I Built a Course from Scratch with AI. Here Is What It Did, and What It Didn't.

This past summer I set out to build a new course from nothing. PSY 440, Scientific Literacy in Life, was meant to answer a question I think every graduate will face: when psychology shows up in your life after college, as a headline, a wellness program, a dating app, or an AI chatbot, can you tell what is solid and what is marketing?

There was no textbook that fit. So I built the course with generative AI as my production partner: 14 readings, a jigsaw for every reading, and a real-world application activity for nearly every week. We are now well into the first offering, with 24 mostly senior psychology majors. Here is what that collaboration looked like, where it went wrong, and what it made possible.

What I asked AI to build

Readings. Each reading is written for my students and for this course's exact sequence. Every section opens with a real claim from popular media, quoted and referenced, then explains where it came from, what the research actually shows, and delivers a verdict: Solid, Useful but oversold, or Marketing as science. The examples come from early-career life, such as a first job, an employer wellness program, or studying for the GRE without a professor setting deadlines.

Jigsaws. On Mondays, students split the reading into expert sections, master one, and teach it to a home group. For every reading, AI helped me produce an expert card for each section, mapped to the reading's structure, along with companion slides and facilitation notes.

Real-world activities. On Wednesdays, students apply the week's tools to a case. The current set includes the mental-health response to 9/11, Theranos as a study in social influence, algorithmic matchmaking at eHarmony, and NEDA's Tessa eating-disorder chatbot. In one session, "The Hype Gap," students ask a chatbot how AI is used in their future field, then test its claims against real job postings, PsycINFO, and professional guidance. AI helps build the course, and the course teaches students to check AI.

What I kept for myself

The division of labor matters more than the tool. I set the learning goals, the pedagogy, and the standards. The AI proposes options, and I choose among them. Then I write the core intellectual content myself: the essay questions on every jigsaw card, the details of every scenario. The AI builds the facilitation structure and formatted documents around what I wrote.

In practice, AI took on the production work that would otherwise have consumed a year. I kept the teaching decisions. Every document passes my review before a student sees it, and every document ends with an acknowledgement describing who did what.

Where it went wrong

The risk of invented sources. Generative AI can produce a convincing citation to a study that does not exist. In a course about scientific literacy, that would be fatal. So every cited source was checked against the original publication, and I ran a formal citation audit across all 14 readings. It found no fabricated sources, but it did catch six smaller errors: a misattributed lead author, citations missing from reference lists, a wrong issue number, and a statistic that needed a second source. Small errors, but exactly the kind students are learning to catch.

Giving away the answers. Early jigsaw cards kept explaining the very conclusions students were supposed to reach on their own. A card that does the thinking for students defeats the purpose of a jigsaw. I had the cards rebuilt from scratch, twice, and that rule is now fixed: a card never contains the answer students are meant to derive.

Prose that sounded like AI. At midterm, students told me some readings were dense and sounded AI-written. They were right, and I took that as evidence that they were evaluating sources the way the course asks them to. We rewrote the social influence reading within two weeks in plainer language, with an everyday example for each principle and a new section on checking whether a classic study has replicated.

Activities that repeated the reading. Students also said the Wednesday activities felt like a rerun of Monday. They wanted something more authentic and harder. That is why the activities are now built around real cases, and why they are shifting toward team reports that students present to the class.

What it made possible

Materials that fit. No publisher writes for senior psychology majors at EKU heading into their first jobs. These readings do, and they follow this course's sequence rather than a textbook's chapter order.

Revision in the same semester. This is the change I did not expect. When students told me what wasn't working, I didn't have to wait until next fall. A rewritten reading or a new activity could be ready within days. Students could see that their feedback changed the course they were taking, not one for a future group.

No textbook cost. Students paid nothing for course materials.

A course that practices what it teaches. Students see in every document how AI was used and how its output was checked. The same verification standard I ask of them is applied, visibly, to their own readings.

If you want to try this

1. 1Decide what stays yours before you start. For me that meant goals, pedagogy, final choices, and the questions students answer. Write it down.
2. Ask for options, not answers. Having AI propose three versions and choosing among them keeps you the designer rather than the editor.
3. Treat every citation as unverified until you check it. Look up the original publication. Statistics are the highest-risk claims.
4. Write down your design rules. "Never put the answer on the card" only became reliable once it was an explicit rule.
5. Ask students early. A midterm reflection told me more about my materials than any amount of my own review.
6. Disclose. An acknowledgement on every document costs nothing, and it shows students what responsible use looks like.

The course is not finished. Next spring I may reorganize it around the four core tools rather than around topics. But I would not go back to building a course the old way.

 

AI acknowledgement: This post was drafted with Claude Opus 5.5 (Anthropic) from the course record and Dr. M. Perry Winslow's decisions and experiences across the PSY 440 build. Dr. Winslow designed the course, made every pedagogical decision described, and is responsible for the final text. Claude drafted and organized the post.

When the Algorithm Thinks It's in Charge: McGilchrist's Master and Emissary in the Age of AI

 

I've been teaching a course called "Living with AI" this semester, and we keep circling back to a question that feels increasingly urgent: Who's actually in charge here?

