Against Arguments from “Why Not?”: Reply to Deveci
Introduction
My old-time friend and colleague İbrahim Ethem Deveci recently published a rather interesting blog post on machine consciousness and biological naturalism, especially organized around Anil Seth’s ideas. The most valuable aspect of the piece is that it does not reduce the debate on machine consciousness to a contemporary and often superficial question such as “can artificial intelligence be conscious?”, but instead connects the issue to more fundamental philosophical distinctions such as medium-dependence, computational functionalism, autopoiesis, life-mind continuity, Dennett’s real patterns approach, FEP-PP frameworks, and the organization/structure distinction of living systems. In this respect, the piece more than deserves careful consideration. Moreover, I agree with a significant part of İbrahim Ethem’s criticisms of Anil Seth. Seth does seem to create a genuine tension when, on the one hand, he relies on frameworks such as FEP and predictive processing, which are highly formal, computational, and open to medium-neutral interpretations, while, on the other hand, he tries to tie consciousness very tightly to the biological organism and the living body. I also think it is mistaken to infer, from the fact that life or consciousness is “very complex,” that they cannot in principle be computationally modeled. Complexity and uncomputability are not the same thing, and a process that is practically difficult to simulate does not have to be a process that is in principle non-computable.
However, the main claim I will defend in this piece is that the incompleteness or inconsistency of Seth’s arguments does not show the truth of computational functionalism, and that some of the conclusions Deveci reaches are not justified. Criticizing one position and grounding its opposite are not the same thing. İbrahim Ethem’s piece may succeed in showing that machine consciousness cannot be easily ruled out by Seth and similar versions of biological naturalism, but it does not follow from this that computation can, or does, genuinely instantiate life, autopoiesis, or consciousness. In other words, at critical points in the piece, the question “why not?” seems to take the place of a positive argument. My objection begins precisely here. The emergence of life-like or mind-like patterns in cellular automata, the scientific/philosophical legitimacy of real patterns in Dennett’s sense, or the fact that autopoiesis can be defined as organizational in a certain sense, are not by themselves sufficient to support machine consciousness. For that, several distinct argumentative steps need to be justified. For example, from emergence to the mental, from simulation to instantiation, from computability to computationality, and from organization to medium-neutrality (I think the word “substrate” creates confusion outside the context of old-school physicalism. I return to this point below, and for this reason I will use expressions such as medium-neutrality or materiality-neutrality). In my view, these steps are not justified in İbrahim Ethem’s piece, they are simply assumed.
What Does Deveci Say?
First of all, İbrahim Ethem does quite a good job of laying out some of the basic fault lines of the debate. He draws both positive and negative connections between von Neumann, Conway’s Game of Life, the life-mind continuity thesis, FEP-PP approaches, Dennett’s real patterns, medium-(in)dependence, and autopoiesis, and summarizes a rather complex picture for us in a very clear way. While doing so, I think he also criticizes Seth and the enactivist tradition at the right points.
İbrahim Ethem’s argument proceeds as follows: 1) Dennett shows us that “emergence” is a meaningful concept, 2) cellular automata and Conway’s Game of Life show that computational systems are suitable for some kind of emergence, 3) therefore, there is no principled reason, at least on this line of thought, to rule out the emergence of “consciousness” in computational systems.
This is the first part of the argument. This part aims to show that “emergence” is a meaningful and reasonable concept. The second part, drawing on his criticisms of Seth, serves to present computational functionalism as a reasonable position. According to this criticism, Seth contradicts his emphasis on embodiment by relying on a computationalist formalism. Moreover, since autopoiesis is already a kind of “emergence,” it does not contradict the computationalist view, and cellular automata even provide a good case here. From this, Deveci argues that, since emergence is reasonable, and autopoiesis itself can be emergent and even computational, computational functionalism is therefore the stronger position, and the burden of proof lies with the other side. In this piece, I will first argue that there are some inadequacies in the appeal to emergence, and then claim that some of the criticisms directed at Seth, especially those concerning autopoiesis and medium-dependence, arise from misunderstandings of the literature.
