Reading about emergence and reductionism and free will and determinism has led me to finally confront a concept I had vaguely heard about but never really looked into before: downward causation, a term that came to prominence in the 1970’s. (Some other views: here, here, here.) I think it’s a misguided/unhelpful notion, but this is way outside my area and I’m happy to admit that I might be missing something.
Physicists are well aware that there are different vocabularies/models/theories that we can use to describe the same underlying reality. Sometimes you might want to talk about a box of gas as a fluid with pressure and velocity, other times you might want to talk about it in terms of atoms and molecules. Philosophers and psychologists might want to talk about human beings as autonomous agents who do things for reasons, while admitting that they can also be thought of as collections of cells and tissues, or even once again as atoms and molecules. The question is: what is the relationship between these different levels? In fluid mechanics/kinetic theory things are pretty clear, but in the mind/body problem things begin to get murky. (Or at least, there are people who take great pleasure in insisting that they are murky.)
Reductionism notes that some of these descriptions are more complete, and therefore arguably more fundamental, than others. In particular, some descriptions are in terms of entities that are literally smaller than the others; atoms are smaller than neurons, which are smaller than people. The smaller-level descriptions tend to have a wider range of validity; we can imagine answering certain questions in the atomic language that we can’t answer (correctly) in the fluid language, like “what happens if we divide the box in half, and then divide that in half, and so forth a million times?” It therefore seems natural to arrange the descriptions vertically: “lower” levels refer to small-scale descriptions, while “higher” levels refer to macroscopic objects. The claim of reductionism is, depending on who you talk to, that the lower-level description is either “always more complete,” or “capable of deriving the higher-level descriptions,” or “the right way to think about things.”
The reductionist paradigm is of course heavily resisted in certain quarters. Emergentists like to argue that “more is different,” and that truly novel behaviors emerge at the higher levels. All the argument then becomes about what is meant by “truly novel.” Do you mean “you never would have guessed these behaviors, just by thinking in terms of lower levels”? If so, most reductionists would readily agree. But if you mean “these behaviors are truly independent from what goes on at the lower levels,” then they would not. It is not even really clear what that would mean.
Downward causation, as I understand it, is an attempt to give some oomph to the claim that higher levels are not simply derived from lower levels. Consider the good old mental/physical divide. A reductionist would claim that the mental can ultimately be reduced to the physical. (I’m gliding over various nuanced divisions of opinion in the two-dimensional parameter space of reductionism/physicalism, but so be it.) But an antireductionist might say: “Look, I can choose to lift up my hand and put it somewhere. That’s the mental acting on the physical, with causally efficacious outcomes. You can’t describe this in terms of the physical alone; the higher level is influencing what happens at the lower level.”
That’s downward causation; the higher levels acting causally on the lower levels. If you get spooked by mind/body issues, think of the snowflakes. Sure, they are made of water molecules that act according to atomic/molecular physics. But the shape that they end up taking is highly constrained by the macroscopic crystalline structure of the snowflake itself. That wouldn’t have been visible if you were just thinking about molecules; the macroscopic structure has influenced the dynamics of the microscopic constituents.
I’m doing my best to present this idea sympathetically, but it seems completely wrong-headed to me. As far as I can tell, a major motivation for thinking about downward causation is to preserve the autonomy of mental causation. We think of ourselves as intelligent beings who do things for reasons. We would therefore like to think of the decisions we make as causing certain things to happen in the physical world. But if the mental can be simply reduced to the physical, we might worry that this way of thinking is just wrong. There aren’t “really” mental states that cause things to happen; there are simply neurons and tissues (or atoms and forces) acting according to the laws of physics/biology. Choices and other mental phenomena are just illusions (according to this line of worry). Jerry Fodor put it most vividly:
“If it isn’t literally true that my wanting is causally responsible for my reaching, and my itching is causally responsible for my scratching, and my believing is causally responsible for my saying… if none of that is literally true, then practically everything I believe about anything is false and it’s the end of the world.”
Don’t worry! It’s not really the end of the world.
