What Counts as Failure Now?

Failure used to be easy to recognise.

A system failed when it broke down, contradicted itself, produced noise instead of signal, or deviated from its intended function. Failure meant error, malfunction, or lack.

Much of modern theory — scientific, technical, organisational — is still oriented around this understanding. We look for breakdowns. We diagnose faults. We repair what does not work.

But the systems that increasingly shape our lives do not fail in these ways.

When systems fail by succeeding

Consider the cases that have occupied us.

Escher’s constructions do not collapse. They obey their local rules perfectly. Each step follows lawfully from the last. The impossibility emerges only when we try to compose these lawful moves into a single world.

Cosmological models do not disintegrate under scrutiny. Their equations hold. Their predictions are precise. The problem appears only when we demand a globally closed picture of the universe.

AI systems do not fail by incoherence. On the contrary, their fluency, consistency, and responsiveness are what make their limits visible. The more capable they become, the more sharply their non-integrability shows.

In each case, failure is not the result of error.

It is the result of success without worldhood.

The inadequacy of older criteria

Traditional criteria of failure presuppose a world that already holds together.

They assume that:

• components belong to a single coherent whole,

• optimisation at one point supports stability at others,

• improvement accumulates rather than destabilises.

When these assumptions fail, the criteria misfire. We keep looking for mistakes because our concept of failure cannot register structural limits.

What we encounter instead are anomalies that refuse to behave like faults.

Failure as non-inhabitability

A different criterion begins to suggest itself.

A system fails, not when it breaks its rules, but when it cannot be inhabited as a world.

This does not mean that the system is useless, deceptive, or meaningless. It means that the coherence it provides is local rather than global. It supports action, inference, or movement within frames that cannot be integrated into a single perspective.

Such systems are not defective. They are bounded.

The failure arises only when we treat those bounds as temporary obstacles rather than constitutive limits.

Why this matters now

Much contemporary frustration arises from the attempt to force worldhood where none is available.

We demand closure, and when it does not appear, we infer hidden forces, missing values, or moral deficits. We respond by adding layers of explanation, repair, or governance.

The result is not resolution, but proliferation: more fixes, more artefacts, more confusion.

Recognising non-inhabitability as a form of failure allows us to stop at the right point.

Failure without drama

This reconstrual also removes a great deal of unnecessary drama.

If failure is not breakdown, then it need not provoke panic. If limits are structural, then they are not scandals. They are features of how coherence is distributed.

Seen this way, failure becomes informational rather than catastrophic. It tells us where a system holds, and where it does not.

A changed landscape

We are entering a landscape in which many of the most powerful systems available to us work perfectly within their domains and yet do not compose into worlds.

To navigate this landscape, we need a revised sense of failure — one that can register excess precision, local success, and global impossibility without immediately reaching for repair.

What counts as failure now is not that systems do not work.

It is that they work, and we expect them to do more than they can.