While doing this, I will of course agree with some of his criticisms of Seth. However, I will argue that these criticisms can at most problematize Seth’s specific formulation, that they cannot by themselves eliminate biological naturalism and the autopoietic approach, and that even if they could do so, this step would not automatically make computational functionalism reasonable, since computational functionalism has so far not been supported by any justification beyond “why not?”
Real Patterns Okay, but Emergence of What?
Debates about the emergence of mind have, to a very large extent, proceeded through the question “what is mind?”, and within philosophy of mind they have almost always been framed through the dichotomy of whether the mind can be reduced or not. As an analytic philosopher, I think that the rather stunted tendency of twentieth-century analytic philosophy to reduce the problems at hand to simple binaries has seriously damaged the progress of both philosophy and science, and that the current debate on machine consciousness has not escaped this damage either. For this attitude has led to the illusion that showing that the mental is “irreducible,” or that some things in the world can be “emergent,” is enough to show that the mind, or some mental properties are also “emergent.”
In Deveci’s piece as well, the glider example he gives through CA shows only that when analyzing a system, we do not always have to begin and end with its lowest-level building blocks. Systems can be multi-scalar, and depending on what we are trying to do -analyze, describe, explain, predict, control, and so on - it may be perfectly meaningful, necessary, and even unavoidable to study the regularities that appear at higher levels as well. But this is not enough to say that the higher-level regularities in question, the “patterns” themselves, are “mental.” In other words, it only tells us that something “emerges”; it does not tell us what exactly has emerged, nor does it explain why this emergent thing should count as “mental” or why we should capture it with a mental vocabulary. It simply assumes that what emerges is mental.
Why? We can make the point more clearly by thinking about the lower-level / higher-level relation in simpler cases. Take a chair and a table, for example. They are both made of atoms, often made of the same materials, and the thermodynamic processes through which they interact with the external world are more or less similar. Even if we bring organization into the picture, they have largely similar “organizations” at the higher level. Yet none of this leads us to say that a table and a chair - or a chair and a stool - are the same things. We can push the point further. Almost everything is made of atoms, and different (emergent) organizations of those atoms give us, at roughly the same spatiotemporal scale - say, from millimeters to meters, and from seconds to hours, days, etc... but not eras, or centuries - all sorts of different “objects”: chairs, pillows, computers, organs, brains, human beings, ants, leafs, and so on. But the fact that all of these appear at roughly the same “level of emergence,” or can be studied at roughly the same “level of explanation,” does not make them identical to one another, nor do we try to identify them with one another. We would not say, for instance, that lungs and my backpack are “the same thing” simply because they have roughly similar dimensions and can be examined at the same “level of emergence.” Our distinctions between life, the mental, and other such categories still remain in place, and the answer to the question of what these things “are” cannot simply be that they are “emergent.”
Yet in Deveci’s piece - and in many appeals to emergence in philosophy of mind- the claim that certain things “emerge,” together with the usual examples given for this claim, is used for precisely this function and purpose. So, returning to İbrahim Ethem’s example, yes, in the context of CA, it may be reasonable to speak not at the level of “cells” but at the level of “gliders,” and to analyze, explain, and make predictions at this level. But the fact that the events and entities at this level have “emerged” cannot, by itself, indicate that the emergent thing at hand is a living system or something mental.The question of why we should approach this observed level with the mental terminology of “folk psychology,” rather than through this or that theoretical lens, remains unanswered. As far as I can see, neither Dennett nor anyone else has been able to give a meaningful answer to this question. When you read authors at the center of these debates, such as Putnam and Fodor, what you also see is that this step has been skipped: it has been assumed that showing that some things can be “emergent” is enough 1) to show that the mind is emergent, and 2) to show that we have to use mental concepts to explain human-level behavioral/physiological phenomena. Yet there is no logical entailment or necessity between these claims. In my view, they are not even particularly “suggestive” of one another.