But before explaining why, let me give a sensible argument that downward causation can’t really work. It’s called the “exclusion argument” (if I’m understanding things correctly), but physicists would simply refer to “closed sets of equations.” The point is that, when we talk about the world in terms of atoms and forces, we have a closed system — any question we can ask in those terms, can be uniquely answered in those terms. (We have the same number of equations as unknowns.) So it can’t be true that we need to account for higher-level processes to follow things at the lower level; indeed, doing so would amount to overconstraining the system, and we would generically expect no consistent solutions. This is how we know that immortal souls require violations of the known laws of physics — those laws are complete by themselves, and aren’t able to support immaterial souls surviving past the body. My language is a little different from that in the philosophy literature, but I take it that this is what’s meant by the exclusion argument.
Why isn’t it, then, the end of the world? I think there are two mistakes being made here. One is to believe that if one phenomenon can be “reduced” to a lower level, then the higher-level phenomenon isn’t “real,” it’s just an illusion. (That’s how I interpret “literally true” in Fodor’s quote.) That’s a very bad way of thinking about the relationship between different levels. This is what I tried to argue in the post about free will and baseball: just because we can think of something macroscopic in terms of its microscopic parts, doesn’t mean that macroscopic thing becomes any less real. Baseball is real, temperature is real, free will is real — all in the sense that they are useful categories for organizing the macroscopic world, whether or not these concepts are nowhere to be found in the vocabulary of fundamental physics.
The second mistake is taking the hierarchy of levels too seriously, with some on top and some on the bottom. (This is related to the previous mistake, obviously.) I would suggest that a better mental image would feature a parallelism of levels with sideways relations between them. So we have a description of a box of gas in terms of atoms and molecules, and another in terms of fluid dynamics. These models sit next to each other, and have arrows moving sideways between them to indicate the map that tells us which configurations in one correspond to which configurations in the other. Sure, one vocabulary may be “more complete” in the sense that it accurately models a wider array of physical conditions, but so what? If another (“higher-level”) description obeys its own autonomous rules of evolution — that is, if we can successfully speak of its properties and outcomes without ever making reference to the any other descriptions (as is certainly true for fluids) — then this description is just as “real” and “literally true” as any others.
I think this way of thinking gets you everything you want. You are allowed to treat mental phenomena (or whatever) as perfectly “real” and causally efficacious. You are also allowed to attempt to “derive” the dynamical rules of one description from the dynamical rules of another plus the map between them. It might be easy, or it might be hard or impossible, but succeeding wouldn’t leech any of the power from the autonomous rules of the “derived” system.
All the mess comes when people try to mix up vocabularies across different levels. You should beware of crossing the streams — total protonic reversal could result, and that would be bad. We can talk about people as animals with minds and reasons, or we can talk about them as collections of cells and tissues, or we can talk about them as collections of protons, neutrons and electrons. It’s only when you start asking “what effect do my feelings have on my protons and neutrons?” that you start getting syntax errors.
This parallelism view gets strong support from dualities in physics. One thing we’ve learned is that you can have completely different descriptions of exactly the same underlying “reality,” but it’s not that one is lower-level and the other is higher-level; they’re simply different. Autonomous vocabularies provide powerful tools for discussing different features of the world in different circumstances. Knowing that you’re made of elementary particles obeying the laws of physics doesn’t make you any less of a person.
@Scott, sounds like you’ve just described the subtle difference between isomorphism and equality. I can only see how making that point being anti-reductionist if reductionism claimed that non-equality implies non-isomorphism, but I doubt many reductionists take such a claim seriously.
@Bee, do you mean that Sean’s argument about the number of equations and unknowns is wobbly or only appears wobbly? I certainly appreciate that it could appear wobbly to the semi-initiated, those familiar enough with infinities to see that there could be a problem but not sufficiently familiar the fact that well-posedness can be still be established in many infinite dimensional systems of equations relevant to physics, once reasonable constraints are imposed on potential solutions (continuity, etc.).
@Phillip#24 and @NickM#15, these are both very difficult examples of code breaking. I think the difficulty there (which I think is unarguable) is much more tied to the notion of what it means to break a code, rather than reductionism per se.
Scott Aaronson > “we could think of natural selection as enjoying a sort of “autonomy” from lower-level physics”
You may be confusing two different kinds of “autonomy”; evolution is “autonomous” from physics in a broad sense: if you have a multiverse of universes, each with different physics, many of them will have evolution. But evolution is not autonomous from physics in detail: in any specific instance of evolution, if evolutionary theory predicts X and physics predicts Y, Y will happen. (For example, an otherwise highly fit genotype could by pure chance cause the DNA molecule which encodes it to become unstable, so fitness would fail to increase as you might predict if your theory didn’t talk about DNA)
As stated by Craig, this conundrum is right in Hofstadter’s wheelhouse.
