The Wrong Question

Every few weeks another government announces its AI strategy, and every few weeks the same argument plays out. One side says this changes everything and we must move faster. The other says this is dangerous and we must slow down. Both believe they are having a debate about artificial intelligence. Neither is. They are arguing about speed when the real variable is institutional capacity.

Can the institution absorb the thing it is deciding about? Can it tell a real capability from a vendor's claim? When the system it deploys makes a mistake at scale, will anyone find out in time - and will the person harmed have any way to push back?

These are questions about governing. AI forces institutions to confront them sooner, at greater scale and with less room to hide the consequences.

The technology is rarely where failure begins. Failure begins in procurement that cannot evaluate what it is buying; budgets and administrative policies that cannot move at the speed of the problem; an organizational culture that treats a pilot as a time-limited exemption rather than an opportunity to learn; and feedback systems so burdensome that the institution stops hearing that it is wrong until the failure is on the front page. None of that is new, and none of it is primarily technical. AI simply removes the slack that used to hide it.

This is why moving faster and slowing down are both answers to the wrong question. An institution with real adaptive capacity can move quickly and safely because it can evaluate what it is told, act under uncertainty, see when it is wrong and change course. An institution without that capacity is dangerous at any speed. Telling it to go faster produces the next scandal. Telling it to stop produces a slow, quiet decline while the world moves on.

Adaptive capacity is not one thing. It is the ability to sense what is coming; see your own systems and data clearly; acquire new capability without becoming hostage to whoever sold it to you; move money, people and rules when circumstances change; act before you are certain; learn that you were wrong from the people you got wrong; and change without losing public legitimacy. An institution can be strong on some of these qualities and hollow on others. The hollow ones are usually invisible until something stresses them.

What makes this hard to talk about is that adaptive capacity is multifaceted. It is the ability to sense what is coming, see your own systems and data clearly, acquire new capability without becoming a hostage to whoever sold it to you, move money and people and rules when you need to, act before you are certain, learn that you were wrong from the people you got wrong, and change without losing the public's permission to exist. An institution can be strong on some of these qualities and hollow on others, and the hollow qualities are usually invisible until something stresses them.

AI is that stress test. It is fast, opaque, probabilistic and consequential. It presses on exactly the institutional weaknesses that slower technologies allowed us to leave unexamined. That is uncomfortable, but useful. It makes those weaknesses harder to ignore.

I built AI in the Public Sector to make them easier to examine. It is a free, openly licensed graduate course organized around a working diagnostic. Students choose a real institution and assess it across seven dimensions of adaptive capacity. By the end, they have an honest account of where that institution can adapt, where it cannot, and what would have to change first.

The course uses cases from Taiwan and Alberta, which approach institutional adaptation in almost opposite ways. It is also built around the discipline I care about more than any framework: naming the trade-offs honestly, including who bears the cost.

The syllabus builds on Jaxson Khan's Applied AI Systems and Governance and on Teaching Public Service in the Digital Age, a project I am lucky to have been part of. Both are credited properly in the course, and worth naming here too.

An AI strategy cannot supply capacities an institution does not have. Before deciding how fast to move, find out whether the institution can govern what happens next.