Personal Multiplicity Theory

Chapter 4 – Specialisation, Modularity and Devolved Function
Introduction

The preceding chapter widened the argument beyond psychology. Multiplicity is not an unusual condition confined to troubled minds or unusual experiences. Living systems repeatedly achieve higher levels of organisation by preserving distinctions among lower-level components and coordinating their activity.

Multicellular organisms depend upon differentiated cells; colonial organisms can combine repeated modules into larger functional units; animal collectives can produce organised behaviour through interactions among individuals that remain distinct. From that perspective, the psychologically important question is no longer simply why a person might contain differentiated processes. A more fundamental question comes first:

Why does complex organisation become specialised at all?

Specialisation has an obvious everyday advantage. If every member of a group had to perform every task equally well, considerable effort would be spent duplicating capabilities. Division of labour permits different components to become particularly competent within narrower functional territories and allows the larger system to combine their contributions.

Biology demonstrates this principle at many different scales, but it also reveals an important qualification. Specialisation works only if differentiated functions remain sufficiently connected to influence one another where necessary. A completely isolated specialist would cease to contribute to the larger organism.

The resulting architecture therefore depends upon two apparently opposing requirements: difference and relationship. Components must be sufficiently differentiated to perform specialised work, but sufficiently connected to participate in a larger system.

This chapter proposes that this principle provides an important foundation for psychological multiplicity. It does not establish that psychological parts are equivalent to cells, brain regions or computer modules. Rather, it asks whether the familiar existence of functional specialisation throughout biology and cognition makes a similarly differentiated psychological architecture scientifically plausible.

Biological functions are often properties of organised modules

At the molecular level, even the individual cell is already far from being one undifferentiated process. Hartwell, Hopfield, Leibler and Murray argued that many important cellular functions cannot usefully be attributed to a single molecule. Signal transduction, regulation and other cellular behaviours emerge through interactions among multiple molecular components organised into what they called functional modules (Hartwell et al., 1999, p. C47). A module, in their definition, possesses a sufficiently separable function that it can be treated as a meaningful level of biological organisation even though the function itself emerges from interactions among proteins, DNA, RNA and other molecular components (pp. C47–C48). (ResearchGate)

This is already instructive for the present thesis. The relevant function does not have to reside inside one privileged component. A module can be a relationship among many components whose organisation produces a capability that none of the components possesses alone.

Hartwell and colleagues specifically noted that biological modules can be either insulated from or connected with other modules: some separation prevents destructive interference between functions, while connections allow one function to influence another.

They argued that higher-level cellular properties can consequently depend upon the pattern of connections among specialised modules rather than upon the elimination of the modules themselves (Hartwell et al., 1999, p. C48). (ResearchGate)

That distinction is particularly important. Specialisation does not necessarily mean physical separation.

Hartwell et al. describe cellular modules whose relative independence can arise through spatial localisation, but also through molecular specificity. The same molecule may sometimes participate in more than one module, and modules can themselves be combined into higher-level functional organisations.

Their example of mitosis involves several interacting modules concerned with spindle assembly, chromosome alignment and cell-cycle regulation contributing to the larger task of accurately distributing chromosomes between daughter cells (Hartwell et al., 1999, p. C48). (ResearchGate)

The resulting picture is not a collection of little isolated machines. It is closer to an organised ecology of functions.

A component can contribute to one domain without every other domain needing unrestricted access to it. A function can be relatively autonomous without being independent of the larger cell. And higher-order behaviour can arise by coordinating the outputs of several lower-order modules.

This gives us an important distinction for what follows:

functional independence is a matter of degree and relationship, not necessarily complete isolation.

Specialisation can increase as higher-order organisation develops

The evolution of multicellularity provides the same principle at another scale. Chapter 3 considered multicellularity mainly as evidence that unity and multiplicity are compatible. Here its significance is slightly different: multicellularity creates opportunities for division of labour.

Ratcliff and colleagues demonstrated this experimentally by selecting initially unicellular Saccharomyces cerevisiae yeast for rapid settling. Within a relatively short experimental period, independently evolving populations developed multicellular “snowflake” clusters with a multicellular life cycle. Importantly for the present argument, these clusters did not simply consist of identical cells remaining together. Programmed cell death increased during subsequent evolution and contributed to the production of multicellular propagules, giving the authors evidence for an emerging, comparatively simple division of labour among cells (Ratcliff et al., 2012, pp. 1595–1596). (ResearchGate)

The experiment should not be asked to prove more than it does. Snowflake yeast do not demonstrate why the human psyche has parts. They show something more general and directly relevant: once lower-level units participate in a larger organisation, differentiation of function can itself become advantageous.

