The Grounding of Normativity

Where Value Enters

The classical formulation holds that normative conclusions cannot be derived from purely descriptive premises. The natural move for a framework like this one is to attack the premise — to argue that physics was never value-neutral, so the gap was illusory from the start. That move is available and it is wrong, and it is worth declining explicitly, because the version of the argument that survives is stronger.

Here is the tempting version. Physics is probabilistic; thermodynamic "laws" are statistical, and individual trajectories can violate them. So physics describes biases — which outcomes are likelier — rather than necessities, and a bias is an asymmetry: $p_\sigma(\text{outcome}_1) \neq p_\sigma(\text{outcome}_2)$. Call that a proto-preference and normativity appears to reach all the way down. But it does not. An asymmetry in likelihood is not an asymmetry in preference, and the prefix does no work: a gas is not faintly in favor of expanding. Free energy gradients bias configurations without anything being better for anything. The universe is not slightly opinionated.

Value enters later, and in two distinguishable steps. The first is viability. Once a system maintains a boundary, some trajectories preserve it and others destroy it, and that is an objective fact about the system rather than a projection onto it — better and worse defined relative to persistence, with no observer required and nothing yet felt. The second is valence, where a system's own regulation runs on that gradient: states are used as better or worse by the thing they are better or worse for. The relevant hierarchy is therefore not a smooth ramp from mildly-preferential physics up to ethics. It has a floor. Below viability there are asymmetries and no value; at viability there is agent-relative better and worse without feeling; at valence there is mattering. That the floor exists is what makes the rest of this section a claim rather than a definition.

Viability and Agent-Relative Value

Made precise: the viability boundary defines conditions for persistence, and so implicitly a proto-value function $V_{\text{proto}}(\mathbf{s}) = -d(\mathbf{s}, \partial\viable)$. "Proto" here marks something specific and defensible — an objective better-and-worse relative to persistence, with no claim that anything is being preferred.

Note carefully what this does not claim. It does not derive obligation from persistence — that would be circular. Normativity enters at the next step: when the system develops a self-model and acquires valence (gradient direction on the viability landscape), it cares about its viability — caring is what valence is. A viability gradient felt from inside cannot fail to matter to that system. The “why should it care?” question is confused: a system with valence already cares; the valence is the caring. But notice what is delivered, and what is not. Delivered: agent-relative disvalue — this state is dispreferred by this system, from inside its own perspective. Not delivered: anything agent-neutral — any sense in which the state is bad full stop, such that some other agent is thereby given a reason to act.

Valence as Real Structure

When the system develops a self-model, valence emerges—not projected onto neutral stuff but as the structural signature of gradient direction on the viability landscape:

Suffering is not neutral stuff that gets called bad by convention. Suffering is the structural signature of a self-maintaining system being pushed toward dissolution. The badness is constitutive, not added.

The correspondence is nevertheless imperfect, and the framework should not pretend otherwise. Evolved or learned proxies misfire: an immediately rewarding trajectory may damage long-term viability, while a painful intervention may preserve it. Valence is best treated as an internal control signal shaped by viability, not as the viability gradient itself.

One firewall must be stated explicitly here, because the result invites exactly the wrong inference. Integration-as-capacity, forged by survived stress, is categorically distinct from any normative endorsement of the stress. Capacity-from-survival is not endorsement-of-suffering. That a furnace forges is a claim about mechanism; it licenses nothing about what anyone should be put through.

Where the Is-Ought Gap Migrates

Let $D_{\text{exp}}$ be the facts at the experiential scale, valence included. Agent-relative conclusions about approach/avoidance then follow directly: that this state is good-for-S or bad-for-S is settled, for S, by the structure of S’s valence. The classical gap between describing and dispreferring closes here — but only first-personally.

Part of what made the classical gap look unbridgeable was a framing error, though not the one just declined. It is that the argument is usually conducted between the bottom of the hierarchy, where physics looks value-neutral because it is, and the top, where culture is explicitly normative — with the viability layer in between left out entirely. Put viability back and the question changes from "how do you get an ought from an is?" to "what kind of is is a viability gradient?", which has an answer. The error has a perceptual dimension too. The problem was formulated by thinkers perceiving the world with low ascription and low coupling — the mechanistic stance that factorizes fact from value, perception from affect, description from evaluation. Raise both and the separation loses its grip: perceiving something as alive already includes registering its flourishing or suffering as mattering. The participatory perceiver never meets two realms needing a bridge, because that stance never split them. Note carefully that this is a fact about perceptual configuration and not about reality. The viability gradient is there regardless of how anyone perceives, and so is the floor beneath it.

None of this delivers what the framework's later ethics needs. The perceptual stance closes the felt gap between fact and value for the perceiver; the valence argument closes the gap between fact and disvalue for the system that has the valence. Both are agent-relative, and the downstream ethics — that a creature's suffering obligates others — requires agent-neutral disvalue. Nothing in "felt as mattering from inside" bridges bad-for-X to bad-period-and-you-ought-to-act. So the gap does not vanish. It migrates, from description-versus-prescription to first-personal disvalue against third-personal obligation, and this relocated gap the framework does not close. It can show that suffering is real and matters to the sufferer. It cannot show, from these premises, that it must matter to you. The honest claim is not that the gap is gone but that we have found where it survives.

One thing this part deliberately does not attempt: quantifying the worth of persons, organisms, or models. Capability, usefulness, experience, and moral status are different variables, and nothing above establishes how to weigh competing interests. That question is taken up in Part II, where it can be argued on its own terms rather than inheriting unearned authority from the physics.

One coda. If attention steers the trajectory and values guide attention, then values are not epiphenomenal commentary on a value-free process but causal participants at a point of leverage. Not that wishing makes it so — the a priori distribution is still physics. But the effective distribution is physics times measurement, and the measurement distribution is shaped by values. In that compatibilist sense "ought" is not a separate domain from "is": it is a component of the mechanism biasing which "is" the system ends up in.