workshop, Cullen says he will present on the program to 100 people from perhaps nearly as many campuses. DiBella, the postdoctoral scholar at Carnegie Mellon who has worked with Cullen on the program, described it as less of a debate program and more a way to help people understand each other better and “hate each other less.”

“If you send a message that’s not civil … Sway will provide a suggestion of a more constructive way of expressing the thought,” DiBella said. In the case of a statement that’s “problematic but not irredeemably so,” DiBella said, users will receive a suggested different statement that they can approve, or they can ask for a new suggestion, accept and edit the suggestion, or reject it outright.

If a user outright rejects the suggestion and sends their original statement, the guide will then send both users a message trying to mediate, DiBella said. But the program has “zero tolerance for when users try to send slurs or threats, graphic content” and won’t send it.

“It’s the AI edition of the Socratic method,” DiBella said.