Why Expertise Still Matters (And What That Means for AI Literacy)

Revisiting Mollick's argument about expertise in the context of disciplinary approaches to student AI literacy

Thanks for reading Educating AI! If you find this article valuable, I’d love it if you’d like and share it with colleagues who might benefit from the conversation.

I publish two in-depth articles every week exploring AI literacy, implementation strategies, and the future of education. Your support as a paid subscriber helps sustain this work and keeps these resources freely accessible to educators everywhere. If you’ve been finding value in these weekly insights, I hope you’ll consider upgrading to a paid subscription.

Now, let’s dive in...

---

It’s been over a year since Ethan Mollick’s Co-Intelligence offered one of the first comprehensive frameworks for thinking about AI in education and work. In that time, we’ve seen an explosion of AI tools, policies, and pedagogical experiments. But I want to return to what I believe is Mollick’s most important—and most misunderstood—argument: his claim that expertise still matters in an age of AI.

Many educators latched onto Mollick’s discussion of how AI democratizes access to certain capabilities, compressing the skill gap between novices and experts. This is real and important. But his argument about the continued necessity of expertise deserves renewed attention, especially as we move beyond the initial excitement phase of AI adoption and toward more sophisticated implementation.

Here’s why: Mollick’s expertise argument provides the theoretical foundation for disciplinary-specific AI literacy. And disciplinary specificity, I’ve come to believe, is the only viable path forward for meaningful AI integration in K-12 education.

---

Mollick’s Case for Expertise

Mollick introduces the concept of the “jagged frontier,” the idea that AI capabilities are unpredictable, excelling at expert-level performance in some areas while failing completely in others. This creates a fundamental problem: AI can produce plausible-sounding outputs that are deeply flawed, and without expertise, users cannot distinguish between the two.

His argument about expertise boils down to four key functions that experts provide in an AI-saturated environment:

1. Recognizing when AI output is wrong: Domain knowledge allows you to catch errors, oversimplifications, and logical flaws that would slip past a novice.

2. Knowing what questions to ask: Expertise helps you frame problems effectively and prompt AI toward useful directions.

3. Integrating AI output into larger systems: Just because AI can generate something doesn’t mean it fits appropriately into a broader context, workflow, or argument.

4. Exercising judgment and taste: Especially in creative or strategic domains, distinguishing between “adequate” and “excellent” requires expertise.

Critically, Mollick isn’t arguing that expertise works the same way it always has. The function of expertise is shifting from production to evaluation, from creation to curation and critique. Experts are less necessary for generating first drafts and more essential for assessing quality, catching errors, and making strategic decisions about what to do with AI-generated content.

---

But What IS Expertise?

Here’s where Mollick’s argument needs extension: when we say “expertise is necessary,” we need to be precise about what expertise means. This isn’t generic critical thinking or general media literacy. Expertise is fundamentally disciplinary.

A historian evaluating an AI-generated historical essay isn’t just fact-checking dates. They’re assessing whether the narrative oversimplifies causation, whether it engages with historiographical debates, whether it treats primary sources appropriately, whether it acknowledges uncertainty where evidence is contested. They’re thinking like a historian.

A mathematician evaluating an AI-generated proof isn’t just checking calculations. They’re examining the logical structure, looking for gaps in reasoning, assessing elegance and efficiency, considering whether the approach reveals deeper mathematical insights. They’re thinking like a mathematician.

A scientist evaluating an AI-generated lab report isn’t just verifying measurements. They’re examining whether the experimental design is sound, whether confounding variables are addressed, whether conclusions are warranted by the data, whether the work is situated properly within existing research. They’re thinking like a scientist.

This is what makes Mollick’s expertise argument powerful for education: it’s essentially an argument for disciplinary thinking as the foundation for AI literacy. Students need to develop the habits of mind, questioning strategies, and epistemic frameworks that define how knowledge is constructed and validated within disciplines. Without this grounding, they cannot perform the evaluative work that expertise requires.

---

The Possibility Literacy Connection

This is where my framework of Possibility Literacy becomes operationally useful. I’ve argued that students need five core strategies for engaging critically with AI: Pattern Recognition, Directed Divergence, Integrative Synthesis, Source Archaeology, and Collaborative Governance. But these strategies aren’t generic—they’re enacted through disciplinary lenses.

Pattern Recognition in an English classroom means noticing how AI handles literary devices, narrative structure, or rhetorical moves. In a history classroom, it means recognizing how AI constructs causation, handles periodization, or privileges certain historical actors. In a science classroom, it means identifying how AI approaches experimental design, data representation, or scientific argumentation.

Source Archaeology (tracing the origins and influences embedded in AI outputs) looks entirely different across disciplines. An English student doing Source Archaeology on an AI-generated poem examines poetic traditions, genre conventions, and intertextual echoes. A history student traces historiographical schools of thought and archival gaps. A science student identifies theoretical frameworks and methodological assumptions.

Directed Divergence (deliberately steering AI away from default patterns) requires disciplinary knowledge to know what directions are worth exploring. You need to understand what’s conventional in a discipline to productively diverge from it.

Integrative Synthesis and Collaborative Governance similarly depend on disciplinary expertise for their enactment. You can’t synthesize AI output with human insight if you don’t understand the disciplinary standards for what counts as insight. You can’t govern AI use appropriately if you don’t grasp the ethical and epistemic norms of your field.

This is why AI literacy cannot be taught as a standalone unit or generic skillset. It must be embedded within disciplinary instruction, where students develop expertise through the work of evaluating AI outputs against disciplinary standards.

