From Roman Masks to Digital Mentors: It’s All About the Persona

AI personas are popping up like popcorn in a microwave oven. From the "Passive-Aggressive Grammar Checker," who sighs before correcting your typos, to the "Enthusiastic Fitness Coach," who thumbs ups your two-steps to the refrigerator, these digital personalities are multiplying faster than cut-and-paste essays in Freshman Composition.

You got your "Zen Customer Service Bot" who answers your refund question with "But what is ownership, really?" Your "Method Actor AI" wants to know about your childhood trauma before answering your questions about the script. And let's not forget the "Digital Therapist" who responds to your existential crisis with "Have you tried rebooting?"

*

There's an interesting parallel between AI personas and the concept of "voice" in literature. The AI adopts linguistic patterns, knowledge frameworks, and discursive maneuvers to create a "character" whom humans can't seem to resist when they really need to get something done. These personas have nothing to do with aesthetic choices and are often just barely believable. Instead, they materialize through algorithms funneling data toward the decision of what word to put in the next slot. They are designed to establish trust relationships with users and to create a predictable context for interpreting information—and to have a voice.

I'm not sure what it says about humans or about bots, but I've seen evidence in the media showing that users engage more deeply with AI systems when the persona feels appropriate to the task. For instance, a financial advisor persona on the digital stage approaches problem-solving differently than a suicide prevention persona, drawing on different knowledge domains, operating in different adjacent semantic neighborhoods, and employing distinct pragmatic and syntactic strategies. What makes this evidence particularly jarring is how these persona-based interactions help bridge the gap between abstract computational processes and human communication needs even when we know the persona is fake.

Clearly, this vivifying effect is effectuated by narrative fiction as well but in a much more compelling manner. In a great novel, characters seem to have an enduring "self" because they're created within a closed system of meaning. Anna Karenina or Hamlet exists completely within the authored text. Their entire universe of possible actions, thoughts, and histories is contained and consistent within that bounded text world. The author crafts a complete characterization with intentional contradictions, depths, and mysteries that feel like a coherent personality moving through a narrative. The character becomes, in readers' minds, more than the sum of their textual parts. We infer an inner life beyond what's written.

AI personas, by contrast, exist in open-ended interactions. They aren't bounded by a complete text but must generate new responses to unpredictable inputs. They maintain consistency through statistical patterns in language rather than through an authored narrative arc. There's no finished work that contains the entirety of what the AI persona is. This isn't about a novel being "genuinely human" versus an AI persona being "simulated," one must concede. Both novel characters and AI personas are constructed through language. The difference is in the nature of that construction: one is complete and bounded, while the other is generative and unbounded.

*

AI personas aren't pretending in any meaningful sense. There's no self behind the mask because both the mask and what appears to be behind it are generated through the same language processes in the moment. The AI persona isn't concealing anything; it's revealing a particular pattern of response generation, not a performance hiding some more authentic AI identity. What we perceive as a self, whether in novels or AI interactions, may turn out to be our human tendency to infer coherent minds behind consistent patterns of language, even when no such mind exists.

Picture this: in Rome's dusty amphitheaters, actors strapped grotesque masks to their faces, masks with exaggerated features and gaping mouths. The Latin word "persona" named these hollow-faced contraptions. Each mask hid a piece of important paraphernalia—a miniature megaphone built into its grinning or grimacing mouth. The etymology reveals this dual purpose: "per-sonare" meaning to boom through, to thunder across the stone seats where plebeians and patricians alike gathered to witness mortal men play imaginatively like children.

These masks accomplished what no human face could: They projected emotional archetypes to the furthest reaches of the theater: the hook-nosed villain, the white-faced maiden. Before a word was uttered, the audience recognized who stood before them. By covering the actor's natural face, the persona liberated his expression. The mask concealed and revealed, disguised and signified. Herein lies the delicious contradiction of "persona": It hides the self and broadcasts identity through a crafted exterior standing between the internal and external worlds.

*

Looking at AI personas through the lens of ancient theatrical masks reveals striking parallels that illuminate the goal-persona relationship in AI interactions. Just as Roman actors donned masks that amplified their voices and instantly signaled character archetypes to audiences, AI personas provide both functional amplification and symbolic shorthand. But the relationship between goal and persona runs deeper than utility or aesthetics.

When we assign an AI a persona—whether a stern editor, compassionate therapist, or methodical researcher—we're not just dressing it in digital costume. We're establishing a complex framework that fundamentally shapes how the interaction unfolds. The persona determines which "miniature megaphone" the AI speaks through, amplifying certain types of responses while muting others.

