I agree that to be conscious, “artefacts may have to be closer to biological systems than allowed for by conventional digital computation,” but not that this implies Biological Naturalism: “closer” can be functionally defined using “unconventional” computation. Intelligence and consciousness partly dissociate, but not because one is functional and the other intrinsic. Unconscious algorithms produce intelligent behaviour, but if consciousness evolved, it is because it does too. If it additionally has intrinsic, non-functional properties, Zombies lurch into view.
Consciousness is essentially about what a system does – we are conscious for a reason. “Feeling like something” does not imply intrinsicality. Experiences have the character they do due to “bottom-up” and “top-down” processes; to form unified experiences with intrinsic properties, these would need to be added together somehow. This is a form of Cartesian Materialism, which there are reasons to avoid (Dennett, Reference Dennett1991). However, my focus first is on the supposed incompatibility of computation and consciousness.
It is true that being usefully describable in terms of computations does not entail the brain computes, just as we might usefully describe a falling brick computationally, without assuming the brick computes. But we have independent reasons to say the brain computes, namely, intentionality. Mental states are about things, they are informational, so the processes they are involved in can be said to compute, in a way bricks do not.
“Coarse-grained” functionalism is too liberal, but finer-grained varieties are available. Turing-computational functionalism is particularly coarse-grained, but “4E” cognitive science supports other kinds of computationalism. Virtual Machine Functionalism (Sloman & Chrisley, Reference Sloman and Chrisley2003; Bowes, Reference Bowes2023) captures “generatively entrenched” organisational features required for mindedness, avoiding overly liberal functionalism: parts play their roles due to (contextual, dispositional) causal powers. The “limited substrate flexibility” of neurobiological functions makes Turing-computational functionalism unsuitable, but there are “broader forms of computation which have similar constraints.” However, broadening our understanding of computation undermines the argument for biological naturalism.
Embodied and embedded forms of computation are not “pale shadows” of life. There may be an “excitable fluidity” to biological entities ripe for the emergence of consciousness, but why think only biological entities can have this? The computers we build are “rigid” because we design them to blindly follow rules to match the semantics we assign to symbols. But computation does not necessitate a strict syntax/semantics separation:
“… a fundamental theory of programs, and of computation, should acknowledge that an essential function of a computer program is to make things happen. Whereas symbolic logic can be viewed as mere playing around with uninterpreted formal calculi…, and computational logic can be seen as the study of abstract timeless relations in mathematically specified ‘machines’ (such as Turing machines), computer science cannot properly be described in either of these ways. It follows… that the familiar characterization of computer programs as all syntax and no semantics is mistaken.” (Boden, Reference Boden1988)
The point is that arguments like Searle’s (Reference Searle1980) depend on the artificial separation of implementation from semantics, whereas in embodied and embedded computational systems, these cannot be separated.
So, natural minds are embodied and embedded physical computers, with representational content derived from engagement with a world, and substrate dependence that enables “relevant patterns of functional organisation.”
Seth uses the lack of substrate flexibility against embodied computation as it “might not be implementable in… non-biological substrates.” But, if radical substrate independence makes Turing computation unsuitable for consciousness, the need to be embodied and embedded in the right way should be seen as an advantage for a broad-minded computationalism.
The way to avoid overly liberal functionalism is to foreground embodiment – momentarily matching an abstract description is insufficient to realise a computation. There needs to be some substrate flexibility, though, so that relevantly similar mental states can be realised in reasonably varied embodiments. Maybe only biological substrates currently implement the right kind of computation, but knowing what makes them suitable should make replication possible. Can we just stipulate that these “replicants” are biological?
Accepting that “excitable” autopoietic processes are responsible for the emergence of life as we know it, then such processes were sufficient for the emergence of consciousness. But were they necessary, and would that support biological naturalism? The claim that consciousness came about through natural processes, and that carbon-based life on Earth is one way this could happen, is just naturalism.
“Life” may be defined by “homeostatic processes… linked to higher-level functional properties,” which are emergent, in that “knowledge of (or interventions at) lower, microscopic levels does not add anything to macro-level predictability, or help determine macro-level outcomes.” Seth says consciousness could depend on functional or intrinsic properties of the biological substrate, but that, given emergence, it is the properties at the level of life that matter. But the dependence is stronger, otherwise, the higher-level properties would be inexplicable - emergent in the “bad” way.
Seth claims that “apparently intrinsic properties at one level might always be understandable in terms of patterns of functional organisation at more fine-grained levels.” This makes intrinsicality perspectival, related to our explanatory interests. But this seems to conflate intrinsicality with explanatory autonomy. Instead of being “intrinsic,” non-relational properties, they are irreducible – emergent, but not in the sense of inexplicably appearing. To be explicable, and so of scientific interest, they cannot be intrinsic, as explanation requires relations (just not necessarily reductive ones). This implies inherently irreducible, rather than intrinsic, causal powers.
The higher-level causal properties of mindedness, then, are computational in a broad sense; being biological is part of the “finer-grained functional substrate” out of which they may emerge. The additional claim of necessity depends on an assumption that only biological entities can have the suitable properties, which has not been established.
Seth’s argument seems to be based on a Cartesian intuition regarding the intrinsicality of consciousness (that it’s not “about what a system does”), against functionalist accounts. This is in tension with his “real problem” approach, and is prone to zombie attack: given the substrate’s “finer-grained” properties, could there still be something missing, some intrinsic property that emerges in a way that is not explained by those functional facts plus their embodied history? If not, we should accept functionalism. If so, then the hard problem “pops up” again.
