Biology is a pragmatic discipline. Like Justice Potter Stewart, who knew it when he saw it, most biologists eschew precise definitions of life, or species, or consciousness. In each case, we are dealing with multiple contributing factors and continua across arbitrary boundaries. From this pragmatic point of view, most biologists would agree that (i) consciousness evolved by natural selection; (ii) feelings are a central facet of consciousness so they must reflect a physical state that selection could act upon; (iii) feelings must offer some benefit, i.e., a selectable advantage to organisms that would be lacking in the absence of the feeling; and (iv) feelings are about the state of an organism in relation to its environment.

Rather than questioning the philosophical veracity of each of these statements, I want to take them seriously on their own terms. To reframe: feelings are physical states that reflect how well an organism is doing and are tangibly different from any other form of feedback. There is nothing here that speaks to panpsychism (Koch, Reference Koch2021); feelings are strictly a property of living systems. Nor does this speak to any need for a neural network. Some neural networks are accompanied by feelings, others are not (Hohwy & Seth, Reference Hohwy and Seth2020). Are neural nets the wrong level of analysis, with the simplest forms of consciousness being widespread across living things, including bacteria (Reber et al., Reference Reber, Miller, Slijepcevic and Baluška2023)? Anaesthesia of single-celled organisms implies a basic mechanism (Turin et al., Reference Turin, Skoulakis and Horsfield2014). If so, what kind of physical state do feelings reflect?

The only distinction between a living cell that responds to its environment and a cell that has just died is metabolism. This word conceals an unceasing storm of reactions: a billion per second, second after second, in a single bacterial cell (Lane, Reference Lane2022). How do bacteria coordinate these reactions, enabling coherent behaviour? Through a handful of thermodynamic drivers such as NADH (reduced nicotinamide adenine dinucleotide) and ATP (adenosine triphosphate), which coerce reactions into equivalent far-from-equilibrium states, reflecting the availability of food and energy (Lane, Reference Lane2022). But there is a problem. The more NADH or ATP in a cell, the more robust but less sensitive it will be to changes in the environment.

Automated feedback loops, as in chemotaxis, do not help much. They are laboriously statistical and may take minutes to gauge large pool sizes of ATP or NADH (Lane, Reference Lane, Martin and Muller2006). Multiple feedback loops can conflict with each other (Hamadeh et al., Reference Hamadeh, Roberts, August, McSharry, Maini, Armitage and Papachristodoulou2011). What should a cell do if one system reports just enough oxygen, another says there is plenty of food, a third registers a metabolic toxin, and a fourth implies it’s a bit too hot? The cell’s decision must depend in part on its own state: how easily can it deal with these challenges? If a feeling is about an organism’s state in relation to its environment, how does a cell know what state it is in, or what it has in reserve, to make a better-than-random decision? If it moves somewhere else, how does it know it is better off now? It needs a real-time, integrated readout of its overall metabolic state, bounded by its own homeostatic limits. Luckily, it has one – fluctuations in membrane potential (Lane & Rodriguez, Reference Lane and Rodriguez2025).

All cells rely on an electrical membrane potential to drive work, notably ATP synthesis. Our own cells transduce energy in mitochondria, once free-living bacteria, so membrane bioenergetics is internalised. Bacteria energise their plasma membrane, the boundary (or ‘Markov blanket’) separating internal from external states (Palacio et al., Reference Palacios, Razi, Parr, Kirchhoff and Friston2020). The environment (e.g., food and oxygen) drives catabolism, which generates membrane potential, which drives anabolism and costly behaviour, notably motility. This means that membrane potential reflects the overall metabolic demands of the cell in relation to its environment – how well it is doing. A healthy cell has a modestly high membrane potential, maintained by the continuous transfer of electrons to oxygen. Deviations from this homeostatic optimum reflect perturbations in metabolism. Membrane potential is an electrostatic field, while the movement of charge generates electromagnetic fields; the balance between these forces could in principle indicate whether the cause of perturbation is primarily environmental (e.g., oxygen availability) or cellular (damage or mutations) (Lane & Rodriguez, Reference Lane and Rodriguez2025). Challenges such as viral infection are literally short-circuited by collapsing membrane potential, where bacteria commit suicide for even loosely related kin (Refardt et al., Reference Refardt, Bergmiller and Kümmerli2013).

In other words, membrane potential provides a real-time, systems-level output that integrates the cell’s metabolic state – how well it is doing right now, in relation to its environment, genetic resources, and life history. Feelings are electromagnetic fields generated by membrane potential – a physical state (Lane & Rodriguez, Reference Lane and Rodriguez2025). That’s not to say all fields correspond to feelings. A cell has a specific topology, and charge moving in relation to that topology holds meaning for the cell as an entity – a self. Fields with the wrong orientation, strength, or oscillation frequency hold no meaning for the cell or for selection over generations. Fields across metals or through space generally lack meaning in relation to a living metabolic entity.

Feelings as fields confer selectable advantage: they can be acted on, for example, by moving, before other systems-level changes such as metabolic regulation or gene expression obscure the current state. If a cell is in trouble and curbs the danger by changing gene expression, its position may be improved, but it no longer knows the magnitude of risk or its capacity to cope. Feelings are then a physical early-warning system, which alerts organisms to metabolic issues, enabling a rapid response: a selectable advantage that is lacking in the absence of the feeling. That makes feelings meaningful, physical, evolvable, and tangibly different.

The use of electromagnetic fields to signal metabolic health has obviously been elaborated in the evolution of complex animals, but fields are unequivocally involved in how the brain works (McFadden, Reference McFadden2023). We know little about their meaning. But Seth is right: consciousness is not fundamentally about computation. We feel the goodness (or badness) of metabolism, whirling in water at billions of reactions a second. We share that with bacteria, but not with Grok.