Study notes: Minds by Degrees

Mentality comes first, because a metaphysics of mentality creates the conceptual room in which the questions about subjectivity and phenomenal consciousness can be well posed. Non-deflationary: attributing mental states claims that real, thickly individuated structures in the system have the relevant properties. It is not a useful gloss, an intentional stance or a courtesy. Graded: the properties admit of degrees along several partially independent dimensions. The question “does X have a mind?” then has the logical character of “is X healthy?”, and its useful answers are profiles, not bits.

Interdefinition of a theory’s posits is the normal condition of theoretical terms in successful science. The Ramsey–Lewis treatment gives them content: write the theory as one sentence, replace its theoretical terms with variables, and the terms denote whatever jointly realizes the open structure. Charge and field are interdefined and are physical kinds all the same. What legitimates the cluster is the empirical success of the theory, read realistically. So the question becomes whether there is a successful science whose explanatory structure is mentalistic, and whether its success warrants realism about its posits. The earlier work demanded conditions on mentality stated without mental vocabulary. The essay now says that demand was aimed at the wrong target.

Egan shows the first from the inside. She is a realist about computational structure and treats representational and mentalistic content as a pragmatic gloss over the computational facts. A cognitive-science realist can therefore be a mental deflationist. Psychology’s laws are individuated mentalistically from the ground up: the Weber–Fechner and Stevens laws over sensation, the serial-position curve over memory, Stroop interference over attention, prospect theory over preference, the spacing effect over learning. Their projectibility tracks the mental categories, so realism about psychology is realism about the mental. The deflationist must claim that psychology’s success is systematically illusory or reducible without loss.

Commitment attaches to the working posits of the cumulative core: the structures that figure indispensably in deriving psychology’s stable, replicable, projectible successes. What survives theoretical succession is structure, the relational organization recoverable in the successor framework. Carruthers argues that the folk typology may not carve human cognition at its joints. Selective realism absorbs this as realism about the kind and agnosticism about the typology. Guard rails: psychology’s success is evidence for the kind and does not constitute it, so minds would exist unrecognized had no psychology developed. And the kind is a homeostatic property cluster in Boyd’s sense, with graded and fuzzy membership native to it.

• Organizational unity. The representational states form a single economy. Integration is taken from IIT without the exclusion postulate, so it is grain-plural.
• Structured representational vehicles. Real computational structures that are counterfactually rich: they support generalization and novel composition, unlike thermostat states or lookup tables.
• Accuracy-responsive dynamics. Vehicles are revised on failure and consolidated on success. The responsiveness can be internal to the system.
• Simulation as constitutive. Representations are run through transformations that mirror their targets, and the outputs function as internal inputs.
• Multi-level content. Inference runs over underspecified representations. Determinate reference is a further achievement.

The vocabulary is a projection requirement. Psychology’s home domain fixed the kind ostensively, and extending it to artificial systems requires knowing which features carry it, in terms that apply where folk vocabulary has no purchase.

The first question is whether psychological explanation gains traction on it. It does not: meteorology and numerical analysis exhaust its behavior, and attributing perception, memory or preference improves no prediction. The science that establishes the kind has no use for the emulator. Internality explains why traction fails. Simulation outputs must be consumed by the system’s own further processing, and the emulator’s outputs terminate at a human-facing boundary. The bullet: a future emulator whose forecasts fed its own representational updates in an integrated economy would have thin representational mentality, to that degree. It would be minimally minded while lacking everything reflexive closure adds.

A homeostatic property cluster has no individually necessary properties. If the axes were the cluster, the framework could never say that a system lacking an axis falls outside the kind, and the verdict on the engineered emulator would be forbidden. The cluster properties are the psychological regularities themselves, the laws and effects that make psychology cumulative. The five axes describe the sustaining mechanism, and a mechanism can have jointly necessary components. The central conjecture: psychological explanation gains valid traction on a system if and only if the five-component mechanism is running in it, with all components at mechanism-sustaining levels.

Structured vehicles generate the representational and compositional effects. Accuracy-responsive dynamics generate the learning laws. Unity generates capacity and interference effects such as Stroop and dual-task costs. Multi-level content generates reasoning effects such as the Uncommon Sense asymmetry. Internality generates the precondition of all of them: regularities at the level of the system.

