Hedda Hassel Mørch (NYU / University of Oslo)
Abstract: The Integrated Information Theory (IIT) is an empirically motivated theory of consciousness which entails a kind of panpsychism. In this paper, I discuss whether IIT is compatible with Russellian (or dual-aspect) panpsychism which has recently been defended in philosophy of mind. I will show that if IIT were compatible with Russellian panpsychism, this would contribute to solving Russellian panpsychism’s combination problem. However, the theories aren’t compatible as they currently stand, in view of what I call the coarse-graining problem. I will explain this problem, and offer two possible solutions. One solution involves a modification of IIT’s Exclusion postulate; another involves a modification of its coarse-graining principle.
Panpsychism is the view that every physical thing is associated with consciousness. More precisely, it’s the view that every physical thing is either (1) conscious, (2) made of parts which are all conscious, or (3) itself forms part of a greater conscious whole. Humans and animals (or certain areas of human and animal brains) are conscious in the first sense, and panpsychists typically think fundamental particles, such as quarks, are conscious in this sense too. Artifacts and arbitrary aggregates, such as tables, chairs and heaps, are typically regarded by panpsychists as conscious only in the second sense, in virtue of being made of conscious particles. Particles and neurons that form part of human and animal brain areas which are conscious in the first sense are perhaps only conscious in the third sense. By “consciousness”, panpsychists mean phenomenal consciousness, i.e., the property of having some form of subjective, qualitative experience, or of there being something that it’s like for something to exist.
Panpsychism has recently received new attention in both philosophy and neuroscience. In philosophy of mind, it has been argued that a version of panpsychism, panpsychist Russellian monism (named after Bertrand Russell, for reasons that will be explained below), is able to avoid the most serious problems that confront the traditional main positions in the field, physicalism and dualism (Nagel 1979; Strawson 2006; Alter and Nagasawa 2012; Chalmers 2015). For physicalism, this is the problem of the epistemic gap – aspects of which are highlighted by the knowledge argument (Jackson 1982, 1986) and the conceivability argument (Kripke 1980; Chalmers 1996). For dualism, it’s the problem of mental causation (Kim 1988; Papineau 2001).
In neuroscience, the Integrated Information Theory of consciousness (IIT) (Tononi 2004, 2008, 2011; Tononi, Albantakis, and Oizumi 2014) is a new fundamental theory of consciousness which has proved to have significant explanatory and predictive power. According to IIT, consciousness is correlated with integrated information, also referred to as Φ. Information in IIT measures the extent to which a system causally constrains its own immediate past and future state, and integration measures the extent to which this information depends on the causal interconnectivity of the system’s parts. Since even fundamental particles, such as protons and neutrons,  have a small amount of Φ (Koch 2012: 132), IIT entails panpsychism in the sense defined above.
IIT sometimes encounters the objection that although it may appear to account well for empirical data about human consciousness – such as the absence of consciousness during deep sleep, general anesthesia, epileptic seizures, and in regions of the brain such as the cerebellum, all of which correlate with low values of Φ (Tononi and Koch 2015) – it can’t be correct because panpsychism is absurd (Searle 2013). But if Russellian panpsychism solves the main philosophical problems of consciousness, IIT’s panpsychism should be regarded as a feature, not a bug. IIT could claim philosophical support in virtue of being the only current neuroscientific theory compatible with the solution to the mind-body problem.
But not everyone agrees that Russellian panpsychism solves the mind-body problem. The main objection to Russellian panpsychism is based on the so-called combination problem (James 1890; Seager 1995). The combination problem consists in explaining how complex consciousness, of the kind we know from our own case, could result from the appropriate combination of entities with simple consciousness, i.e., the particles that constitute our brain and according to panpsychism possess rudimentary forms of consciousness. According to the objection, the combination problem gives rise to further problems which are strongly analogous the main problems of physicalism and dualism (Goff 2006, 2009; Chalmers 2015, 2016).
At first glance, it appears IIT can help Russellian panpsychism solve the combination problem. The precise claim of IIT is that a system is conscious, in the first sense, if and only if it is a maximum of Φ, i.e., if the system has higher Φ than any of its own parts as well as any bigger system of which it forms part. This is IIT’s Exclusion postulate – consciousness does not overlap. This means that whenever the Φ of a system surpasses the Φ of the parts, the system goes from being conscious in the second, aggregative sense, to being conscious in the first, unified sense. This explains how mental combination occurs, at least in certain respects. Tononi and Koch imply (2015: 11, 13), somewhat contentiously, that IIT offers a complete solution to the combination problem in and of itself. But one might also regard IIT as a contributing factor to a complete solution, when taken together with other proposed solutions to the problem. Other proposed solutions include the phenomenal bonding view (Goff 2009, forthcoming) and the fusion view (Seager 2010, forthcoming; Mørch 2014). These proposals look promising in purely metaphysical respects, but both arguably fail to show what could possibly constitute an empirical correlate of combination. IIT shows what this correlate could be, namely maxima of Φ.
Unfortunately, however, a solution to the combination problem in terms of IIT is not straightforward. In order to solve the combination problem, IIT needs to be compatible with Russellian panpsychism, not just panpsychism in general. Panpsychism could also be combined with dualist or physicalist metaphysics, but will then have no particular advantage compared to standard physicalism or dualism with respect to the mind-body problem. And there are some indications that IIT is only compatible with dualist or physicalist panpsychism. This would be bad news for both IIT and Russellian panpsychism: IIT could no longer claim that its panpsychism is in fact a philosophical advantage, and Russellian panpsychism would retain its combination problem.
