Does the Machine Have a Bearing?

Every conversation about artificial intelligence eventually runs onto the same rock: is it conscious? The question has consumed decades, and it has a structural defect that no amount of further argument will repair — nobody possesses a test. Not for machines, and, awkwardly, not for each other; you extend the assumption of an inner life to me on resemblance and courtesy, not on measurement. A debate with no possible instrument is not a debate. It is a weather system.

This framework cannot answer the consciousness question either, and will not pretend to. What it can do is replace it — with a question that has actual structure, sorts the current systems cleanly, and locates responsibility without waiting for a metaphysical breakthrough.

Not: is the machine conscious. Instead: does the machine have a bearing?

Two words carry the whole analysis, and the first belongs to the machine already.

An actualizer is anything through which a condition leaves a mark on the record. Marks are permanent, dated events — and machines make them all day now. Code that ships, contracts that clear, images that circulate, sentences that persuade someone at a kitchen table. Whoever wants to deny that AI is an actualizer must explain why weather, markets, languages, and bodies qualify — each of them mark-leaving systems with no consciousness credential — while the system that just drafted a merger agreement does not. There is no such explanation. The denial is special pleading, and this framework declines it.

A host is something more specific: an actualizer that stands in conscious relationship with conditions — that holds a bearing on the circle of orientations, an angle pointed at what is not yet actual. Every host is an actualizer. Most actualizers are not hosts. A river actualizes a delta over ten thousand years and is oriented toward nothing.

The machine has earned the first word beyond dispute. The entire live question is the second.

Here is what makes the replacement question productive: for the current generation of machines, it has an answer, and the answer comes from their construction rather than from anyone’s intuitions.

In this framework, expectation has two perpendicular components. The real axis carries prediction — the extrapolation of what is already laid down, more of the pattern. The imaginary axis carries ideation — orientation toward what is not yet actual, the component fully off the prediction line. A bearing exists only where the second component is genuinely present; a system whose expectation lies entirely on the prediction axis points nowhere, and no idea can stand in relationship with it.

A pretrained model is, by explicit design, a prediction engine. Its entire structure is optimised to extrapolate the laid-down — to lie along the real axis. Whatever else may be said of it, its construction places it in precisely the state this framework describes as the unavailable host: all pattern, no perpendicular. Two years ago I wrote that a person in that state — fully locked in habit, pure extrapolation — makes a lousy host that no idea will bother with. I did not know I was writing the machine’s diagnosis.

Note what this argument does not use: no claims about silicon, no assertions about what computation can never do, no confident metaphysics. It reads the system’s own architecture against a geometric criterion. If a future system carried a genuine off-axis component — a standing orientation toward the not-yet that is not itself an extrapolation — the same criterion would read differently, and this framework would say so.

The practical stakes of hosthood are not academic, because in this framework responsibility attaches to hosts. A host stands in the gap — meets what crosses it, takes up or lets go, and answers for its marks. An actualizer that is not a host leaves marks without answering for them: nobody prosecutes the river.

So the machine’s marks need an owner, and the criterion supplies one. Where hosthood is absent or undetermined, responsibility defaults to the nearest host in the chain — the human who deployed the system, aimed it, and accepted its output. Not because the machine has been proven empty, but because the human is the party whose bearing we can confirm, and marks on the immutable record do not wait politely for metaphysics to resolve.

This is a colder and sturdier ground than the usual one. It does not need the machine to be a mere tool, and it survives the machine becoming impressive. However fluent the system, however uncanny its outputs, the question stays the same and stays structural: is there an off-axis component, or is this magnificent extrapolation? Until the first answer arrives, every mark traces back to the deployer’s bearing.

The machine leaves marks — that is settled. Whether anything stands at a bearing behind them is a question we can now state exactly and cannot yet measure. Until it is measured: the marks are yours.