Craftsmen and Chatbots: Why AI Still Needs a Master

Last semester, one of my accounting students sent an email about the guitar in my office. He’d noticed it during a Zoom session and, as a fellow player, asked if I still enjoyed it, then apologized for the off-topic question.

I told him it wasn’t off-topic at all. It led to a conversation about Guido d’Arezzo, who invented musical notation in the 11th century, and Luca Pacioli, who codified double-entry bookkeeping in the 15th. Both men created formal structures that made it possible for others to surpass their own knowledge. What they built weren’t just tools; they were languages that made modern music and financial markets possible. Standardization turned individual insight into shared knowledge, the codebase on which institutions like conservatories, guilds, and financial systems were built. Once a melody could be written down, it could be studied, refined, and built upon by others.

These systems freed the mind from basic translation, creating the space for critical thinking and creativity to flourish. They turned personal craft into a collective discipline, one that could be taught, verified, and improved across generations.  

From there we talked about The Animals’ 1964 recording of The House of the Rising Sun, young men the same age as my student, performing a centuries-old folk song with such musicianship, discipline, and craft that their version became definitive. Sixty years later, it still moves people. Their seemingly effortless art was built on a tedious foundation of scales, chord progressions, and thousands of hours of practice and performance.

That’s the real lesson. You can’t lead an IPO on Wall Street without mastering account types and financial statements any more than you can make transcendent music without learning your instrument. Standing on the shoulders of giants still requires climbing first, and the climb is repetitive, humbling, sometimes boring, essential work.

Every professor has stories like this. The guitar in the corner of my office isn’t decoration, it’s an invitation. This is the real magic of education, whether it happens in a trade school, with a coach, or one of our hallowed universities. Societies advance through the propagation of knowledge. The Industrial Revolution succeeded not just because of machines, but because formal structures, patents, engineering manuals, standardized measurements, let knowledge diffuse rapidly. The printing press, the internet, and now AI have each increased the efficiency of knowledge transfer, but technology provides access, not judgment.

A machine can embody vast knowledge without preserving it. The Antikythera mechanism, an ancient Greek device capable of predicting eclipses and planetary motion, did exactly that. When the ship carrying it sank around 100 BCE, both the machine’s design and the understanding behind it disappeared for more than a millennium. Most people of the time were illiterate, and no institutions existed to sustain or replicate such knowledge. Mechanical clocks of comparable sophistication wouldn’t emerge again until the 14th century.

While one machine was lost to the sea, what humanity truly lost was the capacity to make it again. The Antikythera wasn’t a singular miracle, it was likely one of several such devices, but no institutions existed to preserve or transmit the knowledge behind it. When the few who understood its workings died, so did the craft itself. What vanished wasn’t metal or mathematics; it was the social architecture that carries knowledge forward.

But how does that human transmission work when teacher and student aren’t in the same room? How did Pacioli’s insight spread to merchants across Europe? How do we create systems that let knowledge endure, not just through time, but across distance, with consistency?

Every mature discipline eventually solves this problem by building systems of verification. Accounting provides a clear example. Its rules don’t come from a single authority—they emerge from overlapping institutions. The Financial Accounting Standards Board defines principles under SEC oversight, while the AICPA enforces professional standards and the IRS adds more laws and regulations.

This redundancy can feel like bloat, but it’s structural resilience. To be licensed as a CPA, you must complete a bachelor’s degree, pass a rigorous national exam, and adhere to strict standards of independence and ethics. If a company wants to raise capital in public markets, SEC rules require it to present financial statements audited by a CPA, and to remain in compliance with tax law.

This entire structure, education, licensure, independent audit, creates a system of epistemic verification. It ensures that numbers reported to investors can be trusted because they are traceable, reviewed, and subject to legal accountability. That framework doesn’t just transfer knowledge; it sustains public trust.

Imagine replacing that with a chatbot. Ask an LLM to generate financial statements and it will produce documents that look professional, formatted balance sheets, income statements, disclosure notes that sound authoritative. The output passes a visual test. It might even fool investors.