In short, just as, when trying to determine whether something is a table –that is, when trying to come up with a “metaphysics of tablehood,” so to say- we have to answer to the question of what makes the table a table. And the answer cannot be just “it’s emergent” or “it’s organization.” That doesn’t say anything philosophically or empirically meaningful. At such a point what we need is a set of necessary and sufficient conditions or criteria of “tablehood.” And the situation is the same with the “emergence” literature in philosophy of mind and cognitive science. The million dollar question is “what are the necessary and sufficient conditions of the mental, and more specifically of consciousness?” and pretty much everyone seems to be ignoring it and just asserting that it is emergent, or it has “organization,” however underspecified these notions often remain. It seems to me that this is the most fundamental level at which the debate takes place, and neither İbrahim Ethem nor other computational functionalists answer this question.
At this point, one move that is often made is the claim that mental concepts are “successful enough” in explaining and predicting the phenomena in question. The arguments offered for this claim, especially in the Fodorian line, are often derived from the success of these concepts in everyday life. But in my view, this is not a valid argument, and it faces three related problems: 1) Folk psychology has already emerged as a product of the history of life and humanity in such a way that it can work for certain kinds of regularities in everyday life, and the fact that it is “successful enough” in this specific context does not show that it is also the “right conceptual framework” for a scientific research program; 2) related to the first point, scientific regularities always require more complex frameworks than folk theories. Much of what we call scientific progress has involved updating and replacing our folk-theoretical explanations with new, scientific, and technical conceptual sets precisely at the points where they prove insufficient; otherwise, how could anything like germ theory, or quantum mechanics, or relativity, or electromagnetism have emerged? 3) If the “empirical success” of a theoretical conceptual set is evaluated through the lens of that very conceptual set, it always runs the risk of appearing “successful enough,” and if scientists and philosophers are not open-minded about this, they can continuously justify their own conceptual frameworks with ad hoc assumptions. That is, just as we could not move physics beyond Newtonian mechanics if we insisted on looking through the eyes of Newtonian mechanics, we could not have moved astronomy forward if we had looked only through the eyes of Ptolemy’s epicycles either. because, with enough ad hoc additions, such approaches can continue to appear empirically adequate, even when they are no longer explanatorily progressive.
Therefore, if we look from the perspective of a cognitive science in which folk psychology appears everywhere from its hypotheses to its experimental design, then yes, folk psychology will of course appear successful. But an increasing body of empirical and theoretical work shows the inadequacy of the folk-psychological conceptual set for cognitive science.
FEP-PP and Seth’s Inconsistency
We will return to the issue discussed in the previous section toward the end of this piece. For now, let us proceed through Deveci’s criticisms of Seth. İbrahim Ethem first argues that, by placing FEP-PP approaches at the center, Seth relies on a computationalist and, in this sense, medium-neutral framework, and that this is inconsistent. I agree that there is a real tension here. But in my view, there is still an intermediate position that is defensible from Seth’s point of view.
We emphasized that the basic issue has to do with the necessary and sufficient conditions of “the mental.” Seth may adopt an embodied and medium-dependent position regarding the necessary and sufficient conditions of the mental, that is, regarding his stance on the “metaphysics of mind.” And he may claim that the conditions under which the mind emerges cannot be separated from the bodily processes that also constitute life. On the other hand, while doing this, he may at the same time adopt an instrumentalist attitude toward FEP or other computationalist approaches. There is no categorical obstacle to this. For instance, someone like me, who is ontologically a radical anti-computationalist, can perfectly well say that computational simulations and models are important and insightful in scientific research processes. Similarly, in disciplines such as ecology or geology, computer simulations and computational models are perfectly well used. This does not push anyone in these fields to say that mountains, stones, or certain ecological phenomena are computational “in themselves,” or that the relation here is one of “instantiation / realization.”
Speaking for myself, despite finding computational models insightful, I do not adopt such an instrumentalist attitude toward them. The reason is that I think instrumentalism is problematic in many contexts in the philosophy of science. But since my views on this issue fall outside the scope of this piece, I will not go into that point here (As a hint, my view is based, at the most basic level, on the thought that “insightful” and “scientifically explanatory” are not the same thing). My only concern here is to show that Seth’s inconsistency is not, in fact, a categorical contradiction. That is, I do not think the claim that “he cannot have it both ways,” as İbrahim Ethem suggests, is correct.
Organization of What?
Deveci reaches an important point in his criticism of Seth. In my view, the fundamental point on which the whole piece actually rests is laid out in the section titled “Autopoiesis is Computable.” This part of the argument is quite dense and rich, so let us unpack it step by step.
