The Arrow and the Archer: Why a Computational System Cannot Apprehend an End as Good

A computational system can be parametrised by an objective, but it cannot have a purpose. To have a purpose is to apprehend an end as good, and apprehension, on the account defended here, is an operation that only a thing with a nature of its own can perform. An artefact has no nature of its own. It is an arrangement imposed on materials that keep their own natures, and nothing in it is a subject that could grasp anything. I take immanent causation, the activity that arises within a thing and serves that same thing's own maintenance and completion, as the mark that distinguishes a substance from an artefact, and argue that not even a substance produced by engineering could be made to understand. From this one lack, two consequences follow. The first is practical: the machine cannot hold any end as good; its objective is a parameter, answerable to nothing, about which it cannot be wrong. The second is theoretical: no finite physical process means one content rather than its rivals. These results recast the debate over advanced artificial intelligence. The machine is not a rival of our kind, and the danger is relocated rather than removed, since what remains is an unminded objective pursued without limit. The independence of capability from objective is explained rather than asserted, since an apprehended end constrains where a parameter does not. And alignment becomes the work of keeping a human principal in the evaluative office the machine cannot occupy. The result is not a rival mind but an anti-angel: process in the shape of reason, with nothing that understands.