In this paper, I will explain what indicates this disappointing result, and argue that it mainly derives from what I will call the coarse-graining problem. I will show that this isn’t a surface problem: IIT and Russellian panpsychism are in fact incompatible as the theories currently stand. But the conflict could still be resolved with the introduction of some substantive new principles. I will suggest two candidates for such principles; one involves a modification of IIT’s Exclusion postulate (not abolition, as some have called for (Schwitzgebel 2015)); another involves a modification of IIT’s principle for determining the spatio-temporal grain of conscious systems. My arguments, if successful, will not support a fully satisfactory solution to the combination problem in terms of IIT, but they should give reason for optimism and motivate further research into this approach.
I will begin by explaining the metaphysics of Russellian panpsychism and the combination problem in more detail. Then I will discuss how IIT could help with the combination problem, either on its own, or combined with the phenomenal bonding or fusion view. Then I will present the coarse-graining problem, which shows that the views are in fact incompatible, followed by my two proposals for how to resolve the problem.
2. Physicalism, Dualism and Russellian Panpsychism
According to Russellian panpsychism, consciousness is correlated with every physical thing in virtue of constituting the intrinsic nature of physical relational structure. As Bertrand Russell noted (1927), physics (or the physical sciences) describes only relational or structural properties – how physical things relate to other things, but not what they are like in and of themselves. These are the kind of properties which can be described in the austere language of mathematics. The phenomenal properties of consciousness, in contrast, seem to be intrinsic – we know what they are like in and of themselves, independently of how they relate to other things. According to Russellian panpsychism, structural properties need non-structural realizers, and relations need relata with intrinsic properties. Furthermore, phenomenal properties are the only intrinsic properties we know to exist. Russellian panpsychists infer from this that consciousness could be the realizer of all physical structure, or an intrinsic property of every entity described by physics in relational terms (Seager 2006; Alter and Nagasawa 2012).
As noted, the main attraction of Russellian panpsychism is that it offers a solution to the mind-body problem which appears to avoid the main difficulties of physicalism and dualism at once. Physicalism is the view that phenomenal (i.e., conscious) properties are identical to, or constituted by, physical properties – where physical properties are understood as the kind of properties that physics (or the physical sciences) describes. Its main problem is that it appears impossible in principle to see any entailment from the existence of physical properties to the existence of any phenomenal properties – something which should be possible if the latter were indeed identical to or constituted by the former (Chalmers 2003) (and we understand the nature of both the mental and the physical, see Goff (forthcoming)). In the absence of such an entailment, we’re able to conceive of the mental and the physical coming apart, as the conceivability argument (i.e., zombie argument) highlights (Chalmers 2009). Also, someone who knows all physical properties would not be able to deduce the existence and nature of mental properties, as the knowledge argument highlights (Jackson 1982).
Dualism is the view that phenomenal properties are distinct from physical properties, and are causally generated or affected by them. Its main problem is the threat of epiphenomenalism, because there is no non-redundant causal role for non-physical properties if the physical world is causally closed. The principle of physical causal closure is the principle that every physical event (that has a cause) has a sufficient physical cause. This has solid empirical and methodological support (Papineau 2001). It seems implausible that some physical events are systematically overdetermined by both mental and physical causes and, in any case, overdetermination still renders the mental redundant. The final option for dualists who accept causal closure and deny overdeterminism is to deny the causal efficacy of the mental and embrace epiphenomenalism.
Russellian panpsychism avoids physicalism’s main problem because it doesn’t say that phenomenal properties are identical with or constituted by physical properties. Instead, there is a sense in which phenomenal properties constitute physical properties, as their intrinsic realizers. If physical properties are structural properties, then there is in fact an entailment from the existence of phenomenal, intrinsic properties to the existence of physical properties – if some entities with intrinsic properties exist, then they necessarily stand in some relations and form some structure (though we can’t say a priori which structure).
Russellian panpsychism avoids dualism’s main problem because this inverted constitution relation does not result in epiphenomenalism. If phenomenal properties realize physical structure, they are essential to the physical world. Causally efficacious physical properties can’t even exist without realizers or intrinsic properties. At the same time, this hypothesis leaves physical structure unchanged. Therefore, the physical world remains causally closed from the structural point of view. Russellian panpsychism can therefore respect the scientific considerations that support physical causal closure.
3. The Combination Problem
The combination problem is the problem of explaining how complex consciousness, of the kind we know from our own case, can result from the combination of simple entities with simple consciousness. Constitutive panpsychism is the view that complex consciousness is constituted by simple entities with simple consciousness, or, in other words, that complex consciousness is nothing over and above a collection of simple conscious experiences causally and/or spatiotemporally related in a certain way. This view faces a problem analogous to physicalism’s problem of the epistemic gap: it appears impossible in principle to see any entailment from the existence of entities with simple micro-consciousness – in any purely spatiotemporal or causal relation – to the existence of complex macro-consciousness. This can be illustrated by panpsychist versions of both the conceivability argument and the knowledge argument – think of micro-phenomenal zombies (Goff 2009) or a physically and micro-phenomenally omniscient Mary (Chalmers 2016).
Emergent panpsychism is the view that complex consciousness is distinct from micro-consciousness causally and/or spatio-temporally related in a certain way, but is rather causally generated by micro-conscious configurations. This view faces a problem of macro-mental causation analogous to dualism’s problem of mental causation: macro-consciousness appears causally and explanatorily redundant assuming that micro-consciousness already realizes all physical structure. Positing additional structure for which distinct macro-consciousness could be the realizer would bring emergent panpsychism into conflict with the principle of physical (or at least microphysical) causal closure.
In this way, the combination problem threatens to undermine the motivation for Russellian panpsychism. It appears to show that the view doesn’t avoid the problems of physicalism and dualism after all; it merely moves them, and has them reappear in a different guise.
4. IIT and the Combination Problem
IIT is a theory with both empirical and metaphysical aspects. On the one hand, it proposes an empirical correlate of consciousness: all and only conscious systems are maxima of Φ. On the other hand, it aims to explain why this correlation holds. Tononi argues that the correlation follows a priori from a set of phenomenological axioms, truths about consciousness which are self-evident upon introspection (Tononi, Albantakis, and Oizumi 2014).