In this section, Deveci rightly reminds us of Maturana and Varela’s definition of autopoiesis. He emphasizes that, for Varela and Maturana, autopoiesis is not related to a specific substrate but to “organization.” This is indeed true. Maturana and Varela do distinguish between “structure” and “organization” in their work. According to this distinction, while structure is related to materiality (notice that, I’m avoiding the word “substrate”), organization is what allows the cell, or the system, to maintain itself and to continue self-production despite changes in the specific “structure”. The basic intuition here has to do with the fact that a living being can preserve its “identity” and “unity” even though all of its cells, or all of a cell’s organelles, proteins, and so on, change over time. Up to this point, I do not think there is a problem. What actually surprised me, however, was that Varela, in his Principles of Biological Autonomy, uses cellular automata as an example of autopoiesis. For instance, we do not see such an example in Maturana’s Autopoiesis and Cognition. Moreover, compared to Varela, Maturana is much more conservative and, in my view, more consistent in the concepts he uses, especially in his early writings, but I will leave this aside here. Therefore, Deveci’s attempt to connect CA with autopoiesis, and to criticize Seth from this angle, given that Seth places autopoiesis at the center of an argument about the medium-dependence of the mental, is quite reasonable.
While we are on this point, I should say that, when I was reading the enactivist literature, I too thought that this organization/structure distinction was a kind of “functionalist” leakage, and this is one of the reasons why I do not call myself an enactivist today. On the other hand, in my view, over the years, the use of the concepts of “autopoiesis” and organization in the enactivist literature has gradually changed, and, as far as I can see, apart from a few people who try to bring computationalism and enactivism together, almost no one now gives CA as an example of autopoiesis. But going into this point would be unproductive, so I will continue along the path Deveci points to.
I think Deveci uses the word “organization” in a much broader sense than enactivists ever did. It is true that Varela and Maturana propose the concept of organization as something more “abstract” than structure, and as a concept not limited to the structure “itself.” But the point here is to indicate how a living organism can continue to exist as a singular organism, a singular biological whole, a unity, even though all of its cells and proteins -and even its atoms and so on - change from moment to moment, and how it can “renew” its own cells and organs. So the emphasis here is not on an abstract “substrate-neutrality” in the functionalist sense, but on the fact that material structures, because they stand in a specific relation to one another, and in virtue of this particular type of dependency, can continue self-production and maintain their existence. And it is precisely here that we see that this substrate-neutrality is not “unbounded.” For instance, Deveci says that since the mental can arise in very different living beings, and since living beings themselves have very different biological bases, the autopoietic approach also endorses multiple realizability. This is true. I also think this is, in a “sense,” multiple realizability. But is the multiple realizability at stake here the unbounded multiple realizability proposed by computational functionalism? I do not think so at all.
First, what needs to be noticed here is that what makes “organization” possible is a specific material arrangement, a relation of material/physical dependency. When I say “specific” here, I do not say this because I know what it actually is, but whatever this relation is, it seems to be something that living beings have and that sea waves “beating” rocks or strong winds “bending” trees do not. For computational functionalism, however, substrate-neutrality is an unbounded property. What gives rise to mind does not go beyond being “modelable” by a specific kind of computational pattern. Precisely for this reason, while the idea of a giant “mind” made of cheese and bread does not contradict computational functionalism, as long as it functions as a hardware for some software. But it does not seem very plausible for the enactivist concept of organization. To begin with, it is quite obvious that computers do not have such a capacity to “produce and maintain themselves” in virtue of the relations among their own material components, a.k.a. hardware. That is, when we examine a computer, regardless of which algorithm it instantiates, through the lens of the Varelan and Maturanian structure/organization distinction, I see no problem in saying that, with respect to autopoietic self-production, it is closer to a stone than to a living system. More on this below.
At this point, two questions can quite naturally arise: 1) Could the crucial criterion be far-from-equilibrium thermodynamics? 2) Does the fact that we have so far observed the emergence of mind only in carbon-based living beings mean that it necessarily has to be this way?
