Panpsychism Combination Problem: Can Tiny Experiences Make a Mind?

The panpsychism combination problem asks how billions of tiny experiences, if electrons and quarks really have them, could ever add up to one person who sees red, hears a cello and worries about rent. Panpsychism says that experience is a basic feature of matter rather than a late arrival in evolution. That move dodges the classic mystery of how consciousness springs from wholly non-conscious stuff. It buys a new one.

This article sets out what has been settled, compares three families of proposed answers (phenomenal bonding, cosmopsychism, and the integration-and-exclusion approach borrowed from integrated information theory), and then maps the argument premise by premise to show exactly where each answer pays its bill. A short simulation built for this article makes one of those bills concrete. Nothing here resolves the problem. The goal is a clear map of where the open ground lies.

Why the question is still open

Panpsychism remains a minority position. In the PhilPapers 2020 survey of professional philosophers, 7.55 percent of the 1,020 people who answered the consciousness question accepted or leaned toward panpsychism, against 33.04 percent for functionalism (Bourget and Chalmers, PhilPapers Survey 2020). That is small but not negligible, and the view has drawn serious defenders in the past two decades precisely because the alternatives look stuck. Readers who want the background first will find it in our plain guide to the hard problem of consciousness.

Nearly every critic and every defender agrees that combination is the view’s weakest point. David Chalmers states the target compactly:

“The combination problem for panpsychism is: how can microphenomenal properties combine to yield macrophenomenal properties?”

Chalmers, 2016

Outside the journals, the puzzle surfaces in ordinary discussion too. On The Philosophy Forum, a thread titled “Counter Argument for The Combination Problem for Panpsychism” drifts quickly to the obvious follow-ups: does a part’s consciousness persist once it is pooled, and what triggers merging, mere proximity or some functional link? Those are the right questions. They are also the ones the professional literature has not answered.

Why should anyone outside philosophy care? Because any theory that assigns experience to physical systems, including theories now used to argue about animal minds, brain-injured patients and machines, needs a rule for when many experiencers count as one. Without such a rule, the theory cannot say how many minds are in a room.

What is established

Very little is settled about consciousness itself, but several things about this particular problem are well documented.

• The problem is old. William James attacked the “mind-stuff” theory in chapter 6 of The Principles of Psychology (1890) with his image of twelve men each thinking one word of a sentence; no one, he argued, thereby thinks the whole sentence (James 1890). Itay Shani later argued that James’s critique does not actually establish a combination problem for panpsychism (Shani 2010), so even the historical starting point is contested.
• The name is recent. According to the Stanford Encyclopedia of Philosophy, the phrase comes from William Seager’s 1995 paper in the Journal of Consciousness Studies (Allen-Hermanson 2022).
• It is several problems, not one. Chalmers (2016) separates a subject problem (how do micro-subjects make a macro-subject?), a quality or palette problem (how can a small set of micro-qualities yield the huge variety we experience?), and a structure problem (brain structure looks nothing like the structure of experience). Most of the literature, and this article, concentrates on the subject version, which is widely regarded as the hardest.
• The core argument has a standard form. Chalmers’s “subject-summing” argument runs: if constitutive panpsychism is true, the existence of certain micro-subjects necessitates a distinct macro-subject; but no collection of subjects ever necessitates a further distinct subject; so constitutive panpsychism is false.

Two empirical-adjacent facts also matter. Integrated information theory (IIT), a leading scientific theory of consciousness, explicitly includes a rule for deciding which of many overlapping systems is conscious, and its authors present that rule as an answer to combination (Tononi and Koch 2015; Albantakis et al. 2023). And neuroscience gives us real cases where one mind seems to become two, most famously after the corpus callosum is cut, as discussed in our piece on split-brain patients. Those cases do not show micro-experiences combining. They do show that the number of subjects in a skull can depend on wiring.