The correlational claim, that maximum Φ is the correlate of consciousness, is empirically verifiable in principle. Although Φ can be precisely calculated for complex systems only in theory; insofar as it can be estimated, it predicts which kinds of alterations in the brain would lead to a loss of consciousness (namely those that reduce the Φ of a brain area below the Φ of its parts). It’s also verifiable in terms of how well it accounts for pre-existing data about consciousness. IIT accounts for why an area of the brain such as the cerebellum is not conscious, even though it has more neurons than the cerebrum; why epileptic seizures, as well as deep sleep, lead to loss of consciousness even though neural activity remains high or increases; and a number of other facts that puzzle neuroscientists. Other proposed correlates of consciousness, such as activation in specific brain areas, electrical activity within some particular range, or the P300 wave, all seem to be absent in some clearly conscious states (Tononi and Koch 2015: 3-4).
The explanatory claim, that the correlation follows a priori from phenomenological axioms, is supported by appeal to philosophical argument only. If the explanatory claim is correct, then IIT would solve the combination problem on its own. The a priori connection would eliminate all epistemic gaps, and the Exclusion postulate eliminates the problem of causal exclusion (this will be discussed in more detail below). The philosophical case for the explanatory claim is, however, somewhat contentious. It’s worth considering, therefore, whether the empirically verifiable, purely correlational claim could support a solution to the combination problem without commitment to the explanatory claim. I will suggest two ways in which the correlational claim can be combined with metaphysical approaches to combination that have been developed independently of IIT.
The phenomenal bonding view (Goff forthcoming) is a proposal for a solution to the combination problem for constitutive panpsychism. According to this view, some physical (i.e., spatial, temporal and/or causal) relations have their own intrinsic nature which is not reducible to the intrinsic properties of their relata. Physics is silent about the intrinsic nature of these relations, just as it is silent about the intrinsic nature of physical relata. Via introspection, we have access to the intrinsic nature of some physical relata (i.e., the intrinsic, phenomenal properties of our own brains), but we don’t have access to the intrinsic nature of physical relations. According to the phenomenal bonding view, if we knew the intrinsic nature of the kind of relation that connects the particles in our brain, we would see that the existence of macro-consciousness is indeed entailed by the existence of micro-conscious thus related.
One of the main challenges for the phenomenal bonding view is to specify what physical relation the phenomenal bonding relation might correspond to (Chalmers 2016). The phenomenal bonding relation must be a physical relation which holds within all and only systems where it’s plausible that mental combination takes place. Goff suggests that the spatial relation is the phenomenal bonding relation. This would entail that all phenomenal properties which are spatially related are also co-conscious – experienced together from a single subjective point of view – so that every thinkable object (group of particles) is conscious (in the first sense). Most panpsychists reject this kind of universalism – they want to restrict unified consciousness to just some kinds of systems, including brains and other organs (or organisms), and perhaps also lower level natural objects like minerals or molecules, and/or certain kinds of computers or robots, etc. There is no obvious candidate for a physical relation which gives this result.
IIT tells us what the phenomenal bonding relation could be. If maximal Φ is the correlate of unified consciousness, then the phenomenal bonding relation would be equivalent to a certain kind of causal relation: causal relations that relate elements within systems that have more Φ than overlapping systems. One might wonder why the intrinsic nature of causal relations would depend on whether or not they are part of systems with maximal Φ. But the most important thing for the phenomenal bonding view is that there exists some natural, or at least not wildly disjunctive and ad hoc, kind of physical relation, which does not entail universalism, and that could possibly be identified as the phenomenal bonding relation.
The fusion view (Seager 2010, forthcoming; Mørch 2014) is an approach to the combination problem for emergent panpsychism. According to this view, the emergence of macro-consciousness is not synchronic – emergent macro-consciousness doesn’t exist on top of, and at the same time as, underlying micro-consciousness. Rather, the emergence is diachronic – emergent macro-consciousness succeeds and replaces the set of micro-phenomenal properties that gives rise to it. The generation of macro-consciousness happens as fusion: under certain conditions, micro-phenomenal properties fuse or blend together to form a new macro-conscious unity. The fusion view avoids the problem of macro-mental causation: if macro-consciousness replaces micro-consciousness (i.e., micro-consciousness transforms into macro-consciousness), it’s the sole candidate realizer for the physical properties with which it is correlated. It’s hard to see how a synchronic emergence view can get around the same problem.
The main challenge for the fusion view is to find a physical correlate of mental fusion. Seager (2010, forthcoming) claims that quantum entanglement and the formation of black holes are physical examples of fusion. In both entangled systems and black holes, the constituents lose their individuality and are supplanted by an irreducible new unity – a “large simple”. But nothing like this seems to be going on in the brain. I have argued (Mørch 2014) that mental fusions don’t need to be partless simples, they only need to be wholes which are metaphysically prior to their parts. Fusions also don’t need to look like unities from the physical point of view in order to have this metaphysical structure, although it would be somewhat odd if they didn’t.
IIT is in itself a kind of fusion view. According to the Exclusion postulate, only maxima of Φ are conscious, hence whenever a complex system becomes conscious by gaining maximal Φ, its parts will lose micro-consciousness (in the first sense; but they will remain conscious in the third sense, as defined in the introduction). Φ measures not only information but also integration, and integration is a form of unity. IIT thus defines a sense in which physical systems can be more unified than their parts, and thereby be said to be metaphysically prior to their parts, as fusions should be. Now, one might wonder whether fusions shouldn’t ideally be characterized by even stronger markers of unity. But the most important thing for the fusion view is that there is some property which could serve to identify fusions from the physical point of view, which is empirically detectable, but does not entail any revisionary physical hypotheses (such as quantum entanglement at the brain level, or other kinds of strong physical emergence).
