The starting point for Seth’s biological naturalism is predictive processing. Predictive processing describes various computational and cognitive processes in the brain that can be modelled in terms of cycles of predictions and error correction at different spatio-temporal and hierarchical scales (Clark, Reference Clark2013; Hohwy, Reference Hohwy2013). However, it appears that not much is inherently biological about it: while it can be applied to describe certain biological processes, the same appears to be true for several non-biological ones too. It has been shown, in fact, that the core ideas behind predictive processing come from standard signal processing theory (Spratling, Reference Spratling2016) and estimation theory, e.g., Kalman filtering (Rao, Reference Rao1999) (see however Baltieri & Isomura, Reference Baltieri and Isomura2021), and are simply applied to describe neural and behavioural processes of different kinds. While it is hard to overstate the importance of mathematical models of brain processes, it is perhaps even harder to argue that predictive processing can explain the differences between a steam engine (Baltieri, Buckley, & Bruineberg & Bruineberg, Reference Baltieri, Buckley and Bruineberg2020) or a Braitenberg vehicle (Baltieri & Buckley, Reference Baltieri and Buckley2017) and a human brain. One could argue that Braitenberg vehicles (Shaikh & Rañó, Reference Shaikh and Rañó2020) and Watt governors (Van Gelder, Reference Van Gelder1995) have often been used to support the idea of simpler models of biological and cognitive systems and their behaviours. However, using this to vindicate the proposal of biological naturalism would be a category mistake, similar to thinking of Markov blankets as describing physical boundaries in biological systems (Bruineberg et al., Reference Bruineberg, Dołęga, Dewhurst and Baltieri2022a, Reference Bruineberg, Dołęga, Dewhurst and Baltieri2022b).

A second motivating principle behind biological naturalism is the free energy principle (Friston, Reference Friston2010; Friston et al., Reference Friston, Da Costa, Sajid, Heins, Ueltzhöffer, Pavliotis and Parr2023). According to the free energy principle, for a “thing” to exist, it must appear as if it is minimising a variational free energy functional to perform approximate Bayesian inference on the states of the external environment that produce its sensory inputs. This principle however doesn’t come without problems (Aguilera et al., Reference Aguilera, Millidge, Tschantz and Buckley2022; Biehl, Pollock, & Kanai & Kanai, Reference Biehl, Pollock and Kanai2021; Bruineberg et al., Reference Bruineberg, Dołęga, Dewhurst and Baltieri2022a, Reference Bruineberg, Dołęga, Dewhurst and Baltieri2022b; Di Paolo, Thompson, & Beer & Beer, Reference Di Paolo, Thompson and Beer2022; Raja et al., Reference Raja, Valluri, Baggs, Chemero and Anderson2021), and it is unclear once again what would make it relevant for biological naturalism. Over the years, the free energy principle has been proposed as a theory of brains first (Friston, Reference Friston2010), of life and its origins next (Friston, Reference Friston2013), and finally of “things” in more general physical terms, consistent with accounts from quantum, statistical, and classical mechanics (Friston, Reference Friston2019; Friston et al., Reference Friston, Da Costa, Sajid, Heins, Ueltzhöffer, Pavliotis and Parr2023). It is thus surprising to see this proposed as useful to formalise biological naturalism: if the free energy principle is in fact correct, as a (re)description of some key aspects of the physical world, it can be used “to simulate and predict the sentient behaviour of a particle, person, artefact or agent (i.e., some ‘thing’)” (Friston et al., Reference Friston, Da Costa, Sajid, Heins, Ueltzhöffer, Pavliotis and Parr2023). Thus, if we were to take the free energy principle at face value, biological systems ought to be consistent with it much like a “particle” or an “artefact” ought to. Seth further argues that the free energy principle supports an explanation of autopoiesis, thought to be a central aspect of theories about the emergence of biological individuality and normativity (Maturana & Varela, Reference Maturana and Varela1980). It appears, however, that this principle does not in fact explain, model, or even account for autopoiesis in biological systems, conflating levels of analysis and definitions of “autopoiesis” (implying self-creation and a clear description of structure and organisation among other things) and “homeostasis” (Bruineberg et al., Reference Bruineberg, Dołęga, Dewhurst and Baltieri2022b; Di Paolo, Thompson, & Beer & Beer, Reference Di Paolo, Thompson and Beer2022; Nave, Reference Nave2025; Raja et al., Reference Raja, Valluri, Baggs, Chemero and Anderson2021; Suzuki, Miyahara, & Miyazono & Miyazono, Reference Suzuki, Miyahara and Miyazono2022).

Finally, Seth argues that active inference (Friston et al., Reference Friston, Fitzgerald, Rigoli, Schwartenbeck and Pezzulo2017; Parr, Pezzulo, & Friston & Friston, Reference Parr, Pezzulo and Friston2022) seen from a control theoretic/cybernetic perspective well synergises with accounts of interoception, the sense of the body “from within.” However, the cybernetic perspective proposed by Seth is hardly exclusive to biological systems, see the steam engine/Watt governor example already discussed above (Baltieri, Buckley, & Bruineberg & Bruineberg, Reference Baltieri, Buckley and Bruineberg2020). Formally, this simply appears as a restatement of classical works on “good regulators” and its formalisation in the “internal model principle,” which cover classes of systems that can be described as goal-pursuing by an external observer and thus not only biological systems (Baltieri et al., Reference Baltieri, Biehl, Capucci and Virgo2025; Virgo et al., Reference Virgo, Biehl, Baltieri and Capucci2025). One of the leading examples of this internal model principle in particular is integral control (Wonham, Reference Wonham1976), a core part of modern control theory (Åström & Hägglund, Reference Åström and Hägglund1995) with applications in biology (Yi et al., Reference Yi, Huang, Simon and Doyle2000) and neuroscience (Ritz et al., Reference Ritz, Nassar, Frank and Shenhav2017), which can also be easily accounted for in active inference (Baltieri & Buckley, Reference Baltieri and Buckley2019), making the special role supposedly played by active inference for biological systems not clear.

Overall, nothing about predictive processing, the free energy principle, and active inference seems exclusive to biological systems. In light of this, we thus believe that for a coherent theory of consciousness, we are left with one of the following choices:

1. 1. Pursue Seth’s biological naturalism, but with different tools. In other words, build an argument for biological naturalism using something other than predictive processing, the free energy principle, and active inference. This is because none of these seem to be in the business of characterising exclusively biological systems, much less some biological systems, the ones that ought to display the necessary conditions for consciousness in this proposal of biological naturalism.

2. 2. Reject Seth’s biological naturalism. Specifically, abandon the idea that there’s something inherently special about consciousness in biological organisms, fully embracing universality (Kanai & Fujisawa, Reference Kanai and Fujisawa2024) beyond biology, and frameworks that don’t a priori exclude the possibility of consciousness in artificial systems, like the free energy principle and active inference.

3. 3. Accept parts of Seth’s biological naturalism, but aim for biological universality. This means settling for a notion of universality that applies only to biological systems. The aim would be to make biological naturalism “as universal as possible” by providing formal, mathematical criteria for what should be understood as “biological,” e.g., what counts as a “biological neuron” if we assume that these units are somehow important for consciousness, or what counts as “being alive” if this ought to be the “minimal ‘ground state’ of conscious experience.”