Ethics as Practice
Underneath Accountability is Architecture
Students are written about with their permission and identified by industry rather than by name. Anyone who preferred not to be included isn’t. Direct quotes from student work appear only where the student has agreed to that specific use.
Last week the class mapped their own thinking before building anything. Week Four asked them to turn that inward map outward: to write two or three ethical principles drawn from their own field, name the value underneath each one, and then put those principles in the same room with everyone else’s. I called it Ethics Across Fields.
The instruction I cared most about was the one line telling them this was “not a performance of the right answer.”
I meant it. Without that invitation, the exercise had a predictable shape. Fourteen professionals from fourteen corners of the working world post their principles, everyone nods, and we all agree that AI should be used “responsibly.” That’s the version that sounds like ethics and asks nothing of anyone.
The honest story is that the disagreements are where the learning lives, and so are the overlaps nobody planned.
The word almost everyone reached for.
I didn’t tell them to converge on anything. They did it anyway. Someone who works in finance, a QA tester, a real estate agent, a healthcare analyst, a manufacturing supervisor, all writing independently, from fields that barely touch, kept landing on the same idea: that the human, not the tool, owns the final product. One put it as owning “every output I send.” Another framed it as “accountable authorship.” A student pursuing a career in mental health said she stays “in the driver’s seat at all times.”
That was the first thing Week Four taught me that I hadn’t fully expected: accountability might be the nearest thing to a universal here but only until you ask what people mean by it. Almost everyone arrived at the word on their own. What they didn’t share was its weight. For most, accountability meant ownership: my name is on it, so I answer for it. For a few, it meant something far more demanding: that owning the outcome obligates you to build structures that constrain your own future behavior, even when you’d rather trust yourself. One student’s design defaulted every uncertain decision to human review and treated automation as the exception that has to earn its place, rather than the norm that safeguards get added to. That’s not the same value turned up louder. It’s a different answer to the question “what does accountability actually require of me?” And the distance between those two answers turned out to be the real subject of the week.
Where the room split.
If the convergence surprised me, the divergence taught me more. At first it looked like the split was simply about stakes. The healthcare analyst described a claims bot that could deny a member’s claim or expose the wrong person’s medical record (a HIPAA breach with a name attached to it). The student reaching for a mental health degree is worried about the marginalized communities the training data quietly overlooks. When the cost of being wrong is somebody else’s life or livelihood, the safeguards get heavy: structured “negative testing,” bias checks, formal review before anything ships.
The real estate agent’s safeguard, by contrast, was to read the listing out loud and change one sentence so it sounded like him. That is not a lesser ethic. It’s an ethic scaled to its consequences.
But stakes only explain how much caution each person brought. They don’t explain the deeper fault line which is the one I named a moment ago, between accountability as ownership and accountability as architecture. Watch where it actually lands. The real estate agent’s rule lives entirely inside himself: he is the safeguard, and the safeguard works only as long as he keeps performing it. The heaviest designs in the class did the opposite. They moved the safeguard out of the person and into the system, so it would keep working on a bad day, under deadline, or in the hands of whoever inherits the tool later. One student’s design routed every uncertain decision to a human by default; the tool has to earn the right to act on its own, and never earns it for the decisions that matter most.
Put those two next to each other and the split stops being about how careful people are. It’s about where they’ve decided their carefulness has to live. Inside a person, it’s a habit: real, but only as durable as the person’s attention. Inside a structure, it’s a constraint that outlasts the mood you’re in. The room hadn’t disagreed about whether to be accountable. They’d disagreed about whether being accountable is something you do or something you build.
The question I’m keeping.
Then a QA tester asked the question that turns the whole thing over. Most of the class, he noticed, was relying on personal habits such as a mental checklist, a pause, a promise to review. His question was essentially: how do you know your own oversight is actually catching the bias, and not just making you feel like it did? Fairness has more dimensions than a single check can see. You can confirm you’ve represented women, or people of color, and still miss women of color entirely.
That question is really the two meanings of accountability colliding. Ownership answers, “I’ll be careful.” Architecture answers, “careful isn’t enough, because vigilance can’t see what vigilance can’t see.” And what struck me is that the class had already started building the second answer, not individually, but together, in the replies. The business-development founder had proposed an “AI Constitution,” a living document with other people built into the review. Instead of nodding at it, his classmates picked it up and built it outward. One suggested giving the document a heartbeat: “a recurring ‘constitution review’ built into your SCRUM rhythm. Not every week, but maybe once a month the team could bring one real AI use case and ask: Did this reflect our constitution? Did it create any unintended harm? Did anyone closest to the work see something leadership missed?” The point, he said, was to keep the constitution from becoming symbolic: “it would make it operational.” Others pushed on how a document like that could distribute the authority to question it, rather than centralizing one founder’s assumptions under a nicer name.
That was the moment the week stopped being fourteen people posting into the void and became a room. Nobody was assigned to answer the tester. They started answering him anyway, by refusing to let a good idea stay one person’s property, which is itself the architecture move, performed rather than described. The HR professional, separately, wanted structural bias audits rather than good intentions alone. Between them they were circling what the tester named directly: at some point ownership has to be backed by architecture, because the whole point of a structure is to catch what the careful person will inevitably miss.
I don’t have this resolved for them, and I told them so. But it’s the honest frontier of the week. Ownership is where accountability starts. It is not always where accountability can safely end and, it turns out, it doesn’t have to end there alone.
What I’m carrying into next week.
Week Four was supposed to be the “compare notes” week. It turned into something more useful. The students didn’t smooth over their differences to reach a comfortable consensus, and the differences turned out to be the curriculum. A value that survives contact with a field that isn’t yours is the one worth keeping.
If there’s a thesis forming, it’s the same one that keeps returning in this class from a new angle: staying in relationship with your own judgment isn’t a solo act. Sometimes protecting it means building something outside yourself to catch what you’ll inevitably miss.
Onward.