---

Practical Techniques: Disciplinary Expertise Through AI Evaluation

So what does this look like in practice? Here are four classroom strategies that embody the principle of developing disciplinary expertise through AI evaluation:

1. The Disciplinary Audit

Give students an AI-generated text in your discipline and ask them to conduct a “disciplinary audit.” The prompt: What would an expert in this field notice about this text? What questions would they ask? What would concern them?

• In ELA: Students audit an AI-generated literary analysis, identifying where it misreads textual evidence, oversimplifies themes, or ignores narrative complexity.

• In History: Students audit an AI-generated historical explanation, examining how it constructs causation, handles chronology, and engages (or fails to engage) with historical debate.

• In Science: Students audit an AI-generated lab report, assessing experimental design, data interpretation, and whether conclusions are warranted by evidence.

The key is making disciplinary thinking visible and explicit. Students aren’t just saying “this is wrong.” They’re articulating how experts in the field would evaluate it and why disciplinary standards matter.

2. The Jagged Frontier Mapping Exercise

Have students use AI to generate multiple outputs on the same disciplinary task, then map where AI crosses from competent to incompetent territory in your specific field.

• In Math: Generate solutions to increasingly complex proofs. Where does AI’s reasoning break down? What types of mathematical thinking does it handle well versus poorly?

• In World Languages: Generate translations of increasing complexity. Where does AI lose cultural nuance, idiomatic expression, or contextual appropriateness?

• In Social Studies: Generate arguments about increasingly complex ethical or political questions. Where does AI’s analysis become superficial, miss cultural context, or oversimplify?

This exercise develops students’ ability to recognize the boundaries of AI capability in their discipline, which is essential expertise for working with AI effectively.

3. The Expert Comparison Protocol

Show students an AI-generated text alongside an expert-generated text on the same topic (both anonymized). Ask them to identify which is which and, more importantly, how they can tell.

The discussion that follows makes disciplinary standards concrete. Students articulate what makes writing, thinking, or problem-solving “expert” in your field. This isn’t about dismissing AI. It’s about developing the evaluative capacity that Mollick argues is essential.

Variation: Use student work, AI work, and expert work, all anonymized. This removes the binary human/AI frame and focuses attention on quality markers within the discipline.

4. The Disciplinary Revision Challenge

Students generate an AI draft, then revise it to meet disciplinary standards. The revision must be justified: Why did you make each change? What disciplinary principle or standard guided your decision?

• In English: Revise an AI-generated essay to deepen analysis, complicate arguments, attend more carefully to textual evidence, or improve style and voice.

• In History: Revise an AI-generated historical narrative to engage with historiographical debate, acknowledge evidence limitations, or complicate causal explanations.

• In Science: Revise an AI-generated lab report to strengthen methodology, address confounding variables, or make claims more precisely proportional to evidence.

This strategy positions AI as a generator of raw material that requires disciplinary expertise to shape into quality work. It makes the evaluative and curatorial work of expertise central to the learning process.

---

Beyond the Readiness Trap

Mollick’s expertise argument, properly understood, offers a way out of what I’ve called the “readiness trap,” the assumption that teachers and students must master AI tools before they can engage meaningfully with them.

If expertise is about evaluation rather than production, then students don’t need to become AI experts before using AI in disciplinary contexts. They need to become disciplinary thinkers who use AI as an object of inquiry and analysis. They develop expertise through the work of evaluating AI outputs against disciplinary standards.

This is why disciplinary-specific AI literacy (DSAIL) isn’t just one approach among many. It’s the approach that aligns with how expertise actually works in an AI age. Generic AI literacy courses, standalone units on “AI tools,” or technology-first implementations all miss the point: AI literacy is inseparable from disciplinary literacy because expertise itself is disciplinary.

Teachers don’t need to wait until they’ve mastered AI to begin this work. They need to leverage their existing disciplinary expertise, making their own expert thinking visible to students as they evaluate AI outputs together. This is collaborative inquiry, not top-down instruction, but it’s inquiry guided by disciplinary frameworks and standards that teachers already possess.

---

The Stakes

Mollick’s argument about expertise isn’t just theoretical. It has profound implications for how we structure AI integration in schools. If expertise remains necessary, and if expertise is fundamentally disciplinary, then:

• AI literacy cannot be relegated to standalone courses or computer science classes

• Every discipline needs to grapple with how AI intersects with its particular ways of knowing

• Teachers’ disciplinary expertise is more valuable than ever, not less

• Students need extensive practice evaluating AI through disciplinary lenses, not just using AI tools

The educators who are getting AI integration right aren’t teaching “AI skills” as a separate domain. They’re teaching history, English, science, and mathematics with AI as a tool, a text, and an object of disciplinary analysis. They’re using their expertise to help students develop the evaluative capacities that matter in an age where content generation is cheap but judgment remains expensive.

Revisiting Mollick reminds us that the challenge isn’t keeping up with AI’s capabilities. It’s helping students develop the disciplinary expertise to navigate an AI-saturated world thoughtfully and critically. That’s work that belongs in every classroom, embedded in every discipline, guided by teachers’ existing expertise.

And that’s exactly where it should be.

Nick Potkalitsky, Ph.D.

---

What does disciplinary AI evaluation look like in your classroom? I’d love to hear how you’re helping students develop expertise through critical engagement with AI outputs. Hit reply and let me know what’s working, and what’s challenging, in your context.

---

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy