Attending to Subject vs. Output

In the AI Story

The distinction is easier to state than to practice. When a person uses Claude to produce a report, the surface activity is the same in both modes: reading, thinking, revising. The difference is internal — the direction of attention, the reference point against which judgments are made. The person attending to the subject asks 'Does this correspond to what I perceive when I look directly at the reality?' The person attending to the output asks 'Does this read well? Pass muster? Satisfy what I requested?'

The ego consistently prefers the second mode, because it is less demanding. Evaluating a paragraph that Claude has produced is less demanding than producing the paragraph oneself, because evaluation can rely on surface criteria while production requires the deep engagement with the subject that Murdoch calls attention. The ego, always seeking the path of least resistance, will always prefer evaluation to production — and the AI system will always provide the smooth evaluable surface.

What makes the distinction practically important is that the person often does not know which mode she is in. She can produce competent output in the second mode while believing she is in the first. She can feel confident about a report while never having genuinely attended to the subject the report addresses. The confusion is structural, not psychological — it arises from the fact that both modes produce similar-looking artifacts, and the difference is visible only from a perspective the person has to cultivate deliberately.

The diagnostic Murdoch's framework provides is the quality of surprise. Attention to the subject produces genuine surprise — the discovery that reality is different from what one expected, that the problem is harder, stranger, more resistant than one thought. Attention to the output produces only shallow surprise — unexpected phrasings, novel-seeming connections, outputs that differ from anticipation. The first kind of surprise transforms the person's understanding; the second kind merely varies the material she is consuming. If a long period of AI-assisted work has produced only the second kind of surprise, the person has probably not been doing the first kind of work.

Origin

The distinction is Murdoch's, but its sharp articulation for the AI context is the central contribution of her simulated engagement with The Orange Pill. The philosophical ground is the distinction between fantasy and imagination, between consolation-seeking and reality-seeking, that runs through her work from the 1950s forward.

Key Ideas

Same activity, different orientation. Reading, thinking, and revising can be performed in either mode; the difference is internal and not directly observable.

Surface criteria vs. reality correspondence. Evaluation of output uses surface criteria; attention to subject tests correspondence with directly perceived reality.

The ego prefers output. Output mode is less demanding and produces similar-looking artifacts; the ego will default to it unless resistance is deliberately cultivated.

Quality of surprise as diagnostic. Genuine attention produces deep surprise (reality was different than expected); output evaluation produces only shallow surprise (the machine's phrasings varied).

Debates & Critiques

Whether the distinction is sharp enough to guide practice, or whether most AI-assisted work involves oscillation between modes, is an open question. Empirical work on cognitive work with AI tools is still young; Murdoch's framework suggests that whatever the distribution, the dominant mode matters, and deliberately cultivating subject-attention has outsize effects on long-term capability and moral formation.