The higher-order system gains capabilities because every constituent does not have to remain functionally interchangeable.

This gives specialisation two related advantages. First, different components can optimise different kinds of work. Second, changes in one specialised function need not necessarily disrupt every other function simultaneously. Hartwell and colleagues make the latter point explicitly at the cellular level. They argue that modularity can facilitate evolutionary modification because a function contained within a relatively discrete module can be altered, or its connections altered, without requiring every other cellular function to change with it (Hartwell et al., 1999, p. C48).

Later in the same article they emphasise the tension between robustness and evolvability: essential aspects of a module can remain stable while other aspects, including its connections with other modules, remain capable of evolutionary modification (p. C51). (ResearchGate)

This becomes highly relevant to the later theory of psychological development. A differentiated architecture potentially allows one function, relationship or route of communication to change without requiring the organism to redesign every other capability.

We are not yet proposing a mechanism for such psychological change. At this point the biological principle is sufficient:

modular organisation permits both specialised competence and selective modification.

The nervous system is not simply modular in one place

It might be tempting to move directly from cellular modules to mental modules and declare the problem solved. That would be too easy.

The brain itself warns against such simplification.

There is substantial evidence for functional specialisation in the nervous system, but that specialisation does not mean that every psychological operation occupies a single isolated anatomical box. Different questions reveal different patterns of localisation, distributed processing and network interaction.

One particularly striking form of evidence comes from selective neuropsychological dissociation. Goodale, Milner, Jakobson and Carey studied a patient with profound difficulty consciously perceiving object characteristics such as orientation and size. Despite that perceptual impairment, the patient could guide her hand and fingers accurately when reaching for those same objects. The authors concluded that the neural systems supporting conscious perception of object properties could be distinguished from those using visual information to control skilled action (Goodale et al., 1991, pp. 154–156). (Nature)

This result does not mean that “perception” and “action” are two psychological people. It demonstrates something far more defensible: the same environmental information can participate in different specialised processing relationships, and severe disruption to one of those relationships need not eliminate every other use of the information.

Memory research produced an equally influential example decades earlier. Scoville and Milner reported profound and persistent impairment of recent memory following sufficiently extensive bilateral medial-temporal resections involving the hippocampal region.

Yet the impairment was strikingly selective rather than a destruction of all psychological functioning. Their discussion relates the severity of the memory deficit particularly to the extent of damage involving the hippocampal complex while documenting preservation of other capacities in the patients studied (Scoville & Milner, 1957, especially pp. 15, 20–21).

Such dissociations helped undermine the idea that cognition is produced by one undifferentiated mental faculty. If damage can severely compromise one capability while leaving many others comparatively intact, then some degree of functional differentiation must exist in the organisation generating those capabilities.

But dissociation evidence also creates a methodological danger. Finding that two functions can be separated does not tell us that they are normally isolated from one another. Nor does identifying a brain region important for one process prove that the process exists entirely inside that region.

The appropriate inference is therefore again relational:

specialisation exists, but specialisation must be studied together with integration.

Brain networks show modules within modules

Modern network analysis makes this point particularly clear.

Meunier and colleagues examined resting-state functional brain networks in eighteen healthy adults. Rather than simply attempting to divide the brain into one fixed set of regions, they investigated whether its network organisation could be decomposed hierarchically – that is, whether larger functional modules could themselves contain smaller sub-modules.

Their analysis of networks containing more than 1,800 regional nodes found evidence for such hierarchical modular organisation, with large-scale modules decomposable into subordinate structures at lower levels (Meunier et al., 2009, pp. 1–3). (Frontiers)

The significance for this thesis is not that the resulting brain modules are psychological parts. They are not. The important point is that the physical system supporting psychological life already resists a simple choice between:

one integrated network

and

many separate networks.

It can be both.