You have
Access
I agree that to be conscious, “artefacts may have to be closer to biological systems than allowed for by conventional digital computation,” but not that this implies Biological Naturalism: “closer” can be functionally defined using “unconventional” computation. Intelligence and consciousness partly dissociate, but not because one is functional and the other intrinsic. Unconscious algorithms produce intelligent behaviour, but if consciousness evolved, it is because it does too. If it additionally has intrinsic, non-functional properties, Zombies lurch into view.
Consciousness is essentially about what a system does – we are conscious for a reason. “Feeling like something” does not imply intrinsicality. Experiences have the character they do due to “bottom-up” and “top-down” processes; to form unified experiences with intrinsic properties, these would need to be added together somehow. This is a form of Cartesian Materialism, which there are reasons to avoid (Dennett, Reference Dennett1991). However, my focus first is on the supposed incompatibility of computation and consciousness.
It is true that being usefully describable in terms of computations does not entail the brain computes, just as we might usefully describe a falling brick computationally, without assuming the brick computes. But we have independent reasons to say the brain computes, namely, intentionality. Mental states are about things, they are informational, so the processes they are involved in can be said to compute, in a way bricks do not.
“Coarse-grained” functionalism is too liberal, but finer-grained varieties are available. Turing-computational functionalism is particularly coarse-grained, but “4E” cognitive science supports other kinds of computationalism. Virtual Machine Functionalism (Sloman & Chrisley, Reference Sloman and Chrisley2003; Bowes, Reference Bowes2023) captures “generatively entrenched” organisational features required for mindedness, avoiding overly liberal functionalism: parts play their roles due to (contextual, dispositional) causal powers. The “limited substrate flexibility” of neurobiological functions makes Turing-computational functionalism unsuitable, but there are “broader forms of computation which have similar constraints.” However, broadening our understanding of computation undermines the argument for biological naturalism.
Embodied and embedded forms of computation are not “pale shadows” of life. There may be an “excitable fluidity” to biological entities ripe for the emergence of consciousness, but why think only biological entities can have this? The computers we build are “rigid” because we design them to blindly follow rules to match the semantics we assign to symbols. But computation does not necessitate a strict syntax/semantics separation:
“… a fundamental theory of programs, and of computation, should acknowledge that an essential function of a computer program is to make things happen. Whereas symbolic logic can be viewed as mere playing around with uninterpreted formal calculi…, and computational logic can be seen as the study of abstract timeless relations in mathematically specified ‘machines’ (such as Turing machines), computer science cannot properly be described in either of these ways. It follows… that the familiar characterization of computer programs as all syntax and no semantics is mistaken.” (Boden, Reference Boden1988)
The point is that arguments like Searle’s (Reference Searle1980) depend on the artificial separation of implementation from semantics, whereas in embodied and embedded computational systems, these cannot be separated.
So, natural minds are embodied and embedded physical computers, with representational content derived from engagement with a world, and substrate dependence that enables “relevant patterns of functional organisation.”
Seth uses the lack of substrate flexibility against embodied computation as it “might not be implementable in… non-biological substrates.” But, if radical substrate independence makes Turing computation unsuitable for consciousness, the need to be embodied and embedded in the right way should be seen as an advantage for a broad-minded computationalism.
The way to avoid overly liberal functionalism is to foreground embodiment – momentarily matching an abstract description is insufficient to realise a computation. There needs to be some substrate flexibility, though, so that relevantly similar mental states can be realised in reasonably varied embodiments. Maybe only biological substrates currently implement the right kind of computation, but knowing what makes them suitable should make replication possible. Can we just stipulate that these “replicants” are biological?
Accepting that “excitable” autopoietic processes are responsible for the emergence of life as we know it, then such processes were sufficient for the emergence of consciousness. But were they necessary, and would that support biological naturalism? The claim that consciousness came about through natural processes, and that carbon-based life on Earth is one way this could happen, is just naturalism.
“Life” may be defined by “homeostatic processes… linked to higher-level functional properties,” which are emergent, in that “knowledge of (or interventions at) lower, microscopic levels does not add anything to macro-level predictability, or help determine macro-level outcomes.” Seth says consciousness could depend on functional or intrinsic properties of the biological substrate, but that, given emergence, it is the properties at the level of life that matter. But the dependence is stronger, otherwise, the higher-level properties would be inexplicable - emergent in the “bad” way.
Seth claims that “apparently intrinsic properties at one level might always be understandable in terms of patterns of functional organisation at more fine-grained levels.” This makes intrinsicality perspectival, related to our explanatory interests. But this seems to conflate intrinsicality with explanatory autonomy. Instead of being “intrinsic,” non-relational properties, they are irreducible – emergent, but not in the sense of inexplicably appearing. To be explicable, and so of scientific interest, they cannot be intrinsic, as explanation requires relations (just not necessarily reductive ones). This implies inherently irreducible, rather than intrinsic, causal powers.
The higher-level causal properties of mindedness, then, are computational in a broad sense; being biological is part of the “finer-grained functional substrate” out of which they may emerge. The additional claim of necessity depends on an assumption that only biological entities can have the suitable properties, which has not been established.
Seth’s argument seems to be based on a Cartesian intuition regarding the intrinsicality of consciousness (that it’s not “about what a system does”), against functionalist accounts. This is in tension with his “real problem” approach, and is prone to zombie attack: given the substrate’s “finer-grained” properties, could there still be something missing, some intrinsic property that emerges in a way that is not explained by those functional facts plus their embodied history? If not, we should accept functionalism. If so, then the hard problem “pops up” again.
Acknowledgements
None.
Financial support
None.
Competing interests
None.