Not in some dystopian, Skynet-takes-over sense. I mean in the mundane, everyday sense of how we're using these tools. Are we directing AI to serve our purposes, or are we subtly reshaping our purposes to fit what AI can do?

Enter Iain McGilchrist and his brilliant Master and Emissary framework.

The Original Story

McGilchrist tells us about a wise master who sends out a clever emissary to handle specific tasks in the world. The emissary is good at his job—focused, analytical, efficient. But over time, he forgets his place. He starts to believe that his narrow, systematic way of seeing things is the only way. He mistakes the map for the territory, the representation for reality itself.

The emissary, McGilchrist argues, is like our left hemisphere—brilliant at breaking things down, categorizing, manipulating parts. The master is our right hemisphere—attuned to context, relationships, the living whole of experience.

When I first encountered this idea years ago, it felt like a useful metaphor for understanding cognition. Now, teaching about AI, it feels like a warning.

The New Emissaries

Here's what I'm noticing: AI tools are spectacular emissaries. They're narrow, focused, and targeted. They work with the known and fixed, analyzing parts and categorizing information with inhuman speed. They deal in explicit, literal data—the kind of thing that can be measured, optimized, and scaled.

Sound familiar? These are precisely the characteristics McGilchrist attributes to the left hemisphere, to the emissary.

And like any good emissary, AI can be incredibly useful. It can draft emails, summarize articles, generate code, analyze patterns in data. When it serves our broader vision—our "master" capacities for context, meaning, and embodied judgment—it's transformative.

But here's where it gets tricky.

When We Forget Who's Master

I'm watching my students (and myself, if I'm honest) start to defer to AI in ways that should concern us. Not because AI is dangerous, but because we're outsourcing the very capacities that make us human.

A student asks ChatGPT to write an essay about a poem. The AI produces something technically competent—it analyzes parts, categorizes themes, provides explicit interpretations. But it completely misses what the student might have discovered through their own broad, vigilant, open engagement with the text. It can't access the implicit meaning that emerges when you sit with ambiguity. It has no embodied experience to draw from.

The student gets a decent grade. The emissary has done its job. But the master—the student's own capacity for meaning-making—has been bypassed entirely.

Or consider how we're using AI in education more broadly. We're asking it to grade essays, detect plagiarism, personalize learning paths. All tasks that require narrow, focused analysis of known, fixed criteria. But education, at its best, is about attention to the new and unique in each student, about seeing context and relationships, about fostering the kind of living engagement with ideas that can't be reduced to metrics.

When we let AI's capabilities define what education is—when we optimize for what can be measured rather than what matters—the emissary has become the master.

The Trust Students Connection

This connects directly to something I've been writing about for years: the importance of trusting students.

Traditional education often operates from a place of suspicion. We design systems to catch cheaters, to control outcomes, to manipulate student behavior through grades and penalties. This is emissary thinking—narrow, focused on parts (individual assignments, test scores), working with the known and fixed (rubrics, standards, policies).

But real learning requires master thinking. It requires broad, open trust that students are capable of growth. It demands attention to context and relationships—understanding that a student's work exists within the larger story of their development. It means being attentive to the new and unique rather than forcing everyone through the same standardized process.

AI can amplify whichever approach we choose.

If we use it to police and control—to detect AI-generated text, to automate grading, to enforce compliance—we're letting the emissary run the show. We're reducing education to what can be analyzed, categorized, and manipulated.

But if we use it as a genuine tool—to free up time for deeper conversations, to provide scaffolding for struggling students, to handle administrative tasks so we can focus on embodied, relational teaching—then we keep the master in charge.

Staying Master

So how do we stay master in our relationship with AI?

First, we need to recognize what AI is good at—and what it isn't. It excels at emissary tasks: analyzing, categorizing, working with explicit information. It's terrible at master tasks: grasping implicit meaning, understanding context, making embodied judgments about what matters.

Second, we need to use AI to serve our broader purposes, not let its capabilities define our purposes. Ask: "What am I trying to accomplish?" Then ask: "Can AI help with that?" Not the other way around.

Third, we need to cultivate our master capacities. Spend time with ambiguity. Practice seeing wholes, not just parts. Trust your embodied sense of what's meaningful, even when you can't quantify it. These are the muscles that atrophy when we defer too much to algorithmic thinking.

Finally, in education specifically, we need to design for trust, not control. Use AI to support student agency, not to police it. Focus on the living, unique relationship between teacher and student, not just the fixed, mechanical delivery of content.

The Choice

McGilchrist argues that Western culture has been increasingly dominated by left-hemisphere, emissary thinking for centuries. We've privileged what can be measured over what matters, representations over reality, control over connection.

AI didn't create this problem. But it can certainly accelerate it—if we let it.

The good news? We still get to choose. Every time we use these tools, we're making a decision about who's in charge. Every time we design a lesson, grade an assignment, or respond to a student, we're choosing between emissary thinking and master thinking.

The algorithm doesn't get to decide. We do.

Let's make sure we remember that.

---

What's your experience with AI as emissary vs. master? I'd love to hear your thoughts in the comments.

---