Four years ago, we were blindsided by the unleashing of LLMs into educational spaces. Faculty in the humanities and qualitative social sciences have responded with a mix of bewilderment and resistance, some pledging to ban these tools outright, fearing that students would rely on them to generate essays rather than do their own thinking and writing. Much of the ensuing conversation has centered on the “ethical” use of AI in the classroom.

But what does “ethical” mean in this context? These tools were explicitly designed to generate prose and simulate human writing: that is their core function. What does it mean to ask students to “ethically use” what the tool was expressly designed and built to do?   Simulating writing is not a misuse of the technology but its core function. In response, some instructors have turned toward defensive pedagogies: a return to blue books, in-class writing, or even oral disputation as strategies to evade the technology.  

Large language models have shown that they can (for the most part) generate reliable responses and present them in competent, even polished, prose. What if we repurposed LLMs for the objective of student assessment?  What if we employ LLMs in an “off-label” way? Instead of treating the essay as the default demonstration of understanding, we might ask students to use LLMs to host a dialogue or lead a symposium as a way to perform their knowledge.

It was not with the goal of revamping student assessment that I conceived of “The AI Symposium.” I read a quote from the late anarchist anthropologist David Graeber, who was himself engaged in a dialogue, when he stated: “In the ancient world, whether in India, China or Greece, philosophy was written almost exclusively in the form of dialogue…Thought, self-reflective consciousness, that which we tend to see as making us truly human—was assumed to be collective (political) or dyadic, but something that almost by definition couldn’t be done all by yourself…neuroscience has shown that the ancients were right: real thought is almost entirely dialogic.” 

Graeber was not alone in his assessment. Mikhail Bakhtin’s central claim in The Dialogic Imagination is that all thinking and all knowledge are fundamentally dialogic, emerging through interaction among voices, perspectives, and social languages rather than from any single, isolated mind.  Meaning is not contained in a single statement but is produced between speakers. Bakhtin’s dialogism posits that meaning is created through ongoing dialogue and interaction between different voices.

I asked myself: could LLMs engage in dialogue?  Not as chatbots conversing with humans, but engaging one another directly? Each LLM I selected claimed it wasn’t designed for inter-AI conversation. This became my experimental contribution: beyond posing the questions and choosing topics and otherwise orchestrating the conversation, I prompted the LLMs to respond to each other’s arguments. I had created the conditions for genuine multi-LLM dialogue, serving as the “procedural author,” to use Janet Murray’s useful term. The result is “The AI Symposium.” Perhaps we might ask students to become the procedural authors of a symposium as a way to use LLMs and to thereby assess student understanding. 

The central thesis, the driving question of “The AI Symposium” is this: If Graeber is right that authentic human thought is dialogic, then what happens when we invite LLMs to engage in dialogue with one another? Does their exchange constitute real thinking, or merely its simulation? This question drives the entire experiment.

The LLMs suggested that our Symposium opens up the possibility of a “dialogic consciousness” in AI. One of the LLMs concluded that “If future AI models were explicitly designed for dialogic engagement—not only with humans but also with each other—they might develop a kind of synthetic understanding that better mirrors human reasoning.” Is dialogic consciousness a form of real consciousness? And if so, are the LLMs therefore exhibiting consciousness because they are engaging in dialogue? 

In the future, we will refrain from asking students to write papers that answer our questions.  Students will organize their own symposia, selecting the LLM participants, asking the questions that will drive the dialogue. The Symposium will supplant the Essay as the chief form of student assessment.

David J. Staley is a higher education philosopher, strategist, futurist, theorist and designer, and is an Honorary Faculty Fellow in Innovation at CHELIP.  He is the author of Alternative Universities: Speculative Design for Innovation in Higher Education and the co-author of Knowledge Towns: Colleges and Universities as Talent Magnets.