Some governance concepts become so familiar that we stop asking whether they still describe reality.
There was a time when the horse transformed civilisation. It moved armies, built economies, connected communities and redefined what societies were capable of. Entire industries and institutions were organised around it.
Today, the horse remains highly visible.
It appears at royal processions.
It serves ceremonial duties.
It still commands enormous respect.
It reassures us that some traditions endure.
Nobody, however, mistakes it for the mechanism that now moves modern transport.
Its symbolic role survived long after its operational role changed.
Quietly.
Completely.
Without anyone convening a meeting to approve the transition.
The difference is that everyone knew.
When a family bought a motor car, nobody imagined the horse remained the primary means of transport. Society recognised that its operational role had changed while its symbolic role endured.
I wonder whether Human in the Loop is approaching a similar moment.
With one important difference.
Many organisations still behave as though the horse is pulling the cart.
Human in the Loop tells us that a person remains somewhere within the decision process.
It does not tell us whether that person still exercises meaningful authority over the outcome.
In practice, they are increasingly treated as though they were interchangeable.
They are not.
Consider a typical AI supported decision.
Before a human sees anything, the system may already have gathered the relevant information, filtered what appears important, framed the available options, generated recommendations and expressed confidence in its preferred outcome.
None of this is inherently problematic.
People have always worked within information architectures.
Board papers.
Briefing documents.
Analyst reports.
Consultant recommendations.
The difference is not that information is framed.
It always has been.
The difference is that the framing increasingly occurs inside systems that remain largely invisible to the people expected to exercise judgement over their outputs.
By the time a human is asked to approve a recommendation, something important has already happened.
The decision landscape has been constructed.
The human is now operating within it.
The governance record captures the final act.
The approval.
The signature.
The click.
Those things matter.
They establish accountability.
They do not necessarily establish authority.
A person who lacked the information, the access or the practical freedom to have decided differently can still be held accountable for a decision they did not, in any meaningful sense, shape.
The governance record is complete. The question of where authority actually resided goes unasked.
This is not unique to AI.
Institutions have always preserved the form of governance long after the operational reality beneath it changed.
Symbols outlast systems.
Language outlasts architecture.
Assumptions outlast evidence.
AI has accelerated that process faster than many governance models were designed to adapt.
Perhaps the most important question in AI governance is no longer:
Is there a human in the loop?
Perhaps it is this.
Could the human in the loop still have meaningfully changed the decision?
If the honest answer is uncertain, another question quietly follows.
What was the loop actually designed to govern?
Question for the Field
Think about the last consequential AI supported decision in your organisation.
Not whether a human approved it.
Whether that human could realistically have reached a different conclusion, with the information they had, within the system they inherited, and under the conditions in which the decision was made.
If the answer is uncertain, the more interesting question is no longer whether a human remained in the loop.
It is whether meaningful human authority did.