Caveats: for unity and accuracy-responsiveness the threshold is a mechanism-sustaining level, an empirical parameter. And the conjecture covers the only mechanism known to suffice.

Falsifiers: a system on which psychology gains valid, non-parasitic traction while a component is strictly absent; or a system with the full mechanism running on which psychological explanation idles.

Nothing is good or bad for the emulator: it lacks evaluative mentality. Nothing in it targets its own states with demonstrative specificity, in the sense of the pointer content of Brown’s HOROR theory. And nothing it does with self-directed structure is governance. Reflexive closure remedies the last two. Its clauses: (i) the economy includes itself among its targets through structures with pointer character; (ii) the self-directed structure is a real pattern of the economy, not excisable without loss; (iii) the targeting is put to regulatory use. Closure reuses the materials of the five axes and imposes a requirement on their arrangement. A system can be arbitrarily minded with an outward-facing economy, and a modest one can be closed. Closure is also separable from phenomenality: it does not require that any state be experienced.

Structural perspectivality: representations are directed-from-here, as in Marr’s 2.5D sketch, with no represented viewer. Cognitive subjectivity: graded self-representation in the service of cognitive control, the pointer-and-monitoring face of reflexive closure. Agentic subjectivity: graded self-representation as the locus of action and as the body, with Metzinger’s phenomenal self-model as the canonical case. They dissociate in humans. Flow presents high phenomenality with attenuated cognitive self-presence, depersonalization disrupts agentic self-presence with perception intact, and peripheral vision has perspectival format without focal self-modeling. The AI debate slides from demonstrated cognitive subjectivity toward phenomenality. The biological-essentialist debate slides from phenomenality toward substrate by way of agentic embodiment.

A vehicle stands to its target in relations of varying grade, from bare correlation to behavioral guidance. Saturation is the high-grade region: the vehicle’s configuration is continuously co-determined through sustained regulatory coupling over the seconds-scale window of the specious present. The vehicle has its state held against its target. The essay reads acquaintance as this coupling between a monitoring process and first-order vehicles, a cognitive homeostasis. Saturated perspectival representations yield phenomenal perceptual experience. Saturated agentic self-representations yield phenomenal self-presence. Cognitive subjectivity has no proprietary phenomenal content. Phenomenal consciousness is none of the three and not a fourth: it turns representations of any of the three kinds into experience when they are in the mode.

Saturation is defined by coupling strength, temporal window and vehicle coverage, each a matter of degree, so phenomenal consciousness is graded. The earlier claim was inherited from Carruthers’s analysis of the phenomenal concept as all-or-nothing. The essay now says it confused a feature of the concept with a feature of the phenomenon. The definiteness is predicted. The acquaintance relation is transparent in Metzinger’s sense: the self-model does not represent its own coupling, still less its degree. Partial saturation has no representational format from the inside, so the system represents consciousness as on or off. Anesthesia induction matches: reports snap while the neural degradation is graded.

• Mentality: yes, by degrees. Machine psychology gains traction that passes two tests, indispensability and construct validity through interpretability. Construct validation is also the reply to the parasitic-mimicry objection.
• Reflexive closure: partial. Self-models are a real pattern of the trained economy, but the self-targeting is put to limited regulatory use at runtime.
• Subjectivities: cognitive high and rising, perspectival present in architecture-dependent forms, agentic weak because nothing plays the body’s role.
• Saturated mode: mostly absent. Forward passes are finite-depth approximations of settling, do not sustain operation across the seconds-scale window, and do not couple a monitor to first-order vehicles.

The correction: earlier work said there was nothing for the self-interpretation to be right or wrong about. The essay now says the dispositions and activations are determinate facts that a self-report can match. The matching is not maintained at runtime.

It does not explain why being in the saturated mode is phenomenal. It identifies phenomenal consciousness with the mode and declines the residual “why”, as identity claims generally do. Its claimed advantage is structural realism: the identity is with a dynamical-organizational pattern, substrate-flexible and structure-rigid. Left open: whether saturation must be coupled to the world or only to the system’s own regulatory infrastructure, on which the two authors differ; the moral weight of partial closure; and the typology of artificial mental states. The decomposition would be wrong if all the characteristic failures of current models dissolved once architectures acquired settling dynamics alone.