Meunier and colleagues describe the relevant architecture in terms closely related to Simon’s earlier concept of near-decomposability: interactions are denser within some functional groupings than between them, while particular connector nodes and hubs contribute disproportionately to communication across modular boundaries (Meunier et al., 2009, pp. 1–2). (Frontiers)

This is a much more useful picture than the old metaphor of a brain divided into independent boxes.

A system can have local density of interaction without complete informational isolation.

It can have specialised regions or networks while relying upon cross-network communication.

And it can have levels of organisation, such that what appears to be one module at one scale contains several distinguishable sub-modules when examined at greater resolution.

That final point will become crucial in Chapter 6, where we consider the possibility of nested psychological domains. For now, it provides strong external support for the more modest proposition that specialisation can be hierarchical rather than flat.

Cognitive theories have faced the same integration problem

The move from neuroscience into psychology produces an interesting historical repetition.

As cognitive science identified increasingly specialised mechanisms, it encountered exactly the problem that biological modularity creates: if cognition consists of many differentiated processes, how is coherent cognition produced?

Anderson and colleagues address this question explicitly in their presentation of the ACT-R cognitive architecture. They begin by noting the proliferation of evidence and theories concerning specialised mechanisms in areas such as visual processing, memory, language and motor behaviour.

They then identify the obvious weakness of an account containing only separate specialties: it does not explain how those components combine to produce the behaviour of one mind (Anderson et al., 2004, pp. 1036–1037).

ACT-R is one proposed answer rather than an established description of the brain. Its usefulness here lies in the kind of architecture it considers scientifically legitimate.

The model contains distinct modules concerned with different classes of processing, including visual information, manual action, declarative memory and current goals.

These modules do not make their complete internal states available to the rest of the system. Instead, limited information is placed into buffers through which coordination can occur. Anderson and colleagues explicitly compare this limited availability with the fact that people do not have conscious access to everything in the visual field or everything contained in long-term memory at once (Anderson et al., 2004, p. 1037).

Again, the details of ACT-R should not be imported wholesale into Human Multiplicity Theory. In particular, ACT-R includes a central production system coordinating module outputs, whereas our later model will investigate whether some forms of psychological coordination can be more distributed or devolved.

What matters is that cognitive architecture already takes seriously a problem structurally similar to ours:

specialist processes can possess information unavailable in full to the larger coordinating system, while selectively communicating information required for coherent action.

That is a substantial conceptual step beyond merely saying that “different parts of the brain do different things.”

Modularity is not one settled doctrine

The term module itself requires caution because it has accumulated several different meanings.

Barrett and Kurzban reviewed the long-running dispute over cognitive modularity and noted that much disagreement arises because different researchers use the term for different combinations of properties.

Fodor’s influential conception was comparatively restrictive, involving features such as domain specificity, specialised operation and forms of informational encapsulation. Later modular theories often use a less restrictive notion centred more generally upon specialised information-processing systems (Barrett & Kurzban, 2006, pp. 628–630). (Squarespace)

The distinction is useful because Human Multiplicity Theory does not require us to adopt the strongest version of cognitive modularity.

We do not need to claim that every psychological specialist is:

• anatomically localised;
• completely encapsulated;
• genetically specified as an independent unit;
• unable to learn;
• or permanently isolated from other systems.

Indeed, such claims would conflict with much of the architecture we are developing.

Barrett and Kurzban themselves argue that the simple yes-or-no question is this system modular? can become less informative than asking precisely what information a system processes and what its computational properties are.

They point out that different kinds of information are handled by different systems while rejecting the assumption that all such systems must exhibit an identical set of modular properties (Barrett & Kurzban, 2006, pp. 630–631). (Squarespace)

This is close to the approach required here.

The thesis does not need a universal psychological module with a standard shape.

It needs something more modest:

functionally differentiated domains can exist within a larger system, can possess unequal access to information, and can participate differently in the production of behaviour.

The exact degree of autonomy, permeability, learning and localisation may vary from domain to domain.

In other words, specialisation is the important principle; rigid modularity is not.

From specialisation to functional domain

We can now introduce a term that will be used throughout the remainder of the thesis.

A functional domain is the informational and operational territory associated with some differentiated function within the larger organismic system.

The phrase deliberately does not specify a particular physical size or psychological ontology.

A functional domain might involve many brain regions rather than one.

It might participate in several larger systems.

Its boundaries might change with context.

And what appears to be one domain at one level may later prove to contain several specialised sub-domains.

