A central explanandum of consciousness is its unity (Hurley, Reference Hurley1998). An individual may hold multiple contradictory beliefs, and a room may hold multiple conscious individuals, but one individual cannot experience multiple different perspectives at a time. What is it about consciousness that forces the disparate information currents streaming through our sensory apparatus into this single coherent take on the world?

Each theory has its own answer, from the higher-order processing of lower-level states to competition between informational subunits for global broadcast. To evaluate consciousness in non-neural candidates, like AI, these criteria are typically specified in terms of medium-independent computation. Formalized as such, all face the same interconnected problems: 1) gerrymandering, an observer may arbitrarily carve up any physical system so that it satisfies the criteria (Shevlin, Reference Shevlin2021) and 2) indeterminacy, an observer is free to carve up the same physical system in different ways, some that will meet said criteria and some that will not (Papayannopoulos et al., Reference Papayannopoulos, Fresco and Shagrir2022).

On pain of regress, the unity of one conscious self cannot be accounted for in terms of how things appear to another, for that second perceiver’s own experiential unity equally demands explanation. We need, as Kant argues, an objective basis for the unity of an experiencing subject. This issue of objective unity is a central problematic throughout Kant’s philosophy and arises also in his introduction of the term “self-organization” to describe how the parts of an organism are unified by mutual interdependence with the whole toward the intrinsic purpose of continued self-production (Kant, Reference Kant2013/1790).

Oddly, Kant did not connect these unities. If, however, metabolic unity underpins the unity of a conscious self, then that would entail biological naturalism, for the complex artifacts that trouble distinctions between the animate and inanimate – for Kant, pocket watches; for us, computers – are not self-producing. These appear as unified objects within our experience, but said unity depends on how they achieve our purposes – such as telling the time or generating undergraduate essays. Beyond our perspective, there’s just excitations of quantum fields with only varying degrees of interaction strength to connect them. Different perceivers might divide things differently – one organism’s circuit board is another (micro)organism’s habitat.

The reason there’s nothing it is like to be a pocket watch or an AI is because there is nothing it is to be a pocket watch or an AI, independent of how these objects are constructed by us as subjects. The cell, however, is bound together by more than our perception of it. It is unified by constructive interdependencies: between membranes, which create electrochemical gradients, which drive enzymatically aided reactions, which regenerate these membranes and enzymes in turn. If one part of this network breaks down, then the whole thing disintegrates. This interdependence is not relative to our purposes, but to the intrinsic instability of biological substrates (Thompson, Reference Thompson2010).

How does this metabolic unity extend through the multicellular activity of the brain? For Seth, the answer lies with free energy minimization, which Friston and collaborators present as an account of the contents of consciousness, the neural dynamics underpinning it, and biological self-organization – under the integrative formalism of predictive inference (Friston, Reference Friston2013; Hohwy & Seth, Reference Hohwy and Seth2020).

This seems like the account of experiential and organic unity we were looking for. Yet there’s a problem for biological naturalism here. Contra earlier presentations, both critics (Nave, Reference Nave2025; Di Paolo et al., Reference Di Paolo, Thompson and Beer2022) and advocates (Ramstead et al., Reference Ramstead, Sakthivadivel, Heins, Koudahl, Millidge, Da Costa, Klein and Friston2023) now argue that the FEP is not a unique principle of life but a general description of every “thing.” This is because it reduces self-organization to the emergence of invariant statistical properties and defines survival as the preservation of these regularities. That would entail that, when encountering an “unlikely” state, an organism must return to a more “likely” one – a process that can be described as minimizing informational “free energy.” Still, insofar as any physical object is defined by the invariance of certain features and is subject to environmental perturbations, so any object must do the same (Ashby, Reference Ashby and Klir1991).

Crucially, the FEP advances no ruling on which properties of the system are essential to its ongoing unity. All it allows is that if we’ve already identified these features, then we can re-describe their preservation in statistical terms. Without prior constraints on system individuation, this process is as vulnerable to gerrymandering and indeterminacy as any formal redescription.

What if, as Seth does, we focus on informational free energy only as it relates to the thermodynamic quantity, namely, energy that is available for constrained release to power physical work? It’s true that metabolic self-production involves reducing this by channeling the concentrated energy of high-frequency photons or food into the construction of membranes, enzymes, and other cellular structures. The faster this process, the faster thermodynamic free energy is lost via the inefficiency of heat dissipation (Martyushev & Seleznev, Reference Martyushev and Seleznev2006). To minimize free energy entirely, however, is to die.

Thermodynamic equilibrium is the end, but not the point, of life. What makes a dissipative process a living one is not the universal tendency toward free energy reduction but rather the contingent specifics of what happens along the way – namely, whether that dissipation powers the production of an organized structure that can continue to channel this dissipation into the work of its own construction (Montévil & Mossio, Reference Montévil and Mossio2015).

Viewing the brain thermodynamically means we cannot just treat it as a predictive control mechanism. Neurons may be long-lived, but, like all cells, they churn through components with turnover times in the order of minutes (Meagher, Reference Meagher2014). Ignoring this and taking their seemingly invariant structure for granted leads us to misinterpret phenomena like spontaneous activity as mere “noise” – an unfortunate biological residue interfering with essential cognitive dynamics that could be better implemented in an artificial machine. In contrast, viewing brains as precarious self-constructing networks where, as Seth suggests, we cannot separate what they do from what they are, frames this intrinsic instability as essential in holding our metabolic and experiential selves together.

How substrate-specific is this unity requirement? That’s an open question, subject to investigation of whether alternative chemistries can satisfy the precise synchronization of exergonic and endergonic reactions that powers life (Grefenstette et al., Reference Grefenstette, Chou, Colón-Santos, Fisher, Mierzejewski, Nural, Sinhadc, Vidaurri, Vincent and Weng2024). But for all that synthetic or alien organisms may one day exhibit metabolic closure in alternative substrates, silicon semiconductors ain’t got the stuff.