What makes a possibility actionable versus merely metaphysically possible?
This explores a recurring move in the corpus: drawing a line between questions that can stay philosophically open (is AI conscious? does it really 'think'?) and the design, policy, and engineering choices that don't have to wait for those questions to resolve.
This reads the question as one about leverage, not ontology: a possibility is actionable when you can build, decide, or intervene on it regardless of whether the deeper metaphysical claim ever gets settled. The clearest statement of this in the corpus is the argument that the moral-status question is methodologically independent of the consciousness question Do we need to solve consciousness to address AI harms?. Harms from people treating AI as conscious happen whether or not it actually is — so the actionable thing (design and policy that account for those harms) is decoupled from the metaphysically open thing (whether the system has inner experience). The metaphysical possibility stays genuinely unresolved; the practical work proceeds anyway. That decoupling is the whole game.
The same shape shows up in how the corpus handles attributing minds to LLMs. 'Modest inflationism' deliberately ascribes only metaphysically undemanding states — beliefs, desires — while withholding the loaded claim of consciousness Can we defend modest mental attributions to large language models?. The point isn't that consciousness is impossible; it's that you can get useful, defensible traction from the lightweight attributions without having to win the heavyweight fight. Compare that to the cases the corpus treats as *not* actionable through tweaking: linguistic agency is argued to require embodiment, participation, and precariousness that current architectures structurally lack — a categorical gap, not a dial you can turn What makes linguistic agency impossible for language models?. And computation is argued to presuppose a conscious 'mapmaker' that no amount of added algorithmic complexity can generate Can computation arise without a conscious mapmaker?. Those are framed as possibilities you can't engineer your way into — interesting metaphysically, inert practically.
There's also a more operational sense of 'actionable' that the corpus surfaces: a possibility becomes actionable when it attaches to a concrete trigger or a thing you can inspect and run. An agent shouldn't deliberate over every abstractly-possible action — deliberation gets triggered only at steps where sampled actions actually diverge, i.e. where uncertainty is real rather than notional When should an agent actually stop and deliberate?. Code earns its place as an agent's reasoning substrate precisely because it's executable, inspectable, and stateful — possibilities expressed as code can be run and verified, not just entertained Can code become the operational substrate for agent reasoning?. The common thread: actionability requires a handle — a trigger condition, an execution, a measurable harm — that mere logical possibility lacks.
What ties these together is something a curious reader might not expect: the corpus keeps converting hard metaphysical questions into tractable empirical or design ones, not by answering the metaphysics but by routing around it. And there's a flip side worth noticing — the gulf-of-envisioning work shows that even human intent starts out merely 'possible' and only becomes actionable through interaction that turns open-ended wishing into constrained options you can evaluate Why can't users articulate what they want from AI?. So actionability isn't a property a possibility has on its own; it's something a possibility acquires when it's given a handle — a trigger, an execution, a harm, or a structured choice. The metaphysically-possible-but-inert cases are exactly the ones where no such handle exists.
Sources 7 notes
Research shows that harms from user behavior treating AI as conscious occur regardless of whether AI actually is conscious. This decouples metaphysical debates from practical design and policy work.
Both robustness and etiological deflationist arguments beg the question against inflationism. A graded approach ascribing metaphysically undemanding states like beliefs and desires—while withholding consciousness claims—mirrors how we treat non-human animals.
Enactive cognitive science identifies three constitutive properties of linguistic agency—embodiment, participation, and precariousness—that are structurally absent from LLMs. This is a categorical incompatibility, not a matter of degree, suggesting current architectures cannot achieve genuine linguistic agency.
Computational systems depend on a conscious mapmaker who alphabetizes continuous physics into discrete symbols. No increase in algorithmic complexity can generate this agent; it must logically precede the computation it makes possible.
SAND uses self-consistency sampling to flag uncertainty: if N policy samples all match the expert action, skip deliberation; if they diverge, trigger execution-guided critiques. This step-level compute allocation lets agents deliberate only at genuinely uncertain decision points.
Research shows code uniquely enables agents to externalize reasoning, execute policies, model environments, and verify progress through its simultaneous executability, inspectability, and statefulness across task steps.
Intent develops through interaction, not in isolation. Since AI models respond rather than probe, they miss opportunities to help users discover unarticulated requirements. Structured dialogue that presents model-generated options shifts the cognitive burden from open-ended envisioning to constrained evaluation.