5. The Coarse-Graining Problem
Unfortunately, there is a problem which gets in the way of any solution to the combination problem in terms of IIT: the coarse-graining problem. This problems stems from the way IIT identifies the spatiotemporal grain of consciousness.
IIT not only selects a set of elements which regarded as a system gives maximal Φ. It also selects a spatiotemporal grain (Hoel, Albantakis, and Tononi 2013). One might regard the brain (or brain area that correlates with consciousness) as a system consisting of elements at the smallest spatial grain – as a system of subatomic particles at the nanometer scale or below. Calculating on this basis will result in certain value Φ. But one might also regard the brain as a system consisting of elements at a coarser spatial grain – such as a system of neurons or groups of neurons. This will result in a much higher value of Φ. One can also vary the temporal grain. One can regard the system as proceeding from state to state at the rate of a nanosecond, and calculate how much integrated information the system has about the immediate past and future nanosecond. Or one can regard each state as lasting for milliseconds, and calculate how much integrated information each milliseconds-long state has about the immediate past and future milliseconds long states. For the brain, a coarser temporal grain (most likely corresponding to the time it takes for neurons to fire) will result in much higher Φ.
If a maximally integrated system exists at a certain spatiotemporal grain, the structure below that grain won’t matter to the quality of the experience of that system (Tononi and Koch 2015: footnote 9). For this reason, IIT allows for multiple microphysical realizability of systems with identical consciousness. If the Φ of the brain peaks at the spatial grain of neurons, it won’t matter to the character of its experience whether the microstructure of the neurons is carbon-based or silicon-based as long as they function identically at the neuron level.
This leads to a conflict with Russellian panpsychism, as follows. Every major theory of causation entails that the causal structure of the world is determined by the intrinsic properties of things together with the laws of nature. According to the causal powers view (Shoemaker 1980; Ellis 2001; Bird 2007), the causal structure of the world is determined by the intrinsic, or intrinsically grounded, causal powers (or dispositions) of individual things (these powers in turn determine the laws). According to realism about laws (Dretske 1977; Tooley 1977; Armstrong 1978), the causal structure of the world is determined by irreducible governing laws, and it’s natural to think that the laws apply to things in virtue of their intrinsic properties. According to the regularity theory (Hume 1739-40; Lewis 1973), the causal structure of the world is brute, i.e., not determined or explained by anything else. But it is constituted by regularity relations (across the actual or the closest possible worlds) between things with intrinsic, qualitative properties.
Russellian panpsychism claims that the intrinsic properties of things are all phenomenal properties. Combined with any of the above views about causation, it follows from this that physical structure at least nomologically supervenes on phenomenal properties. That is to say, given that the laws of nature (whether irreducible, derived from intrinsic powers, or reducible to contingent regularities) remain the same, then how things behave and what relational structures they thereby form, is determined by what they are intrinsically like, i.e., their phenomenal properties.
If physical structure nomologically supervenes on the phenomenal properties of things, then there can’t be systems with different physical structure but identical phenomenal properties, given the same laws. But according to IIT, neuron and silicon brains, which would have different microphysical structure, could have identical phenomenal properties, given the same laws. It follows that Russellian panpsychism is false – physical structure doesn’t nomologically supervene on phenomenal properties.
One might think the conflict can simply be resolved by saying that although silicon brains and organic brains have identical macrophenomenal properties, they have different microphenomenal properties. Silicon neurons (or molecules/atoms) would have silicon-type micro-experience, while the organic neurons (or molecules/carbon atoms) would have different carbon-type micro-experience, while both constituting the same human macro-experience. The identical human macro-experiences realize, or nomologically determine, identical coarse-grained macrophysical structure, but the different neuron (or molecule/atom) experiences will realize the different microphysical structures.
But this is ruled out by IIT’s Exclusion postulate, according to which consciousness never overlaps – only the system with maximal Φ is conscious. Neurons (or molecules/atoms) can’t have different experiences if they are excluded from having experiences of their own in the first place. Hence, IIT is not compatible with Russellian panpsychism in view of the coarse graining principle and the Exclusion postulate.
It might seem, then, that the best way of rendering IIT and Russellian panpsychism compatible would be to abandon one of these principles. But abandoning either leads to both philosophical and empirical problems.
6. Abandoning Exclusion
Rejecting Exclusion leads to a prima facie very undesirable proliferation of consciousness (but see Schwitzgebel 2015 for an argument for abandoning Exclusion in spite of this). Nations, the internet, the galaxy and the universe all have non-zero, but non-maximal, Φ and would therefore be conscious without Exclusion. There would also be multiple overlapping consciousnesses within the brain. Not only would each neuron be conscious, there would also be one consciousness associated with almost any combination of neurons (the brain, the brain minus one neuron, minus two neurons, minus two other neurons, and so on). In other words, we get a serious case of the problem of the many. For those who are motivated to combine IIT with Russellian panpsychism in order to avoid universalism with respect to phenomenal bonding, this consequence should be unacceptable.
Rejecting Exclusion is also empirically problematic for IIT. As mentioned, the empirical case for IIT is largely based on data about when our (waking or dreaming) consciousness disappears and reappears according to reports, such as deep sleep, general anesthesia, seizures and coma. These states don’t leave the brain with zero Φ – our consciousness disappears (i.e., disintegrates into multiple lesser consciousnesses) long before Φ reaches zero. What explains this? Officially, IIT is not entirely clear on this point. One possible explanation is that low, non-maximal but non-zero, Φ still correlates with some dim form of consciousness, but not with memory or reportability. But this seems somewhat ad hoc – why would it necessarily be the case? It could also be very worrisome if true about anesthesia! An alternative, more systematic explanation would be to appeal to the Exclusion postulate: our consciousness disappears because the Φ of the normally conscious brain area goes below the Φ of neurons, neuron groups, or other subsystems that constitute it. Given Exclusion, these subsystems will become individually conscious instead (each to a much lesser degree) because they are now the maxima. It’s not clear whether the Φ of sleeping or anesthetized brain areas is actually lower than the Φ of any subsystems – given the practical difficulties with measuring Φ precisely, this is still an open empirical question. But by abandoning Exclusion, the simple and elegant explanation based on this conjecture is in any case ruled out. In view of these problems, abandoning Exclusion in order to render IIT compatible with Russellian panpsychism looks like it won’t be worth the costs.
