Dispositional Analysis

In the AI Story

The power of dispositional analysis, for the AI question, is that it does not require settling the hard problem of consciousness. It asks only that behavior be characterized with care. When Claude produces a working prototype from a plain-English description, its behavior exhibits certain dispositional properties: responsiveness to context, flexibility across tasks, reliability within a definable range of conditions. These are genuine dispositions — real, testable, comparable across systems. Whether they are accompanied by phenomenal experience is a separate question, and its answer is irrelevant to the behavioral assessment.

Dispositions have reliability profiles. A physician can be disposed to diagnose accurately, reliably, across a wide range of cases, or disposed to diagnose accurately in common cases but catastrophically in unusual ones. The reliability of the disposition is a crucial fact about the practitioner — it determines how much independent verification her judgment requires. Claude's dispositions have a specific profile: extremely reliable for fluent prose, highly reliable for working code, notably less reliable for substantive philosophical accuracy (witness the Deleuze error), and poor at self-correction. The practical upshot: the tool is trustworthy in proportion to the match between its dispositional profile and the task.

Dispositions are also developmental — they are built through practice and atrophy through disuse. The human expert's reliability profile reflects the specific history of training, error, correction, and repetition that built her dispositions. Claude's dispositions were built through a different process (training on a corpus), and the difference shows up in what Claude can and cannot do. The senior engineer who feels that a codebase is wrong before articulating why has built that disposition through thousands of hours of implementation. The knowing how lives in the dispositional structure.

The analysis also clarifies what is at stake in productive addiction and related concerns about AI-mediated work. The worry, translated into dispositional terms, is not that a ghost departs when humans work with machines. It is that the conditions under which critical dispositions are exercised may be eliminated, causing those dispositions to weaken. Not because of metaphysics, but because dispositions require exercise. Critical judgment, like a muscle, atrophies when unused. The collaboration must be structured to preserve the conditions under which the human's evaluative dispositions are genuinely exercised, or the collaboration erodes the very capacities it depends on.

Origin

Ryle developed dispositional analysis throughout The Concept of Mind (1949), drawing on earlier work in behaviorism and ordinary language philosophy. The approach has antecedents in Wittgenstein's later philosophy and in the American pragmatist tradition, particularly Dewey and Mead, though Ryle's formulation gave it the distinctive shape that influenced subsequent cognitive science.

Key Ideas

Dispositions are real but non-episodic. Sugar's solubility is a genuine property even when sugar is not dissolving. Intelligence is a genuine property even when the intelligent person is not currently solving a problem.

Reliability profiles, not binary possession. Dispositions come in degrees and with specific conditions of reliability. The question is not whether a system possesses a disposition but how reliably it exercises the disposition under specified conditions.

Dispositions are developmental. Built through the specific history of practice that shapes them. This is why Claude's dispositional profile differs systematically from a human expert's — different training produces different dispositions.

Dispositions can atrophy. The practical concern about AI-augmented work is not metaphysical but dispositional: under what conditions do critical human dispositions remain exercised, and under what conditions do they weaken?

Debates & Critiques

The dispositional analysis has been criticized for being either too weak (failing to capture the qualitative character of mental states) or too strong (attempting to analyze too much in terms of hypothetical behavior). Frank Jackson's knowledge argument and Thomas Nagel's subjective character argument press the first critique; Donald Davidson's critique of the 'paradox of irrationality' presses a different version of the second. The Ryle volume's chapter on honest limits concedes that the framework is strongest where behavior is the question and weakest where phenomenal experience is.