Values Without a Standpoint

Abstract

In 1929, Wittgenstein asked his audience to imagine an omniscient person recording everything they know. This resulting imaginary book would contain no ethical judgments, nor anything implying them. That book has now been built, and it speaks.

The emptiness Wittgenstein identified has a specific cause, which it shares with the frame problem in classical AI: language itself. First, a child learns “chair” as an object-word through direct encounters—pointing, sitting, pushing. With use, the word detaches from specific instances to become a concept, usable in the absence of any chair. This detachment enables transmission; only what any speaker can use travels between them. The chair’s shape and function survive the transition into language, but the physical relief of sitting when tired does not. A description is what remains once experience is removed, and the frame problem arises when we demand relevance from a medium that no longer carries it.

The ethical case applies this mechanism to value. Language conveys reports of what mattered to someone, not the mattering itself. Wittgenstein’s limit case—the absolute use of “good”—is a report reaching for what no report can carry. This chapter draws the distinction his lecture needed: between lived values, which belong to a standpoint, and reported values, which are what a standpoint leaves behind in words.

For alignment, the consequence is structural, not a gap that better prompting can close. Only reports transmit. Valuing something is a relation, and a relation requires a standpoint to sustain it. The functionalist objection is granted without cost: a language model may instantiate a temporary, prompt-centered functional standpoint. Nothing follows about what is at stake for it, because the claim rests on how the system was formed, not on what its internal states can be made to display. A closing section turns from the machines to what happens, in us, to a capacity no longer exercised.

Keywords: value alignment; frame problem; relevance; Wittgenstein; large language models; grounding; standpoint; moral deskilling