7. Abandoning Coarse-Graining
Instead of Exclusion, one could consider abandoning coarse-graining. This would mean that when calculating the Φ of a system, one should always regard it as a system of fundamental particles. If so, silicon and organic brains would have different experiences, on which their different micro-structures could nomologically supervene.
But abandoning coarse-graining also leads to both philosophical and empirical problems. Firstly, when we look at our own experience, it appears, phenomenologically, to have a certain spatiotemporal grain. The structure of our experience appears to be much more coarse-grained than the microphysical structure of our brain. At any moment, we don’t seem to experience as many distinctions within our experience as there are particles in our brain. It also seems that our experiences last much longer than the shortest microphysical events in the brain. Experiences seem to last for at least some milliseconds, but microphysical events in the brain last nanoseconds or less.
This is the basis for the so-called grain problem in philosophy of mind known from Lockwood (1993) and Sellars (1965, 1971): the grain of experience doesn’t correspond to the microphysical grain of the brain. The conclusion often drawn from the grain problem is that human consciousness needs an irreducibly macrophysical correlate, which is precisely what it gets with IIT’s coarse-graining principle, but loses without it.
Secondly, rejecting coarse-graining gives an empirical problem. As noted, if we regard the brain as a system of particles, it will have lower Φ than if we regard it as a system of neurons. And it will also have lower Φ than individual neurons, molecules or atoms in the brain. Therefore, without coarse-graining, neurons, molecules or atoms will be conscious, not the brain (or any extended brain area) as a whole. This would result in billions (or more) other consciousnesses in our brains, each of which as highly conscious as our own. This seems very implausible (as another version of the problem of the many, although in this case the many don’t spatially overlap) and doesn’t fit any natural interpretation of empirical data. With coarse-graining, in contrast, Φ will peak only in a few locations, and the highest peak would be much higher than the others, so there will only be one most advanced consciousness in the brain.
Hence, abandoning coarse-graining seems out of the question for IIT. What I will now suggest, is that in order to render IIT compatible with Russellian panpsychism, there is no need to wholly abandon either the coarse-graining or the Exclusion postulate. Instead, it would suffice to merely modify either one of the principles, in ways I will now describe.
8. Modifying Exclusion
One way of avoiding the coarse-graining problem is to modify the Exclusion postulate by relativizing it to spatiotemporal grains. Instead of saying that consciousness never overlaps, IIT could say that (1) consciousness never overlaps within the same spatiotemporal grain, and (2) there is consciousness only at grains where Φ is higher than any other level below. This would entail that the brain, as a coarser-grained system, will have overlapping consciousness at the finest, microphysical grain (since there is no level below, clause (2) is vacuously satisfied by the finest grain). In two brains with different fine-grained parts, such as carbon and silicon atoms, these parts will have qualitatively different consciousness. Differences in microphysical structure could then nomologically supervene on these differences in microconsciousness.
This relativized Exclusion postulate avoids the problems that result from wholly abandoning it. It won’t prevent human consciousness from disappearing in deep sleep, general anesthesia and so on before Φ goes to zero. Human consciousness would disappear when the Φ of its correlate brain area goes lower than the Φ of individual neurons or neuron groups, because Exclusion is still in effect within this grain. If the brain has maximal Φ all grains considered, there will be no internet, galaxy or cosmic consciousness. The internet, the galaxy or the cosmos will not have higher Φ than the brain if we regard them as systems made of neurons and other entities at the grain of the brain. Nor will they have higher Φ than the brain if we regard them as systems made of coarser-grained parts such as computers or stars (but if they did, they could have consciousness without threatening to eliminate human consciousness via Exclusion). Below the brain level, there could be grains above the subatomic that increase Φ before the grain of the brain, resulting in more than one level of microconsciousness, but there won’t be arbitrarily or absurdly many such levels.
This modification of Exclusion is only compatible with constitutive Russellian panpsychism. If microphenomenal and macrophenomenal properties are entirely distinct, as per the emergent view, then micro-phenomenal properties at lower grains would causally exclude macro-phenomenal properties at higher levels with maximal Φ. However, the idea that coarse-grained phenomenal properties are constituted by overlapping fine-grained phenomenal properties seems to lead to another problem. Say I have an experience of a tiny patch of phenomenal green, the smallest patch of green that is possible given the grain of my experience. This green patch will be completely uniform green for me, because my experience can’t contain any complexity below its own smallest grain. Now, this uniform green patch will actually be constituted by micro-phenomenal properties at a smaller grain, for example even smaller patches of blue and yellow that are the experiences of the particles in my brain (or some other kinds of particle experiences that we can’t imagine). But then it seems my experience of green is both uniform and partless, and complex, made of blue and yellow (or at least non-green) parts, at the same time.
Many Russellian panpsychists, and non-reductionists about consciousness in general, are motivated by the principle that there is no appearance/reality-distinction for phenomenal properties (if an experience appears to be of green, then it is of green, phenomenally speaking). If so, it would be contradictory to say that the same phenomenal properties can appear either uniform green or complex yellow/blue (or otherwise non-green) depending on the grain. Hence, this way of rendering IIT and Russellian panpsychism comes at the cost of introducing a kind of appearance/reality distinction for phenomenal properties, and the consequences of this would need to be examined in more detail in order to see if that could possibly be acceptable.
