The Circle of Inquiry: Socratic Seminars
Deep Dives Into Assessment Methods for the AI Age, Part 3
This post follows my standard early access schedule: paid subscribers today, free for everyone on January 20.
This series on AI-resistant assessment began with the design critique, a method that makes thinking visible through the real-time defense of creative work. Part 2 examined video logs, where students record themselves thinking aloud, preserving the embodied reality of cognition through multimodal evidence that anchors assessment in physical presence. Both methods share a common principle: they shift the locus of assessment from artifact to performance, from what students can eventually produce to what they can demonstrate through immediate, observable engagement.
But both the critique and the vlog remain individualistic. The design critique, though often conducted in group settings, ultimately evaluates each student’s defense of their own work. The vlog captures a single student’s thinking process, recorded in isolation and submitted for instructor or peer review. These structures are pedagogically sound, and as I’ve argued, they create robust barriers against AI-generated work. Yet they represent only one dimension of what I’ve called the dialogic institution—the reimagining of education around human interaction as the primary site of learning and assessment.
The following third installment examines a method that distributes both the intellectual work and the evaluative criteria across the entire learning community: the Socratic Seminar. Unlike the vlog, which documents individual thinking, or the critique, which tests individual defense of work, the Socratic Seminar makes the conversation itself the object of assessment. Students must not only articulate their understanding but show their capacity to think-with-others in real time.
From the Atelier of Athens: The Socratic Method’s Educational Lineage
While the method bears Socrates’ name, its contemporary incarnation owes more to mid-20th-century American educational philosophy than to ancient Athenian practice. To understand why the Socratic Seminar works as assessment, we need to trace two intellectual lineages: the classical tradition of dialectic inquiry and the democratic education movement that brought it into modern schools.
The historical Socrates, as depicted in Plato’s dialogues, practiced elenchus—a form of cross-examination designed to expose contradictions in his interlocutors’ beliefs. This was fundamentally asymmetrical. Socrates controlled the conversation, selecting whom to question and determining when an answer was sufficient. The goal was not collaborative knowledge-building but individual recognition of ignorance, the first step toward philosophical wisdom.
The modern Socratic Seminar inverts this power structure. In 1982, philosopher Mortimer Adler published The Paideia Proposal, a manifesto arguing that American education had fragmented into vocational training for some students and intellectual development for others. Adler organized this curriculum around three distinct types of learning: gaining organized knowledge through lectures and readings, developing intellectual skills through coaching and practice, and what he called the “enlargement of understanding of ideas and values”—achievable only through Socratic dialogue.
Adler’s insight matters because it identifies exactly where traditional assessment fails in the AI era. The first two columns are precisely what large language models excel at demonstrating. They can recall facts, summarize texts, and execute procedures. But Adler’s third column requires something different. It demands the weighing of competing values, the synthesis of multiple perspectives, and the adjustment of one’s thinking based on reasoned challenges from others. This work happens only in dialogue, through what Adler called the “maieutic” process.
The constructivist theory of Lev Vygotsky provides additional theoretical grounding. Vygotsky’s concept of the Zone of Proximal Development describes the space between what a learner can do independently and what they can achieve with guidance. In traditional instruction, the teacher provides this guidance. In a Socratic Seminar, the guidance is distributed across the peer group. Students collectively construct understanding through what Vygotsky called “intersubjectivity”—the shared meaning arrived at through social negotiation.
This theoretical foundation reveals why the Socratic Seminar resists AI. A language model can generate text that sounds like analysis. It cannot engage in genuine intersubjectivity because it lacks a subjective position to begin with. It cannot adjust its mental model based on a peer’s emotional reaction or revise its interpretation when confronted with textual evidence it overlooked. The social negotiation of meaning that Vygotsky identified as central to human learning is beyond the reach of current AI systems.
Why the Socratic Seminar Resists Artificial Intelligence
The resilience of the Socratic Seminar stems from its insistence on presence, spontaneity, and social responsiveness. The challenge is not that AI cannot take part in discussions about texts, but that it cannot show the specific human capacities we’re actually assessing.
Consider what happens in a well-designed seminar. A student makes a claim about a text: “Brutus was ultimately a patriot, not a traitor.” Another student challenges this: “But doesn’t his speech at the funeral reveal his self-deception about his motives?” A third student enters: “I think you’re both right—he genuinely believed in the Republic, but Shakespeare shows us he’s also motivated by envy of Caesar’s power.” The conversation continues, with students building on, refuting, and refining each other’s interpretations.
This exchange shows several capabilities simultaneously. Students must track the developing conversation, reference specific textual evidence, acknowledge the validity of opposing views while maintaining their position, and adjust their thinking based on additional considerations raised by peers. Most importantly, they must do all of this in real time, without the luxury of revision or the ability to generate multiple responses and select the best one.
An AI can certainly generate thoughtful-sounding commentary about Julius Caesar. But when asked to reconcile its interpretation with a peer’s conflicting reading, to locate the specific passage that supports a claim made three minutes ago, or to explain how its current position differs from the view it expressed at the beginning of the discussion, the limitations become apparent. While an AI can track a thread within a single session, it lacks the embodied memory of the room’s social dynamics and the cumulative, shared history of the class’s prior discussions.
There’s also the matter of what Donald Schön called “reflection-in-action,” which we covered in Part 1 of this series. In a seminar, students must read the social dynamics of the group, deciding when to speak, when to hold back to let a quieter voice emerge, when to challenge directly and when to build bridges between opposing positions. They must notice when the conversation has stalled and offer a new angle, or recognize when an idea needs more development before the group moves on. These are acts of social intelligence that require presence and real-time responsiveness to subtle cues—body language, tone, pacing, and group energy.
Finally, the seminar assesses something that no artifact can capture: the capacity for intellectual humility and growth. When a student says, “I came in thinking X, but after hearing Maria’s point about the marketplace scene, I now see Y,” they’re showing metacognitive awareness of their own learning process. An AI can simulate this language easily. But in a live seminar, the teacher has observed the student’s initial position, tracked their engagement with competing views, and witnessed the moment of genuine revision. The process is visible in ways that make authenticity verifiable.
The Three Structures of Socratic Dialogue
The Socratic Seminar is not a single method but a family of related structures, each serving different pedagogical purposes and offering different assessment opportunities. Understanding these variations allows you to match the format to your learning objectives and class size.
