The physicality of information processing systems has largely been sidelined by the standard functionalist picture. The primary philosophical upshot of Haueis and Colaço’s analysis of metabolic considerations concerns the functionalist tenets of multiple realizability (Fodor, Reference Fodor1974; Polger & Shapiro, Reference Polger and Shapiro2016; Putnam, Reference Putnam, Capitan and Merrill1967) and medium independence (Drayson, Reference Drayson2025; Garson, Reference Garson2003; Haugeland (Reference Haugeland1993 [Reference Haugeland1985]); Piccinini, Reference Piccinini2015, Reference Piccinini2021). They submit that the metabolic underpinnings of cognition do not provide decisive arguments against these positions. However, in addition to metabolism, a naturalistic account of cognition requires engaging with the physicality of the information processing vehicles. Such considerations problematize the ontological agnosticism adopted by the authors. They propel us to reject medium independence, while allowing for a degree of multiple realizability.

The formal view of computation, encoded in the medium independence thesis (see Drayson, Reference Drayson2025), holds that any physical implementation of a formal model of computation is equivalent to any other. From this perspective, the causally and explanatorily relevant features of physical computation can be abstracted from their implementation. Information, the primary resource of computation, becomes immaterial as well. This view has been captured in structural terms by positing that information is carried by the differences in the medium rather than the medium itself (e.g., Piccinini, Reference Piccinini2015).

While this idealized picture of computation is powerful, grounding various empirical achievements within computational neuroscience, for the purposes of naturalization projects, it must be supplemented with the constraints that physical implementations introduce. The extensive research on aspects of physical computation (e.g., Chirimuuta, Reference Chirimuuta2024; Copeland, Reference Copeland1996; Horsman et al., Reference Horsman, Stepney, Wagner and Kendon2014; Miłkowski, Reference Miłkowski2013, Reference Miłkowski2018; Piccinini, Reference Piccinini2015, Reference Piccinini2021; Shagrir, Reference Shagrir2022) contrasts with the relatively low interest in the physicality of information, with the notable exceptions of Collier (Reference Collier1996, Reference Collier and Sankey1999, Reference Collier2000, Reference Collier, Adriaans and van Benthem2008, Reference Collier, Dodig-Crnkovic and Burgin2011) and Deacon (Reference Deacon2007, Reference Deacon2008, Reference Deacon2011, Reference Deacon2021; Deacon & García-Valdecasas, Reference Deacon and García-Valdecasas2023; Deacon, García-Valdecasas, & Monzón-Marqués, Reference Deacon, García-Valdecasas and Monzón-Marqués2026).

For Collier, information content is specified through the Negentropy Principle of Information, which compares the actual thermodynamic entropy of a system with its theoretical maximal entropy. This is closely related to the Gibbs free energy – the ability of a (closed) system to do useful work or, as Collier interprets it, the ability to control a physical process (Collier, Reference Collier and Sankey1999; see also Rorot, Reference Rorot2025). In a similar vein, Deacon (Reference Deacon2007, p. 146) argues that “a sign/signal with any given Shannon information capacity has achieved it due to extrinsically imposed physical (thermodynamic) work,” highlighting the relational, constraining character of physical information.

Consider the following examples of systems exhibiting cognition-like behaviors, which underline the role of physicality in information processing (see also Harrison, Rorot, & Laukaityte, Reference Harrison, Rorot and Laukaityte2022; Lyon, Reference Lyon2020; Lyon et al., Reference Lyon, Keijzer, Arendt and Levin2021):

1. 1. Oil droplets, non-biological active matter systems, are studied as precursors to biotic materials. As a result of physico-chemical instability, the droplets can move in complex ways. For instance, they can navigate mazes using chemical gradients (Čejková et al., Reference Čejková, Novák, Štěpánek and Hanczyc2014; Lagzi et al., Reference Lagzi, Soh, Wesson, Browne and Grzybowski2010). This capability responds primarily to surface tension, but some droplet systems have been shown to exhibit a degree of autonomy or decoupling from the environment, including memory effects (Čejková et al., Reference Čejková, Banno, Hanczyc and Štěpánek2017; Hanczyc & Ikegami, Reference Hanczyc and Ikegami2010) and metabolism-like homeostasis (Hanczyc, Reference Hanczyc2011), on account of its energetic non-equilibrium state. The thermodynamic relationship between the system and its environment has informational character due to how the chemical gradient constrains and controls the system. These relationships can also be exploited by an external user, as the development of chemical computation platforms shows (e.g., Adamatzky, Reference Adamatzky and Meyers2009).