Deveci gives some answers to both. For example, regarding thermodynamics, he says: “And insofar as the thermodynamic story is made precise, it is made precise through the Free Energy Principle, which is itself a substrate-neutral formalism, already implemented in artificial agents.” There are two problems with this. 1) FEP is not “precise” in the relevant sense, at least not in a way that straightforwardly settles the metaphysical question at hand. 2) The idea that thermodynamics, or more generally physical energy processes, is one of the necessary conditions of life, “meaning/representationality,” and the mental goes back long before FEP (Bickhard, Haken, Atlan, Prigogine, von Foerster, and even Ashby). Seth may not have fully absorbed this history, but if what we are discussing here is not Seth’s ideas as such, but computational functionalism versus a kind of biological naturalism in the context of machine consciousness, then a fair criticism should not depend only on Seth’s arguments. More importantly, however, Deveci says that computers are also physical systems and are already subject to thermodynamic laws, and that therefore, even though energy processes are not intrinsic to the computational formalism, they apply to all physical systems that realize computing. This is true, but in my view this objection stems from Deveci’s missing a fundamental point emphasized by Seth and by the autopoietic tradition.
There is an issue that Deveci himself also mentions by quoting Seth. The quote is this: “Brains may be the kind of thing for which it is hard, perhaps impossible, to separate what they do from what they are.” To be honest, I would not formulate this sentence in such a brain-centered way, because this applies, in fact, to all autopoietic systems, that is, to all systems that can reproduce and maintain their organization - their unity and identity - through the specific relations their material components establish with one another, and that can do so even though the things that compose their structure are in constant change. So the real difference here is this: for the enactivist, an autopoietic system does what it does in virtue of what it materially is, whereas a computational system does what it does in virtue of its algorithm and irrespective of what it materially is. That is, the basic issue in the conception of autopoiesis is that, while in the case of a computer, the algorithm is sufficient to explain the performance and behavior of a computational system or model; on ther other hand, in the case of autopoietic systems, “structure and organization”, by definition, make material processes explanatorily necessary. I should underline here that what I mean by “material” is not the same as what the physicalist literature in analytic philosophy means by “physical” or “substrate.” For enactivists, when we speak of the material, this can include even sociocultural scaffolding in the case of humans or social animals, whereas for physicalists these are certainly not “physical” things. For example, without changing the hardware of the computer I currently have, I can run an entirely different algorithm, in various ways unlike all the algorithms I have run on this computer so far. But in the case of a living being, it is not possible, at some point in time, to suddenly change the algorithm and get an entirely different performance and output. Because what is responsible for the behavior of the system here is already material organization - which includes millions of cells, tissues, organs, and the relations they develop within the life trajectory of that living being. In the case of the computer, at the “level of explanation,” computational description - I do not mean this in the Marrian sense, but in a more general sense - is primary and in most cases sufficient, whereas in autopoietic systems, what is primary is materiality, and computational description is secondary, if it is relevant at all.
The other objection that may come to mind, and that Deveci also addresses, is the fact that we have so far observed the emergence of mindd and consciousness only in carbon-based living beings is not enough to conclude that it must always be this way. This is true. But in my view, this too stems from a misunderstanding. In my readings of enactivism, as far as I can remember, I have never seen an emphasis on “carbon-based” as such (so, this seems to be a very common straw-man). What is emphasized here is “materiality” in the sense I have just described. What does this mean? Let us open this up with a thought experiment.
Suppose, for example, that material science has advanced a great deal and that scientists have developed a mercury-like metal called “Mercurium.” Unlike mercury, however, let this metal have two additional properties besides being fluid: 1) by using other metals and various chemicals, it can function like a “cell,” as an organization with unity and identity that can maintain itself even though its material subcomponents change, and 2) these cell-like units can reproduce and multiply, in virtue of their thermodynamic and other physical interactions with things external to themselves, perhaps in a way similar to DNA and protein synthesis in living beings, and thus become capable of developmental processes of the sort we see in developmental biology, such as growth, compartmentalization, some sort of biological modularity –which is quite different from the modularity of mind or brain- and the development of tissues and organs.
