Epiphenomena, Information and Emergence


I have substantially re-written this page (17-03-20 - updated 22-04-20), after unsuccessfully burying it by taking all its links away. Since it appears in search engines (those that search for 'epiphenomena and hard emergence' will find it) I decided instead to expand and develop this page to become a useful part of the theme.

The reason this page is here is that some scientists and philosophers of science maintain that all the things loosely referred to as 'emergent' are in fact only epiphenomena. After explaining what that means, I go on to consider some consequences, especially relating to information. This page is best read in conjunction with the page on emergence in the biocomplexity theme.

What are epiphenomena?

According to the conventional view of emergent phenomena, they are all perceptual appearances and therefore subjective. This is because the conventional view is one of radical reductionism. This philosophical stance is illustrated with the paragraph below (though it is not the position held by the writers of this website).

The national economy goes through cycles of fluctuating income, debt, employment rates and so on, but this is all really the consequence of millions of individual trading transactions conceptually aggregated. The national economy is an epiphenomenon of those trading transactions because it is a way we understand their overall effect, but it has no existence of its own: it is just the product of our minds. The former UK Prime Minister, Margaret Thatcher is famous for having said (though she did not quite) that "there is no such thing as society, there are just individual people". She may have been expressing a view on epiphenomena, had she not been attacking socialism. In material terms, she was right - there really are only individual people and what we call 'society' is a concept, rather than a real thing. It describes the aggregated effect of interactions among the people taking part. Though it has properties that we can describe, measure and predict with models, it is not really there, it is just a 'course-grained' appearance that emerges when we step back and look at the big picture rather than the detail.  In short it is an epiphenomenon of human relationships of all kinds. Reductionists say that because emergence is the appearance of phenomena at a scale of system organisation that is not present at the lower scales within it, emergence qualifies as a kind of epiphenomenon. Epiphenomena are observed, but are not things in themselves; rather they are ways of thinking about, or interpreting, the things from which they emerge. In the dictionary definition an epiphenomenon is a by-product of an underlying process. Crucially, it is defined as having no power to influence underlying 'real' processes. In particular, epiphenominalism is the dualist position on the nature of mind, saying that mind (or consciousness) is merely a by-product of physiological processes that has no power to affect them. Whether or not the mind, or society, is real or epiphenominal therefore rests on the practical test of whether or not it has any influence on other things that we already recognise as real.

So, is society really not a thing? Is a cell really just a bag of chemicals, or even just a particular arrangement of atoms and what we see as a cell no more than the way we make sense of that in our own minds? ... And are our minds nothing but innefective by-products of underlying neurological processes that have no real effect in the world. If that is true, then there cannot possibly be free-will, nor any point to thinking about society or the economy as a whole, since these are all innefective illusions. (On the emergence page, I give an account of the best known philosophical argument along these lines from Jaegwon Kim).

Recall that emergence refers to properties that are observable at a system level that cannot be expected or understood by studying only the level of the component parts. Taking, for example, the engine and its component 'mechanodiversity', we know that the function of the engine is not accessible when all we have is a jumble of component parts. This is because the precise way they are put together instantiates information which is then responsible for this function (jointly with the functions of the components). A cell is not any bag of chemicals, but a very specific arrangement of them and again this arrangement is information that gives function to the whole. A possible defense of phenomena for real, as opposed to epiphenomena, is that this system-level information is something genuinely additional to that of the sum of the parts. If this were true, then the cell does exist as a real thing in itself and so does a society and perhaps... the mind.

An important warning is needed at this point as we contemplate the difference between what is real and what is mere appearance. Most philosophical scientists appreciate that what we think we know is all just appearance. We only have our perceptions and the thinking which creates models of the world of which we are a part. We have no direct experience or understanding of what is real. Every scientific theory, Newtonian mechanics, quantum mechanics and relativity included, is only a model of reality that is satisfactory in supporting our understanding within limits of applicability that we refer to as its domain. The theoretical physicist David Bohm well illustrated (in Wholeness and the Implicate Order, 1980) the now widely accepted view that underlying the reality that we are familiar with is a strange, but more real, world - not of interacting particles, but of fields of influence and flowing 'essence' that has no familiar counterpart. In this world, there are no separate things as we would understand separateness: the entire cosmos is an indivisible whole of interacting fields. The consequence is that the most concrete seeming ensemble of individuals (atoms, people, neurons, etc.) are themselves an illusion because the deep nature of reality is indivisible wholeness. This means that when we call something real, what we are actually saying is that we experience, or can detect it, in a form that 'makes sense' to us because it is consistent with our experience or the way we have learned to think about the world.