The concept therefore identifies jurisdiction of function, not necessarily location.

This is important because the preceding evidence suggests that functions can be separable in several different senses. Hartwell’s molecular modules can be differentiated through spatial localisation or molecular specificity. Neuropsychological dissociations can reveal functions that depend differently upon particular neural systems.

Network analysis can reveal densely interacting communities without requiring absolute separation. Cognitive architectures can distinguish specialist processors whose internally available information differs from what is communicated to the wider system. (ResearchGate)

A functional domain is our deliberately broader category for asking what these forms of specialisation might mean psychologically.

At this stage we are still not entitled to call such a domain a part.

That term will eventually imply a richer psychological organisation.

For now, we are establishing only that there is nothing scientifically unusual about an organism containing differentiated processing systems whose competencies and informational territories are not identical.

Devolved authority

This brings us to a second project-specific term: devolved authority.

The word authority is potentially misleading if interpreted socially or politically, so its meaning must be narrow at this stage.

Devolved authority means:

a specialised process has sufficient local control to perform its ordinary function without requiring conscious awareness to specify every intermediate operation.

The body’s regulation provides obvious examples. Conscious awareness does not decide each individual muscular adjustment required to remain standing, specify each stage of a familiar reaching movement, or consciously search every stored memory before one becomes available.

The neurological and cognitive evidence considered above instead points towards specialist systems performing large amounts of processing locally and contributing selected outputs to wider behaviour. (Nature)

Calling this authority does not mean that the subsystem possesses rights, political sovereignty or conscious intention.

It means that the larger organism has not centralised every functional decision at the level of reflective consciousness.

This is actually a requirement of competence. If conscious awareness had to calculate every correction involved in grasping a cup, retrieve and compare every memory potentially relevant to a conversation, or supervise every transformation in visual processing, ordinary behaviour would become impossibly cumbersome.

The advantage of specialisation is precisely that the specialist can do work for the wider organism without the wider organism reproducing the specialist’s entire calculation.

That proposition is already visible in ACT-R’s architecture, despite its central coordination mechanism. The production system responds only to limited information made available by specialised modules rather than continuously inspecting all of their internal activity (Anderson et al., 2004, p. 1037).

We therefore propose that devolved functional authority is a useful way of describing an organismic architecture in which specialised systems retain enough control over their own processing to perform their jobs while contributing relevant information to larger-scale coordination.

This is the first point at which the argument begins to approach Human Multiplicity Theory directly.

Different specialists can produce different recommendations

If functions are specialised, another consequence follows.

Different domains do not necessarily possess the same information, operate on the same timescale, or evaluate the same aspect of a situation.

It therefore becomes possible for two specialist systems to produce different outputs without either system being meaningless.

This is already implicit in the neurological evidence. The visual information used to guide an action need not be organised identically to the visual information used for conscious perceptual judgement, as Goodale and colleagues’ patient dramatically demonstrated (Goodale et al., 1991, pp. 154–156). (Nature)

The psychological implication is a proposal rather than an established consequence of that experiment.

Suppose one functional domain is especially concerned with maintaining social attachment while another is especially sensitive to threat. A particular relationship might provide both attachment value and evidence of danger. If the domains possess different functions and different informational emphases, one might favour approach while another favours withdrawal.

The person then experiences:

I want to stay.

and:

I want to get away.

The ordinary language with which Chapter 1 began no longer has to be interpreted simply as one irrational process contradicting itself. Another possibility becomes available:

different specialist functions may be generating different evaluations because they are solving different problems.

At this stage that remains a theoretical extension. We have not demonstrated that attachment and threat operate as the particular psychological domains the later theory will propose. The argument is structural.

If psychological processing is specialised, then difference of output is an expected possibility, not automatically evidence of malfunction.

That provides a deeper foundation for our earlier claim that internal disagreement can sometimes carry useful information.

Specialisation does not imply independence

We should resist one further mistake before moving on.

If a specialist possesses local competence, it is easy to imagine that each specialist becomes a self-contained miniature mind. Nothing in the evidence supports that conclusion.

Every example examined in this chapter points towards partial differentiation combined with dependence.

Cellular modules require connections with other modules.

Neural systems interact in networks.

ACT-R’s specialised processors participate in a larger architecture.