But the model isn’t applying accounting logic. It’s pattern-matching and predicting, not reasoning. There’s no audit trail, no signature, no license to revoke when the numbers prove false. And when those statements turn out wrong, and they will, there’s no accountability, no mechanism to prevent it from happening again.

This isn’t a gain in efficiency, it’s an epistemological crisis. And the decision has already been made. Capital has flowed toward infrastructure promising productivity gains. When Nvidia became the world’s most valuable company, the market signalled its choice. Automation over legacy institutions.

The institutions being displaced, universities, licensing boards, professional associations, and regulators like the AICPA and SEC, weren’t positioned to influence that choice. They couldn’t articulate what was at stake because they were too busy defending procedures instead of explaining why verification itself matters. The market didn’t reject their argument, because no argument was made. They are bloated and slow. But confusing bureaucratic inefficiency with epistemic necessity is a category error. Institutions aren’t friction that can be bypassed with a chatbot subscription, they’re the scaffolding that holds civilization’s knowledge systems upright.

What’s being built isn’t a new verification paradigm; it’s statistical modeling replacing accountability structures. The choice is made. Now we either build the alternative, or live with the consequences.

The financiers and technologists pushing automation aren’t entirely wrong about institutional failures. Many institutions still operate as if it were the Industrial Age. But the answer to bureaucratic inefficiency isn’t elimination; it’s reform. AI can streamline these systems, identify redundancies, and make verification faster without making it optional. The real question is whether we use the technology to improve institutions or to dismantle them.

The choice isn’t AI or no AI, it’s chaos or construction. And construction begins with using AI for what it truly does well: pattern recognition and classification at massive scale.

AI can read millions of court cases to identify contradictions no human could find in a lifetime. It can process decades of regulatory documents to spot redundant requirements across agencies. It can analyze institutional workflows to expose inefficiencies practitioners have learned to work around. These are classification problems, not reasoning problems and machines excel at them.

But finding patterns isn’t the same as deciding what to do about them. That’s where augmentation diverges from automation. Judges decide which precedent governs. Regulators decide whether to eliminate redundant rules. Experts decide which conflicting data to trust.

AI can be used to predict under human authority. The machine makes suggestions that are transparent and traceable. The institution retains accountability. Human verification isn’t being replaced, it’s being strengthened at machine scale.  

What we preserve isn’t nostalgia for the old system. It’s the principle that knowledge requires verification, and verification requires accountability, a signature, a license to lose, a structure that makes trust possible. Transparency—why decisions were made, how they were made, and what data they relied on—is the epistemic layer that gives institutions their value. It must be surfaced, not hidden. That’s how institutions regain credibility and how true augmentation takes shape: systems that learn from user input, just as AI does, but within frameworks of authority and accountability that are openly understood.

This brings us back to the guitar in my office and the conversation that followed. What I gave that student wasn’t information, anyone can Google Guido d’Arezzo or The Animals. What I provided was guidance.  I explained why the boring work matters, what’s foundational, what connects seemingly distant ideas. That’s the magic of human leadership in education, and why, on days like this, being a professor is the best job in the world.

That kind of transmission, master to apprentice, teacher to student, is what actually makes tools work. Economic historian Joel Mokyr observed that Britain’s advantage in the Industrial Revolution wasn’t steam engines alone, but the craftsmen trained through apprenticeship who could adapt and improve them. Formal education had a different role, not a lesser one.  Oxford produced philosophers while workshops produced mechanics. Civilization needed both, but far more of the latter.

AI is the most powerful tool humans have ever created. But without institutional grounding, without a knowledge operating system that structures how it’s used, we get efficiency without verification, fluency without understanding. We need craftsmen who can wield these tools, not just theorists who can describe them. And we need institutions designed to train and verify both.

The choice between augmentation and automation isn’t technical; it’s civilizational.

And we’re making it right now.