Daniel Dennett - Real Patterns (1991)

Dennett elegantly bridges the chasm between cellular automata and human social intentionality by leveraging the concept of a stable pattern and its status of reality within the world. "Real Patterns" a great piece of work, and its logic is worth following closely.

The presentation of a "pattern" is done by recourse to information theory and Chaitin's compressibility of a data stream. Though it's not explicitly mentioned, there's also an assumption of what Shannon and Weaver would term a "sender", encoding some structure based on an underlying pattern (through which the output is compressed - using, say, the shared grammar of a chess game). The receiver then interprets this data, perceiving the pattern through its underlying actuality. Chess pieces placed randomly would, to a professional, be significantly harder to perceive and recall; to a non-player, however, both layouts would appear arbitrary, and no pattern could be perceived.

Thus, patterns are real yet potentially only discernible from a given perspective. Dennett asserts that a pattern exists in some data "if there is a description of the data that is more efficient than the bit map, whether or not anyone can concoct it". We can infer from this, then, that there are relative magnitudes of pattern-ness, correlating with the degree of information compressibility that we can apply.

When we apply our formidable pattern-matching apparatus in the real world, we form what Sellars terms a "manifest image", overlaid onto our sensations through acquired knowledge and folk psychology, which allows us to make judgements as to what is presented to us and so make intentional decisions. This is done through a significantly statistical, inductive process: a highly weighted network of probabilities based on accumulated experience.

Now, back to the Game of Life. Dennett's critical move here is to go beyond glider guns and explain how we can create a Turing-complete machine from aggregates of Life cells, essentially constructing three new levels in the Life hierarchy

• at L=0, individual cells
• at L=1, persistent aggregates of gliders, blinkers, beehives, etc
• at L=2, aggregates of L=1 units which can perform logical operations
• at L=3, aggregates of logic structures capable of playing a (deterministic or pseudo-random) game of chess

The thought experiments that we are left to take away include: what is the ontological status of the patterns (glider guns, logic gates, etc) that have been created through these illusory collections of cells? At what perspectives would we be able to perceive our Life chess-player as such, and at what perspectives would it appear to be a random, chaotic mulch? Does the latter matter?

A really beautiful work, and one which subtly begins to also emphasise the statistical nature of how such patterns (on a vastly complex scale) may function in consciousness and other real-world emergent scenario. Just as Bedau argued previously, what arises are whole classes of macro phenomena which can be grouped by some mean tendencies: the tendency for 2-2 Life to result in a chaotic slime, the tendency of birds to flock in synchrony, the tendency of a human agent to act in loosely predictable, intentional ways. The metaphysical reality of an abstract centrepoint to such tendencies is difficult to confirm, but the broader reality of such persistent, useful patterns is difficult to deny.

This marks the end of the "Philosophical Perspectives on Emergence" section. Next up: Scientific Perspectives.

Mark A Bedau - Downward Causation and Autonomy in Weak Emergence (2003)

In which we are given the first serious and hopeful treatment of 'weak emergence', a term coined by Bedau some time prior to this stellar piece of work. Focusing squarely on the objectivist models of emergence (that is, those which do not rely on some subjective element of surprise), Bedau lays down a convincing argument that the ability to reduce an emergent property to its underlying parts does not make it uninteresting or insignificant.

Weak emergence sits in the middle of a trio of identified categories: "nominal emergence", which stipulates simply that a macro-level property depends on a collection of micro-level parts but cannot be held individually by these parts; and "strong emergence", which requires the existence of irreducible, supervenient macro-level properties and causal powers. Bedau makes his thoughts on strong emergence clear when he states: "Strong emergence starts where scientific explanation ends".

A phenomena which falls within the class of "weak emergence" can, given archangel-like computational powers, be derived from the network of interactions through which it emerges. However, such a network may comprise of "myriad non-linear and context-dependent micro-level interactions", making it unfeasible to forecast its outcome without simply iterating through these interactions. There is no "short-cut" derivation, to use Bedau's terminology; the extreme context sensitivity means we must simply churn through the micro-level processes until the macro-level outcome has been determined. This touches on the Kolmogorov-Chaitin notions of algorithmic complexity and incompressibility: there is no shorter way to calculate the algorithm's output than simply executing the algorithm itself.

One immediate question that surfaces is how complex this complexity needs to be. Surely some derivations are reasonably obvious, with quasi-shortcuts or broadly general solutions. Bedau addresses this by affirming a "spectrum of more or less weak emergence", with prospective properties being intractable to simulate. This fits into the intuition that there is no clear line between emergent and non-emergent properties.

We finally see how "weak emergence" tallies up with the perennial problems of causal exclusion and downwards causation. Since there is no longer the unbridgeable rift of duality that strong emergence imposes between macro and micro, the macro causal structure is equal to the aggregate of micro causal elements, and so no micro causal laws are overturned or out-prioritized. The vicious circle argument (in which a macro property can, at a given time t, theoretically affect and scoop out its own subvenient base) is not applicable because weak emergence is inherently diachronic; a macro pattern can subvert its micro constituents at time t+1, but that's OK -- this is exactly what happens in the real world (we experience neural pain as a headache, we take a painkiller, the underlying neural cause dissipates and we no longer have the conscious experience of pain).

The last worry, then, is that we are back to a plain epistemic mode of emergence: the entire causal structure at the macro level can be predicted, given knowledge of the micro-level constituents and sufficient processing time. This is true. However, given that a macro behaviour can be realised through many different routes, whole new general classes of macro entities can be created, with autonomy from particular micro pathways (and here, it's noticeable that the language switches to talking about the same "kinds of" macro behaviours). The justification is that the same process is used to justify causal autonomy between, say, chemistry and biology. This defence is only somewhat convincing, though feels like it is lacking in rigour.

As an addendum, most of Bedau's novel examples are given by way of Conway's Game of Life automata, its first major appearance in this reader. We'll be seeing more of it in the following chapters.

Sorry to all of you who have been checking back each day, only to be brutally rebuffed by the lack of any new doors. With luck, the missing days should be made up very shortly.

Jaegwon Kim - Making Sense of Emergence (1999)

Read as RTF

Kim's 'Making Sense of Emergence' ploughs thoroughly through a number of the most major questions for the philosophy of emergence:

downwards causation: is it possible or even necessary for a macro-level entity to be able to exert causal powers on micro-level parts (and, beyond that, its own micro-level parts)?

explainability, predictability, reducibility: can these properties be meaningfully decoupled, and which can then be applied to a truly emergent property?

synchronic vs diachronic causality: does it make sense for emergence to be divorced from a temporal base?

The conclusion is that the only well-formed foundation for strong emergence is one that is diachronically causal. A clearly seminal paper, but resulted in another feeling of metaphysical fatigue.

William C Wimsatt - Aggregativity: Reductive Heuristics for Finding Emergence (1997)

Rather than focusing on seeking the essential characteristics of emergence, Wimsatt's paper takes the opposite approach and attempts to pin down the set of properties for a property to be definitively non-emergent. We saw earlier that it's not a straightforward process to distinguish between the two in any case, with certain "obviously" linear-additive properties being a little more complex on inspection, and vice versa. Wimsatt throws in another nice example of nonlinear composition, that of volume under dissolution: the volume of a salt-water solute is actually less than the volume of either of its constituents. Sometimes, more is less.

The key thesis is that emergence is a consequence of certain organisational properties, combined with context-sensitivity of the parts that constitute this organisation. Non-emergent systems properties are termed "aggregates" by Wimsatt. To be truly aggregative, a property must be functionally invariant when its parts are subjected to any of the following transformations:

• intersubstitution (that is, rearranging or substituting parts for others)
• size scaling (adding or subtracting parts)
• decomposition and reaggregation (of parts)
• linearity

For the macro-scale systems property to not vary under any of these transformations, it is pretty clear that it must be radically functionally homogeneous. Wimsatt observes that the only paradigmatic aggregative properties are those governed by major laws of conservation: mass, energy, momentum and charge.

Perhaps the most rewarding movement of this argument is where Wimsatt takes it in the final couple of pages. With a background in the philosophy of biology, he writes on the structures that underlie natural selection and the models that we, as scientists, impose to understand them. Here, he criticises the "nothing but X" language of radical reductionism, such as in the oft-touted "genes are the only units of selection". However, if we take the complex dynamical systems that comprise the natural world and attempt to reduce them to a model based on one underlying constituent unit (the gene, the agent, the neuron), we cannot then make claims to universality of our model: this is what Wimsatt terms the functional localisation fallacy. Such a decomposition is useful to study some aspects of a system, but it should be understood that it must be complemented by other such decompositions from different levels and angles.

Brian McLaughlin - Emergence and Supervenience (1997)

Read as PDF

In terms of its subject matter, McLaughlin's second paper in the collection follows on chronologically from his first. In the wake of quantum mechanics and other modern scientific advances, he affirms that: "On the current evidence, it appears that all fundamental forces are exerted below the level of the atom". So, is it still logically tenable to appeal to truly "emergent" forces which are genuinely radically unexplainably from underlying processes?

Yes, is the result, though it's highly unclear that there actually exist any such forces. In brief, McLaughlin completes the hard work of the later Emergentists by formulating a rigorous definition of what it would mean to be truly, intuitively emergent:

If P is a property of w, then P is emergent if and only if (1) P supervenes with nomological necessity, but not with logical necessity, on properties the parts of w have taken separately or in other combinations; and (2) some of the supervenience principles linking properties of the parts of w with w's having P are fundamental laws.

Got that?

The magic lies in the use of "nomological necessity", which approximates to a semantic relationship of implication (if "the parts have property A" is true, then "the whole has property B" is true) rather than necessity through logical deduction (such as that of "all bachelors are male"); and the "fundamental laws" clause, which are akin to the Emergentist "configurational laws": that is, non-deducible first principles.

This definition is constructed through the use of the supervenience relation (see left), which sees wide use throughout analytic philosophy. To say that mental states supervene on neural states is to say that any change in mental state also entails (or, alternatively considered, requires) a change in neural state. Conversely, many neural states (labelled A on diagram) may potentially map to the same mental state (B).

So, there we go. Through this modal-logic scaffolding, emergence has been shown to be logically valid. However, McLaughlin himself is the first to admit that, even so, the only remaining known candidate for true emergence is consciousness - and this too is only left as an "open question". The resolution will come if it is ever revealed that the principles on which conscious states supervene are "fundamental" (i.e., in accordance with vitalism) or otherwise. My feeling is the latter.

John Searle - Reductionism and the Irreducibility of Consciousness (1992)

So far in this festive season of emergence, we have seen the radically strong position (i.e., emergent properties are real and ontologically irreducible to their parts); and the radically weak position (i.e., emergent properties are illusory, a consequence of our present ignorance of their true causal factors). With Searle's Reductionism and the Irreducibility of Consciousness, we start hitting the nitty-gritty, philosophically refining the space between these two polar points. We do so by walking into the minefield of philosophy of mind.

Consciousness is perhaps the paradigmatic example of a radically emergent phenomenon. By its nature, it is intrinsically subjective and complex (see Nagel's What Is It Like To Be A Bat?). Science is faring little better - no neurophysiological correlate has yet been found to allow us to predict reliably whether a subject is experiencing consciousness.

Searle's account begins with the claim that consciousness is emergent not just from the spatial relationships between the mind's constituent neurons, but from the causal interactions between them. He accepts that mental features (those of experience) are caused by their neural substrate, but denies that they can simply be ontologically reduced to them, in the same way that the liquidity of a substance cannot be reduced to the spatial configuration of its molecules; instead, both rely on "causal" emergence, in which the causal powers of consciousness can be fully explained by the causal powers of its underlying neurons.

Here, linguistic concepts are added to the mix, as Searle looks at how emergent concepts are formalised. Take the example of "redness". Starting with a subjective experience of red things in the world, we advance our scientific knowledge and come to the understanding that "redness" is caused by the reflection of a certain range of wavelengths of light. We then redefine "redness" as this objective, underlying principle, and our subjective experience of red things becomes relative to this real-world fact. In Searle's terminology, we "carve off the surface features" of redness - the surplus contained within a subjective experience - and are left with a relationship between affect and reality.

He proceeds to argue that, given that consciousness is itself the "subjectiveness" of experience, there is nothing to carve off, and no underlying reference point. We can no longer distinguish between the referent and our experience of it - indeed, the underlying phenomena in question is subjectivity itself. So, the two have converged, meaning that this technique of "reduction" cannot apply to consciousness, by definition.

This is all fine. However, I can't help but feel a little short-changed: all we are left with is the outcome of a metaphysical game.

Searle uses the convergence exercise to argue that "consciousness" is an irreducible fact, after whose application "we are still left with a universe that contains an irreducibly subjective physical component as a component of physical reality". That is, consciousness exists, and we cannot use the carving-off technique to attach it to some external pattern. But, in the neural state space, is it not possible that there is some continuous subspace which directly correlates to the experience of X conscious state? If so, would it not be acceptable to come to refer to this fuzzy state space as "consciousness"?

I'm away tomorrow and over the weekend, so normal advent programming will resume on Monday. Apologies for any distress this may cause.

Carl Hempel and Paul Oppenheim - On The Idea of Emergence (1965)

Moving from the previous chapter's account of perhaps the strongest ontological statement of emergence, Hempel and Oppenheim make the counterpoint by arguing that the appearance of "emergent" phenomena is, in fact, a result of our ignorance of intermediary laws. In the tradition of logical positivism, they state that, until we have a micro-theory that gives us insight into the "inner mechanism" of a phenomenon, we do not truly have "real scientific understanding" of it. This epistemic limitation gives rise to our surprise when encountering such a phenomenon, creating illusory emergence, which later dissipates as our knowledge of the world develops.

Emergence is, then, relative to a set of theories, which include a set of bridging laws to map between (say) physico-chemical terms and biological terms. A sufficiently advanced theory-set allows us to deductively infer the relevant macro property, and the emergence vanishes.

What we are left with is the weak emergence in vogue today, against the Emergentist tradition of strong, ontological emergence. Both will be refined further over the next few weeks.

One interesting addendum is their criticism of the resultant/emergent dichotomy adopted by many of the British Emergentists. A property is said to be resultant if it can obviously be deduced from the sum of its parts; for example, the mass of a stone is equal to the additive sum of the mass of its constituent molecules. It is said to be emergent if it is not explainable or predictable from the combination of its parts. However, both of the two classes are really subjective judgments. "Obviousness" is in the eye of the beholder, and a compound behaviour which may seem unpredictable under some set of theories can seem obvious - logically deductable, even - under some other set. Additionally, under the mechanics of relativity, the "obvious" additive property of mass is not, in fact, a linear resultant, and so the argument loses its remaining support.

Note, however, that this is dependent on our beholder having sufficiently broad perspective as to encompass all of the relevant theories and bridge laws, and potentially unlimited computational powers (a la Broad's "mathematical archangel") to be able to follow through the labyrinthine web of causation that may lead to our emergent...

Brian McLaughlin - The Rise and Fall of British Emergentism (1992)

The first and longest of the papers published in Emergence, McLaughlin's Rise and Fall puts the collection in context by providing an overview of the first major, sustained philosophical discussion of emergence: between a series of British thinkers, from JS Mill's System of Logic (1843) to the scientific advances of the 1920s onwards.

First, as the opening of this series, an overview of what's at stake. "Emergence" is the phenomenon of macro-level properties or behaviours that are a product of an aggregate of micro-level parts. Popular examples include nature's synchronised swarming behaviours, physical phase transitions (say, from solid to liquid) at a critical temperature, consciousness and thought, the phenomenological experience of colour, etc. The Stanford Encyclopaedia provides a potted overview.

Emergentism is a strong philosophical brand of belief in emergence, which states that there are emergent phenomenon which can in no way, ontologically or epistemologically, be fully explained from their lower-level constituents. Working from the assumptions that:

• everything can be reduced to matter, with some underlying level of elementary particles; and
• there is a hierarchy of levels above this; from bottom up: physics, chemistry, biology, psychology… (cf xkcd)

..the British Emergentists claim that, though the matter of a level B may be comprised of the same stuff as its underlying A, it may be able to exhibit special "configurational forces" which cannot be explained or deduced from the forces of A. The motivation for such thinking was the current interest in chemical reactions, and the unexplainability of (say) the dissolution of salt in water from elementary particles.

Unfortunately, as McLaughlin observes, although this form of emergence may be logically coherent, it only remains empirically viable so long as we have no scientific way of understanding how such laws can emerge without resorting to some mystical forces; this is a God of the Gaps scenario. McLaughlin refers repeatedly to the "natural piety" that Alexander recommends we adopt for such faith-reliant situations. And, as Schrödinger and Einstein's leaps in quantum mechanics provided explanations of chemical bonding which did, indeed, bridge between levels, the a posteriori basis of British Emergentism collapsed.

The punchline of the chapter is that, today, we must accept with "natural piety" the difficult fact that high-level concepts such as production do indeed supervene on the same minimal set of forces as electromagnetic bonds.