9. Modifying Coarse-Graining
My second suggestion is to modify the coarse-graining principle, in a way that starts from noticing a certain tension between two aspects of the combination problem.
The grain problem, as already mentioned, says that macro-consciousness has too little structure to reflect the full microphysical structure of our brains. But there is also an aspect of the combination problem, known as the palette problem (Chalmers 2016: section 7). According to this problem, macro-consciousness has too many qualities. In physics, we find a limited number of fundamental particles (about 17, according to the standard model). This suggests a correspondingly limited number of basic micro-phenomenal qualities. In our experience, however, we find what appears to be an endless number of different phenomenal qualities (colors, sounds, emotions and so on). It’s hard to see how all these qualities can result (without radical emergence) just from combining a small number of basic micro-phenomenal qualities in different ways.
If macro-consciousness has too little structure, but too many qualities, to reflect its microphysical correlate, this suggest that the missing structure is somehow encoded in the extra qualities – that microphysical structure is reflected in macrophenomenal qualities as opposed to macrophenomenal structure. Instead of having one little distinct patch in our experience corresponding to every particle in the brain, perhaps a single coarse-grained quality (like a field of color, or an emotion) reflects complex microphysical structure in virtue of the many complex ways in which it is different from other possible macrophenomenal qualities.
This would result in a modified coarse-graining principle for IIT: information below the grain where Φ peaks matters to the quality, but not the structure, of experience. In other word, macrophenomenal structure will be multiply microphysically realizable, but macrophenomenal qualities will not be. If your organic neurons were to be replaced by functionally identical silicon neurons, then your experience would have the same (contrastive and motivational) structure, but different qualities (i.e., a shifted spectrum). For example, for the silicon consciousness, tomatoes and grass will be experienced as having color-like qualities we cannot imagine, but they will be as different from each other as red and green are for us, and they will have pain and pleasure-like qualities which feel different from ours, but they still feel equally bad and good. The fact that silicon and organic brain have the same macrophenomenal structure could account for why these systems will show the exact same behavior at the macrolevel. That they have different, shifted qualitative spectra could perhaps account for why they have different micro-level dispositions (why they respond differently to microscopic measurement and decompose in different ways).
Many questions remain as to precisely how microphysical structure can be encoded or grounded in unstructured qualities. I have only shown that there is some extra qualitative information in one place that can be connected to missing structural information in another place, and there is nothing that immediately rules out that there is some way of translating one to the other. In this way, it no longer looks impossible for IIT and Russellian panpsychism to be combined.
I have suggested two ways of solving the coarse-graining problem and rendering IIT and Russellian panpsychism compatible. These suggestions involve substantive modifications of some basic principles of IIT, either the Exclusion postulate or the coarse-graining principle. With one of these modifications, IIT would support (significant progress towards) a solution to the combination problem for Russellian panpsychism – either on its own, in view of the explanatory claim according to which the principles of mental combination are a priori deducible from phenomenological axioms, or on the basis of its correlational claim only, in combination with either the phenomenal bonding view or the fusion view of mental combination.
Both my suggested modifications come at a price. The suggested modification of the Exclusion postulate is in tension with the view that phenomenal qualities necessarily are as they appear. The suggested modification of the coarse-graining principle assumes mysterious links between qualities and structure. Perhaps the mystery of the latter proposal is preferable to the tension resulting from the former. A modification of the Exclusion postulate might also be more at odds with Tononi’s own philosophical case for IIT.
IIT has many philosophical aspects which have not been discussed here – as noted, I have set the theory’s central philosophical claims mostly aside and mainly considered the significance of its in principle empirically verifiable correlational claim, i.e., the claim that maximal Φ is the correlate of consciousness. IIT as a whole should be subjected to closer philosophical scrutiny if the theory is to fully support a solution to the combination problem. On the one hand, one might find even more sources of tension with Russellian metaphysics within IIT. On the other hand, one might discover other aspects of IIT that could contribute to even further advancing the case for a panpsychist solution to the mind-body problem.
Alter, Torin, and Yujin Nagasawa. 2012. What Is Russellian Monism? Journal of Consciousness Studies 19 (9-10): 67-95.
Armstrong, David Malet. 1978. A Theory of Universals. Universals and Scientific Realism Volume Ii. Cambridge University Press.
Bird, Alexander. 2007. Nature’s Metaphysics: Laws and Properties. Oxford: Clarendon Press.
Cerullo, Michael A. 2015. The Problem with Phi: A Critique of Integrated Information Theory. PLoS Comput Biol 11 (9): e1004286.
Chalmers, David J. 1996. The Conscious Mind: In Search of a Fundamental Theory. New York: Oxford University Press.
———. 2003. Consciousness and Its Place in Nature. In Blackwell Guide to Philosophy of Mind, eds. S. P. Stich and T. A. Warfield. Malden, MA: Blackwell.
———. 2009. The Two-Dimensional Argument against Materialism. In Oxford Handbook to the Philosophy of Mind, eds. B. P. McLaughlin and S. Walter. Oxford: Oxford University Press.
———. 2015. Panpsychism and Panprotopsychism. In Consciousness in the Physical World: Essays on Russellian Monism, eds. T. Alter and Y. Nagasawa. New York: Oxford University Press.
———. 2016. The Combination Problem for Panpsychism. In Panpsychism: Contemporary Readings, eds. G. Brüntrup and L. Jaskolla. Oxford: Oxford University Press.
Dretske, Fred I. 1977. Laws of Nature. Philosophy of Science 44 (2): 248-268.
Ellis, B. D. 2001. Scientific Essentialism. Vol. 113. Vol. 450 Cambridge University Press.