Although there is much to think about in this thought experiment, and much here that could inspire science fiction novels, I will not go into that for now. What matters for us is that Mercurium is not carbon-based but is nevertheless an autopoietic system. That is, provided that the necessary energy conditions, material requirements, or whatever else this system needs are supplied, it can develop just as other living beings grow and develop. Perhaps the example of “life and mind” displayed by this Mercurium will not resemble anything we have previously known about life or mind. But from an enactivist point of view, there is no obstacle to this system’s being alive, being minded, and even being conscious. What I am trying to emphasize here is that the autopoietic approach does not impose “carbon-based biology” as an ad hoc and groundless criterion. It is to draw attention to the fact that what is meant by biological regularities and materiality is different from what is meant by “physical” or “biological” in computational functionalism.
At this point, Deveci says that if this medium-dependent relation that constitutes mindedness is not simply autopoiesis, because he thinks autopoiesis is already computational, which we have discussed, and if it is not thermodynamics either, then we need to be told what it is. And he is right. Based on the explanation I have given here, I think the concepts of autopoiesis and materiality already point to the right kind of relation. But does this need to be made even more precise? This is always necessary and required. Science progresses only in this way. And as work in this area develops, the answers given on this matter will undoubtedly develop as well. But what actually surprises me is this: does computational functionalism, in relation to the ontological claim it makes, offer any precise and well-grounded justification beyond “why not?” I have not encountered such a justification either in Deveci’s text or in the literature I have been intensely reading for nearly ten years. Therefore, in the face of this legitimate demand, I think we are also entitled to expect from computational functionalists a reasoned and positive ontological framework, beyond criticizing the arguments of those who object to them and then saying “why not?”
Before closing this section, I want to address one more argumentative move by Deveci. Immediately after saying that it is contested whether computational models that resemble lifelike regularities are “genuinely autopoietic,” he lists some relatively recent studies in the field and examples he finds successful, and then states that these computational models do not simulate “lifelike” properties but instantiate them, and that claiming otherwise would be contrary to the evidence. To be honest, I could not quite understand what he bases this inference on. After all, what we expect from a good simulation is always that it simulates and performatively displays lifelike or mindlike properties well. Looking at a model, stating that the simulation vs. instantiation / realization question is legitimate, and then asserting instantiation/realization without giving any argument for that answer seems like a non sequitur.
The Proposition That Something Is Computable Does Not Mean That It Is Computational
After the section titled “Autopoiesis is Computable,” Deveci argues that complexity and uncomputability are not the same thing, and even that they are orthogonal to one another, and he defends this point at length. To be honest, I agree completely. It is the kind of section that should be taught in schools. But I think there are still points here that are passed over with respect to the focus of the debate, and that therefore lead to an ungrounded “leap of faith.”
Since I agree with almost all of the arguments in this section, I will not focus on Deveci’s claims one by one. What I want to talk about here is an important nuance that I was surprised not to see addressed, especially given that the relevant distinctions in this debate are discussed in such detail.
To be honest, I agree that life, the mental, and even the entire universe may be computable in the sense of being “computationally modelable.” There is no obstacle to this. And indeed, in the very near future, there may be artificial agents or computational models that we cannot distinguish from human beings at all, and that perhaps have much more advanced cognitive capacities, models that “simulate” intelligence better than we can even imagine. Let us go even further: in principle, if we can meet the technological requirements necessary for this, we may one day be able to “simulate” a universe as complex as ours from scratch with a computational model, and we can do so in a way that captures all of its complexity. I have no objection to any of this. But, as should be clear from my insistence on the word simulation here, the fact that these things are “in principle” computable, and that such one-to-one models can be created, says nothing about whether those things are “computational” in themselves. This is precisely why I see the leap from computability to computationality as a leap of faith, and I do not think I have encountered an argument for this anywhere in the piece.
But why so? This brings us to the next section, and to the rainstorm example.