Given that, and being practical, we would not want to imply that all the things claimed to be emergent are mere epiphenomena, nor that they are all something more substantial. Clearly, if some are more than epiphenomena, then there should be a way to distinguish between those that are and those that are not. The standard distinction is between hard emergence and soft (or strong and weak), but this is really a definition of epiphenomena (the soft or weak emergence) and "real" emergence (that is hard or strong). But from the standard definition of epiphenomena, we can see that if the emergent is able to do something that cannot be strictly attributed to the behaviour of its component parts, then it will be a case of hard emergence. Since it is necessarily something existing at a higher level of organisational structure than its constituents, this means that it must be capable of exercising downward causation if it were more than mere epiphenomenon. The trouble just now is that there is still some debate about the very existence of downward causation and the debate is very much divided along the reductionist / constructivist lines. Here we have an information-based view on this that we think helps.

An Information-based Interpretation
First, let us admit that modern physics currently conceives the whole of physical reality as a mere appearance which arises from an underlying quantum reality to which we have no direct access. For those not professionally involved in that level of description, particularly for biologists, it is usually sufficient to think of matter as atomic, though in the back of our mind we should remember that atoms are an approximation. Still, from that point on, the less than 100 different kinds of atom, with locations and momenta in space and time are the foundation of most familiar phenomena (anything else is given the specific title of a quantum effect). If we just think of atoms, then the specific type, location and momentum of each is an item of information. The way they bind together to form molecules can be explained in thermodynamic terms and the consequence of this for the embodied information can be quantified via a change in entropy. What underlies the mutual actions of atoms is physical causation, that is, efficient cause. Physical forces act on atoms (and of course their elementary components) to arrange them in their least energetic states. The forces come from the sub-atomic particles and act upon the particles and the strength and direction of these forces is determined by relative positions and these are the embodiment of information. Efficient cause is nothing more than the constraint of physical forces by the configuration of matter. The interactions among molecules are similarly understood.  Cascades of chemical reactions can be very complicated in biochemistry, but still, collectively, they can be understood as interactions among configurations of atoms in space and time, and therefore of patterns of information embodied in material form. These are patterns that constrain the action of physical forces and thereby constitute efficient cause (a good example is the way an antibody interacts with an antigen, a single binding site clamps onto a single epitope via some combination of hydrogen bonds, ionic bonds, hydrophobic interactions and van der Walls interactions, all of which are variants of electrostatic interactions among the electron clouds of individual atoms, constrained by being set in particualar locations within the whole molecules).

We can apply the same reasoning up the nested hierarchy of life: metabolic and signaling pathways; the 'cellular operating system'; the cell itself; organisms and communities and finally ecosystems. At every level, we really have patterns of information that exist because they are embodied in patterns of atoms. Crucially, these patterns provide Aristotelean formative causation. They act as the constraint on the direction of forces and in combination with those forces, they constitute efficient causation.

At what point do we then say, below this level of organisation, phenomena are 'real' and above it they are merely epiphenomenal? I think the answer is that there is no level for which we can objectively make such a claim. To use the terminology of Denis Noble "there is no privileged level" in the causal architecture of living systems. This is because every level is the same, in that patterns of information embodied at any one level are constraints for physical forces and these together act as efficient cause - at the same level, the next above and (potentially) below. Then the difference between a genuine emergent level and a mere epiphenomenon can be decided upon whether or not the information embodied at that level constrains physical forces to make efficient cause (same level, upwardly or downwardly). Wherever a physically embodied pattern at level L constrains physical forces arising from level L, L-n, or L+n (where n is any integer), we can say that L is emergent from L-1 and not a mere epiphenomenon.

In this view, what differentiates epiphenomena from real phenomena (i.e. those having causal power) is whether or not they embody information over and above that embodied by their constitutent parts. If not, then the appearance of emergence is no more than an illusion created by course-graining. If there is additional information, then the emergence is a phenomenon that can, potentially, exercise efficient cause (including over its constituent parts). By far the most likely place to find this hard emergence is among the complex systems of life.


