The Harm Is In The Trajectory

AI governance is framed as control over isolated events: approvals, audits, thresholds, guardrails. This is a reduced practice — governance as a series of operational gates applied to discrete deployments.

The deeper problem is that human-machine interaction unfolds over time. People offload cognition to systems that shape how they decide, reason and act — subtly, cumulatively, and largely unseen.

Cognitive offloading is not pathology. It is a natural feature of human adaptation and tool use, and always has been. But harmful dependency is undetectable without temporal-linking: connecting moments across time, and practising second-order observation.

Governance did not shift from a hydraulic model to an atmospheric one. It has always contained both. Practice narrowed to one of them, and it narrowed toward whichever was easier to count.

The failure is not that controls exist. It is that in narrowing to the measurable, we stopped attending to the conditions and environment of a failure — which is where failures actually form.

Governance as praxis must reintegrate both: what shapes behaviour, the agents who navigate, the long view of purpose, and the pathways our choices create over time.

The bucket is not a caricature of bad governance. It is a picture of governance that is working exactly as designed — gates operating, pipework sound, console staffed, every reading nominal.

It contains a vessel that cannot go anywhere, watched by an operator who has confused the barrier for the voyage. The star is still on the drawing, but it has faded to an outline. Purpose did not get overruled; it stopped being an input.

Most people reading this believe they are the figure on the right. The instrument that would tell them otherwise is the one this paper argues they do not have.

It is the human-machine feedback loop: the evolving relationship among human judgment, machine assistance, environmental pressure, institutional purpose, and action over time.

This distinction is not academic, and it decides whether your programme can see anything that matters. Cognitive harm rarely originates in a single interaction. A prompt may be acceptable, an output accurate, an action authorised — while the sequence gradually alters how the human reasons, verifies, trusts, delegates and decides.

Temporal-linking is therefore a requirement, not an enhancement. Only by linking interactions across time can governance observe dependency drift, changes in verification behaviour, shifts in decision authority, and the quiet loss or strengthening of human agency.

Without it, a governance function is not merely incomplete. It is structurally unable to detect its own primary failure mode, and will continue reporting health for as long as the gates keep operating.

Monitoring isolated interactions confirms that each event was compliant, safe or authorised. It says nothing about whether the practice remains healthy — and the two diverge quietly, over quarters, in one direction.

Event logs answer the wrong question. They report what happened at a point in time, not what the pattern of interactions is doing to the people who rely on it. The sequence, not the single event, is the relevant unit of analysis — and almost nothing in a modern governance stack is built to hold a sequence.

Temporal-linking connects interactions across time. It is not telemetry, and a data lake is not a substitute for it. It is the contextual linking of interactions so that governance can form associations between conditions and outcomes.

The system must learn, adapt and detect patterns across a time horizon on its own — because the humans inside the loop are, by construction, the least able to notice what is happening to them.

This is what makes governance diagnostic rather than retrospective. It produces leading indicators of risk and the ability to intervene before harm emerges, transforming governance from a ledger of transactions into an instrument of stewardship.

This is not a distinction between bad controls and good ideals. Controls remain necessary. Gates, approvals, thresholds, audit logs, risk ratings and enforcement points are all part of responsible governance. The failure begins when these instruments are mistaken for governance itself.

Reduced practice treats governance as the management of discrete objects and events. Is the system approved? Is the action authorised? Does a control exist? Did the output pass a threshold? Can a record be produced after the fact? Every question is legitimate. Every question is answerable at a point in time. That is precisely the limitation.

Your programme cannot answer any of these. Not because it was built carelessly, but because it was built to inspect artifacts, and every one of these is a property of a trajectory.

AI does not merely produce outputs. It participates in the formation of judgment — framing options, reducing effort, accelerating decisions, shaping confidence, altering verification behaviour, and gradually changing what humans notice, trust, ignore and delegate. None of that is visible in a transaction. All of it is visible in a trajectory, to anyone holding an instrument that can see one.

Without that broader discipline, governance becomes operational control masquerading as stewardship — and the masquerade is convincing from the inside, because every indicator it produces is genuinely true.

Praxis integrates theory, action, observation, reflection and correction. AI governance becomes praxis only when it observes how the human-machine relationship evolves over time, and adapts accordingly.

Praxis learns from experience, adjusts to context, and strengthens human judgment through positive and negative feedback reinforcement — while steering development and use of AI in service of a shared mission and purpose.

Control of individual gates, vessels and transactions is necessary but insufficient, because cognitive offloading and temporal effects diffuse decisive influence across time and context. Influence stops living where your controls are.

Effective governance must therefore preserve human agency, detect hidden dependency, and sustain the conditions for wise judgment. This is a shift from controlling flows to stewarding trajectories — and it is not a refinement of the current practice but a different instrument entirely.