As Seth argues, the only example of consciousness that we can confidently confirm as real is the byproduct of billions of years of natural selection – that is, consciousness as we know it is inherently biological. In his provocative commentary, Seth argues for biological naturalism, which is the view that the subjective, phenomenal aspect of consciousness is not separable from the biological machinery that keeps an organism alive. This makes life itself a necessary condition for consciousness. Seth presents this position as his preferred view rather than a definitive conclusion, and we share his caution. Others have argued that life itself is computational (Agüera y Arcas et al., Reference Agüera y Arcas, Alakuijala, Evans, Laurie, Mordvintsev, Niklasson and Versari2024; Kurian, Reference Kurian2025), so we would argue there is further work to do to show that consciousness is not inherently computational, even granting Seth’s premise that consciousness and life are intrinsically linked.

Specifically, it is unlikely that all aspects of biology are crucial for consciousness. So then, the central questions from our point of view become: which exact details of biology matter for consciousness, and can these specific details be computationally implemented (or not)? In simple terms: while every instance of consciousness that we can currently confirm with confidence is inherently biological in nature, this does not mean that every last detail of biology is required in order for a system to be conscious. Indeed, some aspects of biology seem to be important, e.g., the reticular activating system, higher-order thalamic nuclei, and pyramidal neurones oriented orthogonally to the layers of the cerebral cortex (Larkum et al., Reference Larkum, Petro, Sachdev and Muckli2018), whilst others do not (e.g., Purkinje neurones in the cerebellum or the dura mater). That is, they may be simply part of the substrate that is “along for the conscious ride” (i.e., conscious “spandrels”). Thus, work separating the conscious wheat from the unconscious chaff is needed before we can draw conclusions about the limits of computationalism. If we can separate the crucial substrates from the supporting architecture, it may be possible (in principle) to implement the crucial features of (neuro)biology on a non-biological substrate. At this stage, our best guess is that this will require far more biological detail than we currently know how to build. Not only would we need to include L5-ET cells and their complex dendritic tree (Aru, Suzuki, & Larkum, Reference Aru, Suzuki and Larkum2020) which we have argued may provide a cellular mechanism for cortical associations (Larkum, Reference Larkum2013), we’d also need to integrate them with a working thalamus (Fang et al., Reference Fang, Dang, Ping, Wang, Zhao, Zhao and Zhang2025; Whyte et al., Reference Whyte, Müller, Aru, Larkum, John, Munn and Shine2025) and an ascending arousal system (Munn et al., Reference Munn, Müller, Medel, Naismith, Lizier, Sanders and Shine2023). From there, we have argued through the lens of neuroscience that a disembodied cortico-thalamo-reticular system would not be conscious of anything much at all (Aru, Larkum, & Shine, Reference Aru, Larkum and Shine2023) – for the system’s consciousness to be as rich as ours, the system must be integrated into a body that can provide an interface with the complex, multi scale nature of the surrounding world.

Of the biological features that seem to play an essential role in brain function, and potentially consciousness, scale-free organisation is likely the most difficult to instantiate in a substrate-independent manner. At rest, brain activity exhibits scale-free organisation, meaning its dynamics span multiple temporal and spatial scales without a single dominant scale. Depending on task demands, this baseline organisation can be selectively biased to emphasise certain scales of activity (Munn et al., Reference Munn, Müller, Favre-Bulle, Scott, Lizier, Breakspear and Shine2024). Crucially, such scale-free behaviour causes problems for computational theories which privilege a particular scale (e.g., population coding; Averbeck, Latham, & Pouget, Reference Averbeck, Latham and Pouget2006), challenging the idea that we can draw a clear distinction between hardware and software as the computations being performed may require contributions from distinct or (multiple) scales. It may not be possible, therefore, to have human-readable software that can be implemented on any substrate if the (scale-free) substrate is itself essential for the computations being performed. For us, this also means that consciousness cannot be understood at some higher, substrate-independent level of description (Gidon, Aru, & Larkum, Reference Gidon, Aru and Larkum2022, Reference Gidon, Aru and Larkum2025).

Thus, while we support Seth’s line of reasoning (Aru et al., Reference Aru, Larkum and Shine2023), we would not announce the end of computationalism yet. Instead, we would argue that computationalism comes in degrees. There is naive computationalism, which states that present-day understanding of the computations underlying consciousness might be enough to build conscious machines (Butlin et al., Reference Butlin, Long, Elmoznino, Bengio, Birch, Constant and VanRullen2023). However, computationalism could also, in principle, be much more complex, taking into account details about the brain – we prefer the term “biological computationalism,” in order to separate it from the naive view. For instance, one could in principle build a computational theory that incorporates the fact that the brain’s organisation and dynamics are fundamentally scale-free, while also allowing for task-dependent modulations that temporarily weight certain scales more heavily (Munn et al., Reference Munn, Müller, Favre-Bulle, Scott, Lizier, Breakspear and Shine2024). Consciousness might be a phenomenon that only exists when a complex, multiscale nervous system is embedded within a living organism that is itself embedded within the natural world, which is itself replete with a dizzying amount of multiscale structure. By incorporating these key relational properties, a future theory of biological computationalism could, in principle, capture the essential ingredients necessary for consciousness.

The way forward is theoretically simple, yet painstaking in its execution. For one, we need to devise experiments across multiple scales and then build causal links between the different scales. And at all stages, we will have to carefully evaluate whether a constituent part is necessary for consciousness or just a byproduct or background condition. In silico models will be extremely helpful in this regard, as they can act as a crucial, bidirectional bridge connecting biologically based theories of consciousness with the crucial empirical tests (Whyte et al., Reference Whyte, Müller, Aru, Larkum, John, Munn and Shine2025).

Eventually, we hope that Seth’s article brings a change to consciousness science. Due to the rise and popularity of neurobiologically underspecified theories, biology has been in the background for too long. Let’s not pretend that understanding these specific processes that underlie consciousness should be simple. Seth’s article shows that we still have a long way to go and much exciting science ahead of us.