We've made peace with AI writing our emails, reading our X-rays, driving our cars, and drafting our contracts. Ask most people whether a machine should handle logistics, diagnostics, or paperwork, and the answer is an easy yes. Ask the same people whether a machine should run for mayor, or stand in front of a classroom as the teacher, and something in them recoils.

That reaction is worth taking seriously. It isn't nostalgia, and it isn't simple fear of new technology because we're not actually that afraid of new technology anywhere else. It's a signal about what we think leadership and teaching are actually for.

The trust paradox

Here's the strange part: it's not that we trust politicians more than AI. We don't. In Gallup's long-running honesty-and-ethics rankings, members of Congress routinely land near the very bottom of the list in the single digits for public trust, while teachers sit near the top, among the most trusted professions in the country.

Meanwhile, surveys on AI adoption show something almost contradictory: a growing share of people say they're comfortable letting AI move beyond answering questions and actually act on their behalf booking things, filtering things, deciding things.

So put those two data points side by side, and you get a genuine paradox:

We trust politicians less than we trust AI to take real-world action, and yet almost no one is proposing an AI candidate for city council. We trust teachers deeply, and still we hesitate to hand a classroom fully over to a machine, even a highly capable one. Trust alone doesn't explain the resistance. Something else is doing the work.

Two different jobs wearing the same title

The mistake is treating "teacher" or "leader" as a single job that can be benchmarked and automated like any other. In reality, both roles are bundles of two very different kinds of labor:

AI is genuinely good at the left branch. It can explain a concept four different ways until one lands, flag a student who's falling behind before a human would notice, and never get tired, impatient, or distracted. That's not a small thing, it's a real and growing capability.

But the right branch is where the actual role lives. Ask anyone about a teacher who changed their life, and they almost never describe a well-explained lesson. They describe a moment of being seen someone who noticed they were struggling, believed in them before they believed in themselves, or made a classroom feel like a safe place to be wrong. That's not a delivery mechanism for information. It's a relationship.

A good policy can be modeled by an algorithm. A country in grief cannot be addressed by one.

Why "optimize the role" is the wrong goal

There's a quiet assumption buried in a lot of ed-tech and gov-tech pitches: that teaching and leading are inefficient versions of an information-transfer or decision-making process, and that efficiency is the thing to chase. Personalize the content, remove the friction, optimize the pipeline.

But if the core value of a teacher or a leader is relational trust, recognition, shared identity then "optimizing" it isn't neutral. It's a subtraction. A perfectly efficient, infinitely patient, flawlessly personalized AI tutor that never actually knows the student is not a better version of a teacher. It's a different product wearing a teacher's job title.

This is why science fiction keeps returning to the same unsettling image: a society run entirely by rational systems, technically flawless, quietly hollow. Not because the machines fail at their jobs, but because they succeed at a version of the job that was never the whole job.

What this means for anyone building AI in these spaces

None of this is an argument against AI in classrooms or civic life. It's an argument about where the line sits.

• Build for the informational half, in service of the relational half. The best use of AI in a classroom isn't replacing the teacher's attention it's freeing it up. Automate the grading and the drafting so the human has more bandwidth left for the parts only a human can do.
• Design for visible humans, not just visible answers. A tool that quietly makes a mentor more responsive is doing something very different from a tool that quietly makes the mentor unnecessary. Products should be judged on which one they're actually building toward.
• Treat the discomfort as data, not an obstacle. When users resist an AI feature that "should" be more efficient, that resistance is often pointing at a relational function the feature accidentally threatens. It's worth listening to before shipping around it.

We're not going to run out of jobs AI can help with. But leadership and teaching were never really jobs in the narrow sense they're relationships wearing job titles. That distinction is the design constraint. Build for it, rather than around it.