Confusion over causation, both top-down and bottom-up

I’m becoming convinced that many disputes in the philosophy of science are merely manufactured, arising from people interpreting words to mean different things. A good example is the concept of “reductionism”, where the meaning intended by those defending the concept usually differs markedly from that critiqued by those who oppose it.

A similar situation arises with the terms “top down” versus “bottom up” causation, where neither concept is well defined and thus, I will argue, both terms are unhelpful. (For examples of papers using these terms, see the 2012 article “Top-down causation and emergence: some comments on mechanisms”, by George Ellis, and the 2021 article “Making sense of top-down causation: Universality and functional equivalence in physics and biology”, by Sara Green and Robert Batterman.)

The term “bottom-up” causation tends to be used when the low-level properties of particles are salient in explaining why something occurred, while the term “top-down” causation is used when the more-salient aspect of a system is the complex, large-scale pattern. But there is no clear distinction between the two, and attempts to propose one usually produce straw-man accounts that no-one holds to.

Let’s start by adopting the usual counterfactual definition of “causation” as being good-enough for current purposes, namely: happening A causes the later happening B if, had A not been the case, then B would not have occurred. There are then multiple “causes” of B, any number of things that, had they not been the case, would have meant that B didn’t occur.

I’ve previously defined reductionism — that is, the only form of reductionism that people defend — using a thought experiment: imagine a Star Trek-style transporter device that knows only about, but everything about, low-level particles, atoms and molecules. Crucially, in order to work, it needs to know the pattern of how the particles are arranged. Thus, for each atom, the device needs to know where it is in relation to its neighbouring atoms (and piecing such information together would then completely specify any larger-scale pattern). Reductionism asserts that, if such a device replicates a system, then the replicated system would exhibit the same high-level behaviour as the original (even though the “high-level behaviour” is not something the transporter device knows about).

Again, the fact that the larger-scale pattern is replicated is crucial. Let’s suppose the transporter device instead included a scrambler, so that, when it replicated each atom of a system, it moved it a random amount in a random direction, scrambling any pattern. Obviously the replicated entity would be very different from the original and would behave very differently. Captain Kirk is very different from a random heap of the same number of the same types of atoms.

What “causes” the behaviour of the replicated system? Again, one can assign the term “cause” to many aspects of the system. One can note that, were the low-level atoms of a different type, then the high-level behaviour could be different; or one can note that, were the arrangement of the atoms different, then, again, the high-level behaviour could be different.

So why might the terms “bottom-up” or “top-down” causation be useful? One might, for example, use the term “bottom-up” causation if one thought that only the type of atom mattered, and that their larger-scale arrangement did not. But no-one actually believes that.

Or perhaps one might use the term “top-down” causation if one supposed that the high-level pattern were the only thing that mattered, and that the nature of the particles was irrelevant. But, again, I’m not sure that anyone believes that in the general case. It is true that one can often build a functionally equivalent system out of a range of different substrates, such that more than one type of atom could suffice in producing a given high-level behaviour. But it does usually matter whether an atom tends to form four covalent bonds (like carbon) or two (like oxygen), precisely because that affects how it interacts with its neighbours, and thus affects the larger-scale pattern.

While high-level function is often multiply realisable, it is never the case that the component parts become irrelevant, and that an account of causation in terms of patterns of parts becomes invalid. If one were to define “top-down” causation as the doctrine that you could swap in any low-level component and it would make no difference, then clearly that doctrine would be wrong — you can make a spanner out of iron or nickel, but not out of nitrogen or neon.

Alternatively, one might reject the concept “top-down causation” if one interpreted it as denying that the above transporter-style reductionism would work. Thus, if one identified “top-down causation” with, say, a non-material, dualistic “soul” that was telling matter how to behave, then one might reject “top-down causation”. But I suspect that few people would argue for such these days.

So why might one reject “bottom-up” causation? For a long while I was puzzled as to what people disliked about this concept. That was clarified for me by a Tweet by Professor of Neuroscience Kevin Mitchell, who gave an example of “bottom-up” causation as being the behaviour of a flock of starlings. In such a flock the larger-scale pattern is incidental, even if spectacular. That is, the flocking behaviour is explained by the decision-making of each individual bird, and its spatial awareness of its near neighbours, but with no larger-scale organisation. Any large-scale pattern appearing in the murmuration is then serendipitous — in complete contrast to the large-scale pattern of material that we call an “elephant”.

Extending the above thought experiment, suppose we instantaneously moved each individual starling a random number of inches in a random direction, and then let the flocking continue. A minute or so later we likely could not tell whether or not we had done that. Nothing essential about the flocking behaviour would have changed. That is very different from the elephant, where moving the constituent atoms about randomly would destroy the elephant-ness of the pattern.

So, if bottom-up causation implies that only the low-level matters, with the larger-scale pattern being unimportant, then obviously we reject bottom-up causation (other than for a few special cases such as the flocking behaviour of starlings). And if “top-down” causation means that the large-scale pattern of the material in a system is crucial then obviously we plumb for top-down causation.

But I’m fairly sure that no-one has ever advocated bottom-up causation in that sense as a general account of how the world works, it’s a strawman. If someone argues for bottom-up causation and against top-down causation they’re instead likely rejecting an immaterial soul or some-such notion. And I’m fairly sure that most disputes over such terms arise from mis-construing what other people intend them to mean.

Such disputes sometimes — and quite erroneously — get translated as being about how a physicist might see things (bottom-up?) versus how a biologist or social scientist might see things (top down?). And, of course, the large-scale patterns are exactly what the biologist and social scientist study (life-forms being complex self-replicating patterns of simple components), whereas the physicist is more concerned with the material that such patterns are composed of.

But it is crucial to realise that the importance of the arrangement of atoms, and thus of the “top-down” view, holds just as much in physics. The different forms of carbon (graphite, graphene, diamond, fullerenes) are made of exactly the same stuff, only the arrangement of the atoms differs. And so the different properties of those materials (hardness, ductility, conductivity, transparency, etc) are attributed to the different arrangements of the atoms, not to the properties of the atoms themselves. If you tried the “randomly move the atoms” thought experiment with diamond you’d then have amorphous carbon, not diamond.

The importance of the different patterns of low-level atoms holds throughout physics, though of course the patterns that physicists study are vastly simpler than the patterns we call “elephants”. Indeed, as Feynman stated at the start of his “Lectures”, the most important concept in all of science is that everything is made of atoms, simple little particles that are all pretty much the same, and that it is the vast range of arrangements of these atoms that give rise to the complexity of the world.

To sum up, both of the terms “bottom-up” and “top-down” causation are unhelpful, since they can only be distinguished by reference to strawman ideas that no-one actually holds to. We shouldn’t use such terms as though they have a clear and agreed-on meaning. Instead, we need to clarify our meanings by specific and concrete examples, and a good way of doing that is through thought experiments. I’ve attempted to do that here with the idea of scrambling a pattern, contrasting the effect by randomly moving atoms in an elephant versus randomly moving starlings in a flock.