We've Been Looking at AI the Wrong Way

Artificial intelligence did not create our biggest problem. It simply turned up the heat.

Most governance discussions about AI focus on what the technology can now do: generate reports, write code, summarize research, answer questions, create increasingly convincing fakes, and sustain conversations. Here, I want to steer back to how we work with AI—or more precisely, with large language models (LLMs)—in the cognitive tasks of thinking, writing, analyzing, and deciding. Our knowledge work.

Understandably, most user-related governance discussions have followed a similar path. We ask whether AI was used, how much of a document it generated, and how those contributions should be disclosed.

These are important questions. But they are no longer the most important ones.

The deeper challenge is not the technology itself. It is what happens when increasingly capable AI amplifies the way humans already think, communicate, and make decisions together.

AI does not create cognitive collisions. But it does greatly increase the speed and scale at which they happen. And this matters.

A cognitive collision occurs when ideas meet but understanding does not. Assumptions pass unexamined. Certainty sidelines curiosity. Confidence outruns comprehension. We begin talking past one another rather than thinking with one another. I feel that pull myself. The attraction of certainty is not someone else's problem. It is one of the conditions we are all learning to navigate. 

These collisions are not new. They have always shaped relationships, organizations, governments, families, and friendships. Today, AI is changing their velocity—and accelerating the reactive decisions that follow.

A vessel that is one degree off course at three knots may drift only slightly before the error is noticed and corrected. The same one-degree error at forty-five knots carries the vessel somewhere entirely different. The course has not changed. The consequences have.

AI is doing something similar to human judgment.

It magnifies existing problems far faster than it creates new ones.

That is why the conversation about AI governance is beginning to change.

The central question is no longer simply:

Did AI help produce these words?

It is becoming:

Who remained capable of exercising, explaining, and revising the judgments those words embody?

That is a different kind of question.

It shifts our attention away from documenting the origin of language and toward understanding the quality of the navigation that produced it. Because working with a capable language model rarely resembles simple writing.

A question changes the answer. And that answer changes the next question. New possibilities appear. Assumptions are tested. Sometimes the destination itself changes. That is navigation. And navigation has never depended upon having perfect charts or absolute rules.

Navigation depends upon remaining in contact with reality while continually adjusting course. This matters far beyond AI.

Today's public conversations increasingly resemble cognitive traffic moving at extraordinary speed. Information arrives faster than any individual can evaluate it. Opinions harden before evidence catches up. Unexplored agreement becomes easier than considered understanding. And difference is increasingly treated as the problem, when often the real problem is the collision of ideas and thinly held beliefs.

Difference is where learning begins. We cannot help but be different. Our experiences, perspectives, knowledge, and values ensure it. Difference is not the problem. Cognitive collisions occur when difference loses relationship—when we stop becoming curious about one another and begin defending positions instead.

The power and economics of these tools still matter. But the neglected question is how we engage them as users. They are not going away, nor should these remarkable large language models. But we certainly do need some better conventions for navigating with them so we can reduce collisions. 

I spent years training young people at sea, where the ocean was our classroom and cathedral. These conventions come from that water. That is what led me to develop AI COLREGs—Seven Navigation Conventions for AI Collaboration. Not rules for using AI correctly. Rather navigational conventions for helping keep human judgment visible while collaborating with increasingly capable AI systems.

Because however capable our cognitive instruments become, the most important question remains profoundly human:

Can we still explain, defend, revise, and answer for the judgments we ask others to trust?