Even striking neuropsychological dissociations show that one capability can be impaired while others survive; they do not establish that normal cognition consists of independent little agents operating without interaction. (ResearchGate)

Human Multiplicity Theory therefore begins from specialisation, not separatism.

A psychological specialist could possess information unavailable to another system without becoming an independent organism.

It could have a characteristic function without possessing a complete personality.

It could process information without being conscious.

It could contribute to a decision without knowing the entire context in which that decision will ultimately be made.

This is why the term domain is useful. It allows us to speak about function before making stronger claims about personhood.

Later chapters will ask whether some domains become sufficiently elaborate to merit descriptions such as part, self-aspect or eventually personified interface. Those are additional propositions. They do not follow merely from functional specialisation.

The unresolved problem is access

We have now reached a more precise version of the problem that began this chapter.

Complex systems gain capabilities by differentiating functions. Biology provides examples of molecular modules that are simultaneously separable and interconnected. Experimental evolution demonstrates the emergence of division of labour within newly multicellular systems.

Neuropsychology reveals selective disruptions suggesting that some cognitive functions can be separated from others. Network neuroscience finds hierarchical modular organisation rather than either complete uniformity or complete isolation. Cognitive architectures explicitly model coherent cognition as arising through the interaction of specialised processors. (ResearchGate)

These findings do not establish Human Multiplicity Theory.

What they establish is the plausibility of its starting architecture:

a complex organism can contain specialised systems whose informational access, competencies and functional responsibilities differ while all remain constituents of one larger organism.

We propose the phrase devolved authority for the local functional competence this makes possible.

But this immediately creates another problem.

If specialist domains simply shared everything with everything else, much of the advantage of differentiation would disappear. Processes irrelevant to one another would continually interfere. Information important to one specialist could swamp systems that had no use for it. Yet complete isolation would be equally disastrous: specialised functions could no longer contribute to coordinated behaviour.

Some mechanism must therefore determine:

what information stays local, what information crosses a boundary, where it goes, and in what form it becomes available.

Hartwell and colleagues encountered essentially this problem at the cellular level when they distinguished insulation from connectivity between functional modules (Hartwell et al., 1999, p. C48).

ACT-R encounters it cognitively by allowing modules to make only limited information available through buffers (Anderson et al., 2004, p. 1037). Brain-network research encounters it topologically through dense within-module relationships combined with more selective connections between modules (Meunier et al., 2009, pp. 1–3). (ResearchGate)

These are not identical mechanisms.

Their convergence nevertheless exposes the next architectural question.

How can information remain sufficiently compartmentalised for specialisation while still being routed appropriately through the larger organism?

To address that question, we need to move from the existence of functional domains to the architecture of informational boundaries.

That brings us to the schema mask.

References

Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y. (2004). An integrated theory of the mind. Psychological Review, 111(4), 1036–1060. Particularly pp. 1036–1037. (PubMed)

Barrett, H. C., & Kurzban, R. (2006). Modularity in cognition: Framing the debate. Psychological Review, 113(3), 628–647. Particularly pp. 628–631. (Squarespace)

Goodale, M. A., Milner, A. D., Jakobson, L. S., & Carey, D. P. (1991). A neurological dissociation between perceiving objects and grasping them. Nature, 349, 154–156. (Nature)

Hartwell, L. H., Hopfield, J. J., Leibler, S., & Murray, A. W. (1999). From molecular to modular cell biology. Nature, 402 (Supplement 6761), C47–C52. Particularly pp. C47–C48 and C51. (Nature)

Meunier, D., Lambiotte, R., Fornito, A., Ersche, K. D., & Bullmore, E. T. (2009). Hierarchical modularity in human brain functional networks. Frontiers in Neuroinformatics, 3, Article 37, 1–12. Particularly pp. 1–3. (Frontiers)

Ratcliff, W. C., Denison, R. F., Borrello, M., & Travisano, M. (2012). Experimental evolution of multicellularity. Proceedings of the National Academy of Sciences, 109(5), 1595–1600. Particularly pp. 1595–1596. (PNAS)

Scoville, W. B., & Milner, B. (1957). Loss of recent memory after bilateral hippocampal lesions. Journal of Neurology, Neurosurgery & Psychiatry, 20(1), 11–21. Particularly pp. 15 and 20–21. (jnnp.bmj.com)