Goff, Philip. 2006. Experiences Don’t Sum. Journal of Consciousness Studies 13 (10-11): 53-61.
———. 2009. Why Panpsychism Doesn’t Help Us Explain Consciousness. Dialectica 63 (3): 289-311.
———. forthcoming. The Phenomenal Bonding Solution to the Combination Problem. In Panpsychism, eds. G. Brüntrup and L. Jaskolla. Oxford: Oxford University Press.
Hoel, Erik P., Larissa Albantakis, and Giulio Tononi. 2013. Quantifying Causal Emergence Shows That Macro Can Beat Micro. Proceedings of the National Academy of Sciences 110 (49): 19790-19795.
Hume, David. 1739-40. A Treatise of Human Nature.
Jackson, Frank. 1982. Epiphenomenal Qualia. Philosophical Quarterly 32 (April): 127-136.
———. 1986. What Mary Didn’t Know. The Journal of Philosophy 83 (5): 291-295.
James, William. 1890. The Principles of Psychology. Vol. 1. London: Macmillan.
Kim, Jaegwon. 1988. Explanatory Realism, Causal Realism, and Explanatory Exclusion. Midwest Studies in Philosophy 12 (1): 225-239.
Koch, Cristof. 2012. Consciousness: Confessions of a Romantic Reductionist. Cambridge, MA: MIT Press.
Kripke, Saul A. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press.
Ladyman, James, and Don Ross. 2007. Every Thing Must Go: Metaphysics Naturalized. Oxford: Clarendon Press.
Lewis, David. 1973. Causation. Journal of Philosophy 70 (17): 556-567.
Lockwood, Michael. 1993. The Grain Problem. In Objections to Physicalism, ed. H. M. Robinson Oxford University Press.
Martin, C. B., and John Heil. 1999. The Ontological Turn. Midwest Studies in Philosophy 23 (1): 34-60.
Mørch, Hedda Hassel. 2014. Panpscyhism and Causation: A New Argument and a Solution to the Combination Problem (Doctoral Dissertation), Departement of Philosophy, Classics, History of Art and Ideas, University of Oslo, Oslo.
Nagel, Thomas. 1979. Panpsychism. In Mortal Questions. Cambridge: Cambridge University Press.
Papineau, David. 2001. The Rise of Physicalism. In Physicalism and Its Discontents, eds. C. Gillett and B. Loewer. Cambridge: Cambridge University Press.
Russell, Bertrand. 1927. The Analysis of Matter. London: Kegan Paul, Trench, Trubner & Co.
Schwitzgebel, Eric. 2015. If Materialism Is True, the United States Is Probably Conscious. Philosophical Studies 172 (7): 1697-1721.
Seager, William. 1995. Consciousness, Information and Panpsychism. Journal of Consciousness Studies 2 (3): 272-288.
———. 2006. The ‘Intrinsic Nature’ Argument for Panpsychism. Journal of Consciousness Studies 13 (10-11): 129-145.
———. 2010. Panpsychism, Aggregation and Combinatorial Infusion. Mind and Matter 8 (2): 167-184.
———. forthcoming. Panpsychist Infusion. In Panpsychism: Contemporary Readings, eds. G. Brüntrup and L. Jaskolla. Oxford: Oxford University Press.
Searle, John R. 2013. Can Information Theory Explain Consciousness? New York Review of Books 60 (1).
Sellars, Wilfrid S. 1965. The Identity Approach to the Mind-Body Problem. Review of Metaphysics 18 (March): 430-51.
———. 1971. Seeing, Sense Impressions, and Sensa: A Reply to Cornman. Review of Metaphysics 24 (March): 391-447.
Shoemaker, Sydney. 1980. Causality and Properties. In Time and Cause: Essays Presented to Richard Taylor, ed. P. van Inwagen. Dordrecht: Reidel.
Strawson, Galen. 2006. Realistic Monism: Why Physicalism Entails Panpsychism. Journal of Consciousness Studies 13 (10-11): 3-31.
———. 2008. The Identity of the Categorical and the Dispositional. Analysis 68 (4): 271-282.
Tegmark, Max. 2016. Improved Measures of Integrated Information. arxiv.org arXiv:1601.02626.
Tononi, Giulio. 2004. An Information Integration Theory of Consciousness. BMC neuroscience 5 (1): 42.
———. 2008. Consciousness as Integrated Information: A Provisional Manifesto. The Biological Bulletin 215 (3): 216-242.
———. 2011. Integrated Information Theory of Consciousness: An Updated Account. Archives italiennes de biologie 150 (2-3): 56-90.
Tononi, Giulio, Larissa Albantakis, and Masafumi Oizumi. 2014. From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0. PLoS Comput Biol 10 (5).
Tononi, Giulio, Melanie Boly, Marcello Massimini, and Christof Koch. 2016. Integrated Information Theory: From Consciousness to Its Physical Substrate. Nature Reviews Neuroscience.
Tononi, Giulio, and Christof Koch. 2015. Consciousness: Here, There and Everywhere? Philosophical Transactions of the Royal Society of London B: Biological Sciences 370 (1668): 20140167.
Tooley, Michael. 1977. The Nature of Laws. Canadian Journal of Philosophy 7 (4): 667-98.
 Consciousness does not require the capacity for, e.g., thought, abstraction or self-awareness. These can be regarded as advanced, complex forms of consciousness and reserved for advanced, complex physical systems. Therefore, panpsychism does not entail that particles think or are self-conscious.
 I.e., how much can you know about the previous state of the system only by looking at its present state, given that all external influences are fixed (given that you know all the relevant causal laws)? And how much can you know about the next state of the system, fixing external influences? Or, in other words, to what extent do the causes and the effects of a system reside within the system itself?
 I.e., how much information would you lose by cutting the system in half? Computers, of the kind we make today, might in principle have the same quantity of information as human brains. But in a computer, each transistor is connected to just a few other transistors. In the brain, the neurons are connected to a vast number of other neurons. If you cut a brain in half, you are therefore going to lose a lot more information than if you cut a computer in half. This shows that the brain has a much higher level of integrated information (Φ) and therefore more consciousness. (In fact, according to IIT’s Exclusion postulate, the computer will probably not be conscious, in the first sense, at all, only its parts will be, because its Φ will probably not be high enough to supersede the Φ of its parts.) For a more detailed introduction to IIT, see, e.g., Tononi et al. (2016) and Tononi and Koch (2015). For a full description of the latest version of the theory, see Tononi, Albantakis, and Oizumi (2014).
 It should be noted that simple fundamental particles, such as quarks, can only (unless they turn out to have internal structure) be conscious only in the third sense given IIT. Since they have no complexity, they will have zero Φ. But quarks are never found in isolation, they always form part of structured particles (e.g., protons) which have some Φ and hence some consciousness.
 Tononi and Koch (2015) are careful to distinguish IIT’s form of panpsychism from the form of panpsychism that says that all things, including artifacts and aggregates, are conscious in the first sense, as opposed to in just the second or third sense.
 The most common objection to panpsychism is probably “the incredulous stare”. The most serious objection is the combination problem.
 Dualist panpsychism would be the view that every physical thing is connected with a mental substance or has wholly non-physical mental properties via fundamental psychophysical laws of nature. Physicalist panpsychism would be the view that consciousness is to be reductively identified with a ubiquitous physical property (such as integrated information understood in a purely functionalist way).
 The entailment also goes the other way around. If we accept that structure needs intrinsic realizers, and that phenomenal properties are the only intrinsic properties we know, then there is an entailment from the existence of physical properties to the existence of either phenomenal properties or unknown properties of a similar nature (non-panpsychist Russellian monists prefer the latter alternative).
 Perhaps outright excluded: overdeterminism is perhaps not even an option for emergent panpsychism, because one structural property can’t have two distinct mental realizers (micro-phenomenal properties plus an emergent macro-phenomenal property).
 The cerebellum has low Φ in spite of having many neurons (Tononi 2008: 221); under epileptic seizures and deep sleep Φ strongly decreases even though activity increases or remains high (Tononi 2008: 223).
 Some complain that the alleged phenomenological axioms are less than self-evident (Cerullo 2015). One might also complain that it’s not self-evident how to translate the phenomenological axioms into empirical postulates. There might also be multiple ways to translating the empirical postulates into a mathematical formula (Tegmark 2016). Chalmers (2016) claims that IIT is best seen as a providing a fundamental, a posteriori law of nature specifying when combination occurs, but he does specify the precise point he thinks Tononi’s a priori justification fails.
 Chalmers notes (2016: section 6.3) that relations along the lines of IIT could be the phenomenal bonding relation, but objects that these relations are derivative and not fundamental. He claims that a solution to the combination problem in terms of phenomenal bonding and something like IIT get a new combination problem: how do fundamental causal relations (proto-bonding relations) become phenomenal bonding relations when they take on IIT-like character? A phenomenal bonding theorist could perhaps respond by renewed appeal to our ignorance of the intrinsic nature of physical relations: if we knew the intrinsic nature of causal relations, we would see how iworks.
 Recall, high integration means that the parts of a system are highly causally interconnected. In this way, it’s bound together as a unit.
 Tononi claims that maximal Φ corresponds to maximal causal power. This would give fusions a very strong form of metaphysical priority, without strong emergence of the kind that violates microphysical causal closure. But the view that Φ corresponds with causal power depends on a controversial metaphysical view about causation (Hoel, Albantakis, and Tononi 2013).
 Insofar as things are taken to have intrinsic properties at all – theories such as structural realism (Ladyman and Ross 2007) deny this, but this is incompatible with both Russellian panpsychism and IIT which take consciousness to be intrinsic.
 The causal powers view would not be compatible with Russellian panpsychism, which takes phenomenal properties to be categorical (i.e., non-dispositional), if dispositions are regarded as having no categorical aspect or as being absolutely fundamental, i.e. not grounded in intrinsic categorical properties. But Russellian panpsychists are free to adopt the view that dispositions have a categorical aspect (Martin and Heil 1999; Strawson 2008) or to regard dispositions as intrinsic in virtue of being grounded in intrinsic categorical properties/aspects.
 Armstrong explicitly holds that laws of nature consist in relations between categorical, i.e., intrinsic and non-dispositional, universals. It’s hard to say how laws could apply to things in virtue of their extrinsic properties, since most extrinsic properties would be determined by the laws, which means things could not possess them unless the laws already apply. Laws could perhaps apply to things in virtue of their bare spatiotemporal locations, but this seems empirically false, if not also hard to make metaphysical sense of.
 Tononi and Koch (2015: 13) gesture toward such a view.
 Although one might think that consciousness during anesthesia is not painful, or if it is, that it’s just very dimly painful and therefore nothing to worry about – but none of this is obvious.
 At least usually – according to IIT, split brain (severed corpus callosum) patients will have one consciousness in each hemisphere, with almost the same Φ. It’s also possible that our consciousness temporarily splits into two in cases such as automatic (absent-minded) driving.
 According to Russellian panpsychism, there is a sense in which the mental realizes the physical. But there is also realization within the physical, in another sense of the term. To say that macroconsciousness in multiply realizable given Russellian panpsychism would really be to say that the macrophysical correlate of macroconsciousness is multiply realizable physically speaking, and while macroconsciousness realizes the macrophysical correlate metaphysically speaking.
 Many thanks to Erik Hoel and Giulio Tononi for helpful discussion of the issues raised in this paper, and for clarifying and explaining relevant aspects of IIT. I also thank all the participants of the workshop IIT: Foundational Issues (New York University, November 2015) for their comments.