Rainstorms and Hologram Pizza
Deveci opens this section with the following quote: “Simulating a rainstorm doesn’t make anything wet, so simulating a brain doesn’t make anything conscious.” But immediately afterward, he raises a legitimate objection: the only way this example makes sense is if we already assume that consciousness, like wetness, is a substrate property. Without this assumption, the example is not meaningful. True. But what surprises me is that, in order to accept the example he gives for the opposite claim, and to compare consciousness or the mental to chess, one also has to assume that consciousness and/or the mental has nothing to do with its substrate. Here I should add a note: in my view, enactivists already do not say that consciousness and the mental are substrate properties, but they do say, in the sense I described above, that they are materiality properties. Accepting the chess-game metaphor as a good example for mind and consciousness likewise requires assuming that they are not materiality properties either. We need to pay attention to this nuance. Therefore, the defect raised as an objection to the enactive approach, namely the problem of question-begging, is present for computational functionalists as well. In order for examples of this kind to be meaningful for them too, they must already have made a very strong assumption about mind and consciousness.
A hologram pizza may look like a pizza and may even guide our expectations in pizza-like ways, but unless it is made of dough, cheese, heat, and edible ingredients, it is not a pizza in the relevant sense. The point is not that consciousness is obviously like pizza rather than chess, the point is that analogies of this kind work only after we have already decided what sort of property consciousness is. And the criticism here leveled against Seth works both ways.
Concluding Remarks
In the rest of the piece, Deveci turns to the hard problem of consciousness and says that it applies to all physicalist approaches, not only to computational functionalism. I agree with this completely. But with one nuance: because of the points I made above about the word “physicalist,” I would prefer to formulate this claim in terms of “all materialist” or “all naturalist” approaches.
The position I have defended here is that, while some of Deveci’s criticisms of Seth are correct, others arise from misunderstandings of the autopoiesis literature, and that, in its current form, this rejection of Seth does not contain arguments that support computational functionalism. For this reason, throughout the piece, I have tried to preserve three distinctions. First, the possibility or plausibility of emergence does not show the emergence of the mental. Gliders in cellular automata, or other higher-level patterns, may show us that multi-scalar explanation is legitimate, but a further argument is needed to show that these patterns are mental, conscious, or representational. Second, the fact that a phenomenon can be computationally modeled does not show that the phenomenon is computational in itself. The move from computability to computationality needs to be justified sepearately. Third, the fact that a running simulation is a physical process does not mean that it instantiates the phenomenon it simulates. The difference between simulation and instantiation cannot be closed by saying, “but the program is also physically running.” Similarly, the fact that autopoiesis is about organization does not make it substrate-neutral (or more accurately, medium-neutral or materiality-neutral) in the sense understood by computational functionalism. The organization/structure distinction in Maturana and Varela is not a hardware/software distinction. The issue here is not that an abstract form works in the same way in any material carrier, but that certain material processes are organized in such a way that they can maintain their unity and identity even as their components change, and they do so in virtue of their material organization, not in virtue of their “algorithms”. For this reason, what is meant by materiality is not an ad hoc commitment to carbon-based biology. As in the Mercurium thought experiment, a non-carbon-based system that produces its own subcomponents, grows within energy processes, differentiates, and maintains its own organization may very well be autopoietic, alive, and perhaps even mental. But this possibility does not point in favor of computational substrate-neutrality, it points to the importance of material organization.
For this reason, in my view, the most reasonable position here is a kind of agnosticism. But this is not an empty and indifferent “we do not know” attitude. On the contrary, it is an agnosticism that demands stricter conceptual distinctions. Machine consciousness may be possible, but its possibility cannot be taken philosophically seriously simply by pointing to the weaknesses of the opposing side. The failure of one argument is not, by itself, the success of its opposite. Each of the transitions from mere emergence of anything to the emergence of mind, from computability to computationality, from abstraction/multi-scales to the hardware/software distinction, and from simulation to instantiation must be justified separately. Unless this is done, what computational functionalism offers is not a strong ontological argument, but at most food for thought.
In conclusion, I find İbrahim Ethem Deveci’s piece valuable because it brings the debate on machine consciousness to the philosophical level at which it really needs to be discussed. But precisely for this reason, I do not think the question “why not?” is enough. When it comes to concepts such as consciousness, life, representation, and mind, scientific and philosophical progress cannot be achieved merely by opening up spaces of possibility, but by showing under which material, organizational, and explanatory conditions these possibilities are genuinely possible. The debate on machine consciousness, too, can be taken philosophically seriously only insofar as it does not leave these crucial transitions undefended.