2. 2. Slime mould Physarum polycephalum, a unicellular amoeba, has complex navigational and decision-making capacities. It consists of a single, macroscopic, mobile cell with multiple nuclei and an intracellular cytoskeleton that scaffolds a network of veins. It exhibits oscillatory chemical signaling, which is hypothesized to underpin its ability to forage for food, as well as learning and memory (Boussard et al., Reference Boussard, Fessel, Oettmeier, Briard, Döbereiner and Dussutour2021; Sims, Reference Sims2025). As such, these oscillations are taken to enable information processing and storage, thereby serving as the focus of a functional description of P. polycephalum’s behaviour. This motivates research exploring the connection to electrical oscillations in the nervous system (Boussard et al., Reference Boussard, Fessel, Oettmeier, Briard, Döbereiner and Dussutour2021; see also Hanson, Reference Hanson2021) and the development of robotic models of slime mould navigation (Umedachi et al., Reference Umedachi, Idei, Ito and Ishiguro2013). Such models draw from the framework of morphological computation, underscoring the role of “physical form” (Paul, Reference Paul2006; see also Rorot, Reference Rorot2022) and the medium-dependent character of information processing.

These examples bolster two crucial insights regarding multiple realizability: first, if information comprises certain thermodynamic constraints between the system and its environment, only systems with a particular energetic makeup (active matter systems, see Harrison, Reference Harrison2024, Reference Harrisonforthcoming; Harrison, Rorot, & Laukaityte, Reference Harrison, Rorot and Laukaityte2022; McGivern, Reference McGivern2020) can be considered to be processing information. Second, at least for some forms of (biological) computation, their implementation relies on the physical properties of the system. Together, these features suggest non-arbitrary, medium-dependent conditions for cognitive processing, which significantly limit what types of systems can in fact realize the relevant cognitive functions. Relatedly, they might also impact our taxonomies of cognitive functions, calling for a more fine-grained classification (Chirimuuta, Reference Chirimuuta2018, Reference Chirimuuta2024; Godfrey-Smith, Reference Godfrey-Smith2016; Harrison, Rorot, & Laukaityte, Reference Harrison, Rorot and Laukaityte2022).

We maintain that the inherent physicality of information processing systems supports the view of metabolism as constitutive of (rather than merely contributing to) cognition. Pace the agnostic perspective of Haueis and Colaço, the thermodynamic underpinnings of information processing significantly limit the scope of the medium independence thesis. In addition to the general point drawn from Thagard (Reference Thagard2022, p. 84) that metabolic “energy depends on material substrates,” work in active matter physics ups the ante by indicating the requisite energetic structure for information processing. In so doing, it specifies what forms of physical media are capable of supporting cognitive functions.

Rejecting the medium independence claim but accommodating a degree of multiple realizability, our proposed perspective opens up a new viewpoint for cognitive science. For example, foregrounding the energetic characteristics of cognitive systems and their role in information processing can inform research in psychiatry (e.g., Mertens et al., Reference Mertens, Wang, Kim, Yu, Pham, Yang, Zheng, Diffenderfer, Zhang, Soltani, Eames, Schafer, Boyer, Marchetto, Nurnberger, Calabrese, Oedegaard, McCarthy, Zandi, Alda, Nievergeldt, Mi, Brennand, Kelsoe, Gage and Yao2015; see Palmer, Reference Palmer2022), enhance understanding of glial function in the nervous system (e.g., Hasel et al., Reference Hasel, Dando, Jiwaji, Baxter, Todd, Heron, Márkus, McQueen, Hampton, Torvell, Tiwari, McKay, Eraso-Pichot, Zorzano, Masgrau, Galea, Chandran, Wyllie, Simpson and Hardingham2017), among other underexplored research avenues (Rae et al., Reference Rae, Baur, Borges, Dienel, Díaz-García, Douglass, Drew, Duarte, Duran, Kann, Kristian, Lee-Liu, Lindquist, McNay, Robinson, Rothman, Rowlands, Ryan, Scafidi, Scafidi, Shuttleworth, Swanson, Uruk, Vardjan, Zorec and McKenna2024).