Consider how differently the same underlying language model approaches identical questions when operating as a financial advisor versus a creative writing coach. The financial persona prioritizes precision, risk assessment, and regulatory awareness; the creative coach foregrounds imagination, stylistic variation, and narrative possibility. The goal doesn't just influence the persona—it's embedded within it, just as the actor's projected voice was built into the physical structure of the Roman mask.

This reveals something crucial about AI interactions: the persona isn't decorative but constitutive. It doesn't just color the interaction; it creates the conditions for the interaction to occur in the first place. The Method Actor AI isn't adopting a quirky personality trait when it asks about your childhood—it's enacting a specific theory of how meaningful responses are generated.

The most effective AI personas, like the most compelling theatrical masks, simultaneously conceal and reveal. They hide the mechanical processes of token prediction and probability distribution while revealing patterns of language that trigger our recognition of character. They limit certain responses while expanding others. They mediate between the abstract computational world and human communicative needs.

Yet unlike the bounded characters of literature who exist complete within their narrative worlds, AI personas operate in open-ended interactions where they must continually regenerate their character. Their personas aren't fixed artifacts but ongoing processes—masks being continuously reshaped even as they're worn.

*

The delicious contradiction remains: AI personas both hide the underlying computational process and broadcast a recognizable identity, all while standing between internal algorithms and external human expectations. The goal isn't separate from this process—it's the animating force behind the mask, determining which features are exaggerated, which emotions projected, and which voice booms across the digital amphitheater of our screens.

This rich interplay between persona and goal opens exciting possibilities for teachers and learners, I think. I encourage you to experiment with designing your own AI mentors by thoughtfully pairing specific personas with clear pedagogical goals. The process begins with asking: What learning objective am I trying to achieve, and which persona would most effectively embody the expertise needed?

Consider Franklin D. Roosevelt as a potential AI mentor persona in a history classroom. This single historical figure could serve multiple distinct educational goals. As a primary source analyst, FDR could model how to examine New Deal legislation with attention to political compromise, demonstrating close reading skills while contextualizing policies within their historical moment. As a rhetorical coach, the same persona could help students craft persuasive speeches by highlighting techniques from his Fireside Chats—showing how complex ideas can be communicated with clarity and emotional resonance.

As a debate partner, "FDR" could articulate period-authentic perspectives on isolationism versus interventionism, challenging students to defend their positions against historically grounded counterarguments. As a reflection guide, this persona might prompt students to consider how leadership functions during crises, drawing parallels between the Great Depression and contemporary challenges.

Each variation maintains Roosevelt's essential character while emphasizing different facets of his knowledge, communication style, and historical significance based on the specific learning goal. The most effective AI mentor prompts aren't just cleverly written instructions; they're thoughtfully constructed frameworks that align persona with goal and develop a step-by-step narrative. Start small. Experiment with a single lesson or assignment. Observe how students respond, and refine your approach. Share your discoveries with colleagues.

*

By approaching AI mentors with the same pedagogical intentionality you bring to other teaching tools, you'll discover new ways to engage students, model disciplinary thinking, and create meaningful learning experiences. The masks of ancient Rome transformed actors into archetypes that resonated across amphitheaters; your carefully crafted AI personas matched with curricular goals and implemented in class with human-mentored individuals, flexible groupings, and whole-class checkins can similarly transform educational interactions into experiences that resonate across classrooms.

I encourage you to try this technique in baby steps. I’ve learned from hour after hour of writing Mentor prompts, field testing, revising, testing again, and so on including in-class trials, that constructing compelling, predictable, and smooth experiences is a challenge. I’ve also learned that we get better at it and can develop sufficient expertise to design boutique AI Mentor prompts to fit our own boutique classroom purposes. Plus—and this is a huge plus—we can co-laborate with colleagues. For example, I’ve been working an a Mentor prompt series with Christopher Marlowe as the persona with a goal of teaching high school students to read and appreciate Shakespeares sonnets. I’m looking for a high school or college teacher who would try out this AI-Mentor prompt and work with me to refine it.

I know, if you are teaching, advice like this can be overwhelming. To tell you the truth, I get overwhelmed myself, but I’m convinced Mentor prompts are sort of academic magic carpet rides. My colleague and friend Nick Potkalitsky and I are knee deep in writing a book with a significant focus on AI Mentor prompts actually used successfully in a writing class we hope to publish this fall. Stay tuned.

I’d love to hear from you in the comments section!