 Is information an epiphenomenon?

To answer this, we have to refer to the definitions of information concepts.
There you will see that information is most precisely not a thing, but a measure of a thing (in a way analogous to how temperature is a measure of atomic dynamic energy). References to information in the paragraph above strictly refer to the pattern in the configuration of matter. To be precise, then I will rephrase the question: is pattern an epiphenomenon?

We already established that pattern is most fundamentally the physical location in space and time of physical things relative to one another. It is useful to think of the pattern of different polarities of  magnetisation on a computer hard disk, the string of amino acids in a DNA molecule or the pixels in an image. Clearly we can conceptualise physical information in ways that do not depend on any particular kind of (substrate for) its instantiation. Pattern is conceptually separate from the physical world (elsewhere we call it cybernetic), but its existence still depends on something physical to embody and instantiate it. When an image is printed, a surface is covered with small dots of ink, but when we look at this from a suitable distance, we perceive a pattern formed among the dots: the pattern which is the image, is in our mind, not on the paper. Does this then make it an epiphenomenon, a perception emerging from a pattern of dots?  In short is the description of the configuration of dots as a pattern merely subjective?



Looking at the series of images above, we see a pattern made from individual differences of light level arranged in space (underlying pattern any configuration of differences is data). So, starting with a single difference between level zero and one (a single bit of binary data), we proceed to add more data and gradually a running Irish setter dog appears (it is the symbol used for the Irish national bus service - not as fast as a greyhound, but just as lovable). This appearance is of course entirely subjective. So called artificial intelligence algorithms can be trained to recognise it, but the extent to which any artificial intelligence can 'know' what they have recognised is highly debatable. With our 'real' intelligence we do know what it is and to many, this is the consequence of consciousness (see e.g. Consciousness explained by Daniel Dennett (1991)). 

Still, on the face of it, the Irish setter pattern is mere epiphenomenon, because according to our discrimination rule above, it does not exercise constraint over physical forces so as to make efficient cause (at any level of organisation). However, the perception of it is - fundamentally - a matching of the pattern; first by the observing system, then by its perception process and by that, the pattern is matched to some stored or remembered pattern and the result of this could be an action.

In a simple application, a robot could be watching chocolate biscuits rolling past on a conveyor belt fresh from the coating machine in a biscuit factory. Its job is to spot biscuits that did not get covered in chocolate properly. It perceives a pattern that matches the general shape of a biscuit, but not the correct shape of a properly coated one. In only this case, it activates a sucker on a gantry to quickly swoop down and suck the faulty biscuit out of the way. The question is, did the pattern, formed of pixels in the robot's electronic camera system, have causal power and did the pattern data in its memory and did the algorithms, or maybe all of these? In every case, a positive answer would mean that information pattern is having causal effect in the physical world. The image will have changed from epiphenomenon to information constraint and thereby will have acquired the status of 'real' phenomenon. What this illustrates is that patterns of information are only effective if they are matched (by forming mutual information) in a system that has agency - the ability to act on the pattern it matches.

Taking a more general example, the Universal Turing Machine was first conceived (by Alan Turing) as a mental tool to help understand the basic principles of computation. It is essentially a mathematical concept. The UTMís properties were thought about in abstract terms: the UTM as a concept depended on brains and pencil and paper, it was not a physical thing. Only later was a UTM actually instantiated with a physical embodiment (as a computer), but the UTM is not a physical thing any more than the number four is. The UTM is a mathematical pattern, one that reveals to us many interpretable properties of information. In this example and all others, the properties of physically embodied pattern can be described without reference to the physical elements with which they are embodied. In general, data and pattern are independent of the kind of things that are arranged in space and time to embody them. This is why elsewhere I refer to the purely informational part of a complex system as a  transcendent complex: it transcends the material world and can be embodied in a variety of substrates, or be an abstract concept like the UTM is. The point at which it becomes more than epiphenomenon is when, embodied as a material system, it has the ability to effect efficient causation. It is important, though, not to confuse transcendence with epiphenomenon.

References

Bohm, D. (1980, with multiple reprints). Wholeness and the Implicate Order. Routlidge.

Dennet, D. (1991). Consciousness Explained. Penguin Books.