Hypothesis one: phenomenal bonding

Philip Goff’s proposal is the most direct reply to subject-summing. It denies the argument’s second premise. Some relation, which Goff calls phenomenal bonding, holds among micro-subjects, and when it holds they jointly constitute a further subject (Goff 2016; Goff 2017). On one reading the bonding relation is what physics describes from the outside as spatial or causal relatedness. Physics, on this Russellian picture, tells us how things are structured but is silent on what they are intrinsically like.

For: it keeps panpsychism constitutive, so the human mind is nothing over and above its parts, and it avoids brute emergence. Gregory Miller (2017, Dialectica) tried to supply the positive concept that Goff admits we lack, which shows the idea can be developed rather than merely named.

Against: the relation does its work by stipulation. Goff concedes that we have no transparent grasp of what bonding is, so the combination is not intelligible from the armchair, which was supposed to be panpsychism’s advantage over physicalism. S. Siddharth argues the proposal therefore fails to answer the subject-summing argument at all (Siddharth 2021). Itay Shani and Heath Williams go further, arguing that subject combination for constitutive micropsychism is not merely unexplained but incoherent (Shani and Williams 2022).

Hypothesis two: cosmopsychism

Cosmopsychism flips the direction of explanation. The universe as a whole is the fundamental conscious subject, and human minds are aspects or parts derived from it (Shani 2015; Nagasawa and Wager 2016; Goff 2017). The inspiration is Jonathan Schaffer’s priority monism, in which the whole cosmos is metaphysically prior to its parts.

For: there is no bottom-up summing of subjects, so the subject-summing argument has nothing to bite. Yujin Nagasawa and Khai Wager suggest it also eases the structure problem, since macro-level structure no longer has to be built from micro-level structure.

Against: the problem returns upside down. Chalmers calls it the decomposition problem: how does one cosmic subject give rise to many separate, private human subjects? He also flags the plainest worry, that the view “seems to entail that there is only one conscious subject.” A later exchange in the Journal of Consciousness Studies argues the cosmopsychist’s problem differs from priority monism’s analogous puzzle about how parts get their distinct properties (Miller 2021), which suggests the monist literature cannot simply be borrowed.

Hypothesis three: integration and exclusion

IIT does not set out to rescue panpsychism; its authors say it shares the intuition that consciousness is fundamental and graded while denying that everything is conscious (Tononi and Koch 2015). Its relevance here is a rule. IIT’s exclusion postulate says a conscious substrate must be definite:

“Its cause–effect power must be definite: it must specify its cause–effect state as this whole set of units.”

Albantakis et al., 2023

In practice, among all overlapping candidate systems, only the one with maximal integrated information counts as a complex, that is, a subject. Its overlapping rivals are excluded. Tononi and Koch apply the rule to two people talking: each brain is a complex, and the pair is not, so there is nothing it is like to be two people. A heap of sand grains, likewise, never becomes a subject.

For: it is the only proposal that yields a computable, in-principle testable answer to “how many subjects are here?” Hedda Hassel Mørch (2018, Erkenntnis) has asked whether IIT can be combined with Russellian panpsychism, which suggests the two literatures can meet.

Against: strictly speaking, exclusion does not combine anything. The parts of a complex are not subjects whose experiences are pooled; they simply fail to be subjects while they are parts. That replaces combination with competition. Whether this counts as solving the problem or abandoning it is exactly the dispute. The rule also makes subjecthood depend on what surrounds a system, an odd consequence for a property meant to be intrinsic.

An argument map: where each solution pays

The original analysis for this article is a premise map. Each proposed answer keeps most of a set of individually attractive claims and gives up at least one. Listing the claims explicitly makes the trade visible.

1. P1, micro-subjects. Fundamental physical entities are subjects of experience.
2. P2, bottom-up constitution. Human subjects are wholly grounded in micro-level facts.
3. P3, intelligibility. That grounding should be intelligible in principle, not a brute law.
4. P4, no summing. No collection of subjects, by itself, necessitates a further subject.
5. P5, privacy. One subject’s experience cannot also be another subject’s experience.
6. P6, many minds. There are many distinct human subjects, with sharp enough boundaries.
7. P7, intrinsicness. Whether something is a subject depends only on what it is, not on its surroundings.

P1 through P5 jointly entail that no human subject exists, which is false. So at least one must go. The table records which premise each view rejects, and what it costs.

Two rows need comment. Sam Coleman’s move is the most radical for a panpsychist, because it denies that the ultimate constituents are subjects at all. As he puts it,

“a phenomenally-qualitied ultimate is not ipso facto a mind or subject of experience.”

Coleman, 2012

Luke Roelofs (2019, Combining Minds) takes the opposite path. He accepts that composite subjects are everywhere and argues that overlapping subjects do not add extra consciousness to the world; they are different ways of carving up the same experience.

Reading down the cost column gives the map’s main result. Every route either loses intelligibility (bonding, emergence), multiplies or redefines subjects (Roelofs, Coleman), relocates the problem (cosmopsychism), or swaps combination for selection (IIT). No proposal keeps P3, P5 and P6 together. That is a stronger claim than “the problem is hard”: the three conditions look jointly unsatisfiable under constitutive micropsychism, which fits Shani and Williams’s incoherence verdict.

A small simulation of the exclusion rule

To see what the IIT row’s cost means in practice, I wrote a toy model in Python. Six units form two clusters of three. Within a cluster the coupling is fixed; between clusters a coupling strength, c_out, is dialed up from zero. Each subset’s “integration” is the mutual information across its weakest bipartition, computed for a stationary Gaussian system. The exclusion step then keeps the highest-scoring subset, removes anything overlapping it, and repeats.

Caution belongs up front. This proxy is not IIT’s Φ, which is defined over cause-effect structure rather than static correlations, and six units are not a brain. The model shows the logic of a maximize-and-exclude rule, nothing about real neurons.

import itertools, numpy as np
N, A, B = 6, [0, 1, 2], [3, 4, 5]          # two clusters of three units

def cov(c_in, c_out):                       # Gaussian model, precision = I - K
    K = np.zeros((N, N))
    for g in (A, B):
        for i, j in itertools.permutations(g, 2): K[i, j] = c_in
    for i in A:
        for j in B: K[i, j] = K[j, i] = c_out
    return np.linalg.inv(np.eye(N) - K)

def mi(S, X, Y):                            # mutual information, nats
    ld = lambda ix: np.linalg.slogdet(S[np.ix_(ix, ix)])[1]
    return 0.5 * (ld(X) + ld(Y) - ld(X + Y))

def phi(S, sub):                            # integration across weakest cut
    return min(mi(S, list(X), [u for u in sub if u not in X])
               for k in range(1, len(sub)//2 + 1)
               for X in itertools.combinations(sub, k))

def complexes(S):                           # exclusion: non-overlapping maxima
    subs = [list(s) for k in range(2, N+1)
            for s in itertools.combinations(range(N), k)]
    chosen, used = [], set()
    for s in sorted(subs, key=lambda s: -phi(S, s)):
        if not used & set(s):
            chosen.append(s); used |= set(s)
    return chosen

for c_out in [0.0, 0.05, 0.08, 0.088, 0.090, 0.10, 0.15]:
    S = cov(0.25, c_out)
    print(f"{c_out:.3f} phi(A)={phi(S, A):.4f} phi(all)={phi(S, list(range(N))):.4f}"
          f" complexes={[len(c) for c in complexes(S)]}")
for c_in in [0.15, 0.25, 0.30]:             # where does 2 subjects -> 1 happen?
    for c_out in np.arange(0, (1 - 2*c_in)/3, 0.001):
        if len(complexes(cov(c_in, c_out))) == 1:
            print(f"c_in={c_in}: one complex from c_out={c_out:.3f}"); break

With within-cluster coupling at 0.25, the verdict flips from two subjects to one at c_out = 0.089. Repeating the scan with within-cluster coupling of 0.15 and 0.30 moved the switch to 0.062 and 0.097 respectively. Near the switch, the two candidates differ by about a thousandth of a nat, yet the answer to “how many minds?” jumps from two to one, and the two small subjects simply stop existing.

What the run shows: under any maximize-and-exclude rule, subject boundaries are knife-edged, and the integration of the parts does not disappear when they cease to be subjects. Nothing is pooled. What it does not show: that real IIT behaves identically (the true Φ calculation differs), that brains sit near such thresholds, or that any experience is present at all. It illustrates a structural cost, not a fact about consciousness.

The author’s reading, and what would settle it

Opinion, clearly labeled. My reading is that the subject version of the problem is real and has not been solved. Phenomenal bonding names the missing relation rather than explaining it. Cosmopsychism is the more elegant option, but trading a summing problem for a splitting problem looks like a sideways move until someone states a principle that individuates human minds within the cosmic one. The exclusion rule is the only proposal precise enough to be wrong, which I count in its favor, though it wins by dropping combination rather than by explaining it.

If that is right, the field’s most productive work will come less from new metaphysical relations and more from subject-counting rules that make predictions. Specific tests and projects that would move the debate:

• Graded disconnection. A maximize-and-exclude rule predicts that, as connectivity between hemispheres or brain regions is reduced, signs of a single unified subject should fail abruptly rather than fade smoothly. Careful study of partial callosal section, or of reversible pharmacological disconnection, could favor graded or step-like change. A clearly graded pattern would count against exclusion-style rules.
• A positive bonding concept. Bonding theorists could specify a relation from which a macro-subject’s existence follows a priori, and show it applied to a concrete case. If no such derivation appears, the view stays a placeholder.
• An individuation principle for cosmopsychism. A cosmopsychist rule that counts human subjects correctly, and does not also count unwanted ones, would answer Chalmers’s decomposition worry. Its absence is the view’s main gap.
• Formal tests of P3, P5 and P6. The map above suggests these three are jointly unsatisfiable for constitutive micropsychism. A rigorous proof, or a counter-model, would settle that sub-question outright.

The stakes extend past philosophy seminars. The same counting question reappears whenever someone asks whether a colony, a corporation or an AI system is one mind or many, a theme running through our comparison of the main scientific theories of consciousness.

Common questions

Does panpsychism mean rocks have thoughts?

No. Panpsychists typically claim that the most basic physical entities have some extremely simple form of experience, not thoughts or beliefs. A rock, on most versions, is an aggregate of experiencing particles rather than a single experiencer, and the combination problem is partly the problem of explaining why a brain differs from a rock in this respect.

Who first raised the combination problem?

The idea is usually traced to William James’s 1890 criticism of “mind-stuff” theories in The Principles of Psychology. The name “combination problem” comes from William Seager’s 1995 paper. Some scholars dispute whether James’s argument really targets panpsychism as it is defended today.

Is integrated information theory a form of panpsychism?

Not exactly. IIT agrees that consciousness is fundamental and comes in degrees, and it allows very simple systems a little of it, but it denies that every aggregate is conscious. Its exclusion postulate is precisely what stops a pile of conscious parts from forming a larger subject.

Could an experiment ever settle the combination problem?

Not directly, since experience in particles cannot be measured. But specific subject-counting rules, such as IIT’s, make predictions about when unified consciousness should split or merge as brain connectivity changes. Testing those predictions would at least eliminate some candidate solutions.

Further reading

• Galileo's Error: Foundations for a New Science of Consciousness by Philip Goff. A readable book-length defense of panpsychism by the author of the phenomenal bonding proposal. See it on Amazon (affiliate link)
• Combining Minds: How to Think about Composite Subjectivity by Luke Roelofs. The most thorough academic monograph devoted specifically to the combination problem. See it on Amazon (affiliate link)

As an Amazon Associate I earn from qualifying purchases. A commission may be earned if you buy through this link, at no extra cost to you.

References

1. Albantakis, L., Barbosa, L., Findlay, G., Grasso, M., Haun, A. M., Marshall, W., et al. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms. PLoS Computational Biology, 19(10), e1011465. https://doi.org/10.1371/journal.pcbi.1011465
2. Allen-Hermanson, S. (2022 revision). Panpsychism. Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/panpsychism/
3. Bourget, D., and Chalmers, D. J. (2020 survey). PhilPapers 2020 Survey, question on consciousness. https://survey2020.philpeople.org/survey/results/5010
4. Chalmers, D. J. (2016). The combination problem for panpsychism. In G. Brüntrup and L. Jaskolla (Eds.), Panpsychism (pp. 179–214). Oxford University Press. Author’s version: https://consc.net/papers/combination.pdf
5. Coleman, S. (2012). Mental chemistry: Combination for panpsychists. Dialectica, 66(1), 137–166. Repository version: https://uhra.herts.ac.uk/id/eprint/3115/1/Combination_for_Panpsychists.pdf
6. Coleman, S. (2014). The real combination problem: Panpsychism, micro-subjects, and emergence. Erkenntnis, 79(1), 19–44.
7. Goff, P. (2016). The phenomenal bonding solution to the combination problem. In G. Brüntrup and L. Jaskolla (Eds.), Panpsychism (pp. 283–302). Oxford University Press.
8. Goff, P. (2017). Consciousness and Fundamental Reality. Oxford University Press.
9. James, W. (1890). The Principles of Psychology, chapter 6, The mind-stuff theory. https://psychclassics.yorku.ca/James/Principles/prin6.htm
10. Miller, G. (2017). Forming a positive concept of the phenomenal bonding relation for constitutive panpsychism. Dialectica, 71(4), 541–562.
11. Miller, G. (2021). The decombination problem for cosmopsychism is not the heterogeneity problem for priority monism. Journal of Consciousness Studies, 28(3–4), 112–115.
12. Mørch, H. H. (2018). Is the integrated information theory of consciousness compatible with Russellian panpsychism? Erkenntnis, 84(5), 1065–1085.
13. Nagasawa, Y., and Wager, K. (2016). Panpsychism and priority cosmopsychism. In G. Brüntrup and L. Jaskolla (Eds.), Panpsychism (pp. 113–129). Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199359943.003.0005
14. Roelofs, L. (2019). Combining Minds: How to Think about Composite Subjectivity. Oxford University Press.
15. Schaffer, J. (2010). Monism: The priority of the whole. Philosophical Review, 119(1), 31–76.
16. Seager, W. (1995). Consciousness, information, and panpsychism. Journal of Consciousness Studies, 2(3), 272–288.
17. Shani, I. (2010). Mind stuffed with red herrings: Why William James’ critique of the mind-stuff theory does not substantiate a combination problem for panpsychism. Acta Analytica, 25(4), 413–434.
18. Shani, I. (2015). Cosmopsychism: A holistic approach to the metaphysics of experience. Philosophical Papers, 44(3), 389–417. https://doi.org/10.1080/05568641.2015.1106709
19. Shani, I., and Williams, H. (2022). The incoherence challenge for subject combination: An analytic assessment. Inquiry, online first. https://doi.org/10.1080/0020174X.2022.2124541
20. Siddharth, S. (2021). Against phenomenal bonding. European Journal of Analytic Philosophy, 17(1). https://doi.org/10.31820/ejap.17.1.3
21. Tononi, G., and Koch, C. (2015). Consciousness: Here, there and everywhere? Philosophical Transactions of the Royal Society B, 370, 20140167. https://royalsocietypublishing.org/doi/10.1098/rstb.2014.0167
22. The Philosophy Forum. Counter argument for the combination problem for panpsychism (discussion thread). https://thephilosophyforum.com/dashboard/profile/comments/3/bert1/440