Hofstadter’s proposition takes your parallel equivalence a step further. He also names the trade off, although not as rigorously as some would like: Moving “Up” hierarchical levels sacrifices some informational resolution, but increases comprehensibility. He, like you, contends that descriptions at either the level of physics or the level of percepts. I really like that you noted that the “higher level” must be a fully autonomous causal system – another way of framing this self-consistency may be to say that the noise in the lower levels is insufficient to affect causal relationships at the higher level.
What’s particularly interesting to me is the relationship between levels of physical reality and “conceptual reality”, if we are to take the mind as a real thing. Hofstadter has a lot to say about this in terms of the emergence from “level crossing” feedback loops, and draws an analogy to the Godel incompleteness theorems, which is sort of a favorite stone to throw at reductionists (Penrose hurls it, as well, in “The Emperor’s New Mind”).
Myself, I don’t think such arguments about level crossing are all that interesting. What is interesting is the structure of information flows at the physical level, and how it relates to the information geometry of the mind. These types of correlations are what are ultimately useful to those with the engineering ambition to build (and therefore understand) minds that include the features of the human mind. The computer example is a very good one, and I’d recommend Charles Petzold’s “Code” for anyone wanting to explore that particular example in excruciating detail.
Rationalist #28:
You may be confusing two different kinds of “autonomy”
I was careful to say a “sort of” autonomy, since I know there’s also an obvious sense in which evolution can’t be autonomous! (Namely, once you’ve fixed all the facts about the laws of physics and the initial/boundary conditions, you’ve also fixed the observable higher-level behavior, at least probabilistically. That’s true essentially by definition: if there were some other determinant of the higher-level behavior, we’d simply add it to our description of physics!)
in any specific instance of evolution, if evolutionary theory predicts X and physics predicts Y, Y will happen
Not so fast: if our theories of evolution and physics were both “perfect”, then clearly their predictions would never disagree! (In particular, a “perfect evolutionary theory” would need to take as input every possible contribution to fitness, which certainly includes DNA-level contributions.)
If, on the other hand, both theories were imperfect, then it’s of course possible that the evolutionist would predict X, the physicist would predict Y, and X would be right (the usual example being Lord Kelvin’s despite with biologists about the age of the sun). So maybe what you really mean is that we can imagine a “perfect” theory of fundamental physics, but not a similarly “perfect” theory of evolution?
No high level phenomena is independent of physics — nothing is independent of physics. The point, however, is that really being able to “explain” some high level phenomena generally is only possible by reference to other high level phenomena. Of course, these other high level phenomena too are not independent of physics. And the physics underlying each phenomena is related — physically, not just metaphorically — but almost always in some very attenuated way so that, as far as we know, even if we had a complete description of initial conditions, the actual physical relationship can never be computed, if only because such computation is practically intractable. Moreover, assuming that it could be computed, what exactly would such a “physical” description tell us about the high level phenomena?
My favorite example of this come from David Deutsch:
“There are explanations at every level of hierarchy. Many of them are autonomous, referring only to concepts at that particular level (for instance, ‘the bear ate the honey because it was hungry’). Many involve deductions in the opposite direction to that of reductive explanation. That is, they explain things not by analyzing them into smaller, simpler things but by regarding them as components of larger, more complex things– about which we nevertheless have explanatory theories. For example, consider one particular copper atom at the tip of the nose of the statue of Sir Winston Churchill that stands in Parliament Square in London. Let me try to explain why that copper atom is there. It is because Churchill served as prime minister in the House of Commons nearby; and because his ideas and leadership contributed to the Allied victory in the Second World War; and because it is customary to honor such people by putting up statues of them;
and because bronze, a traditional material for such statues, contains copper, and so on. Thus we explain a low-level physical observation– the presence of a copper atom at a particular location– through extremely high-level theories about emergent phenomena such as ideas, leadership, war and tradition.
Presumably a reductive ‘theory of everything’ would in principle make a low-level prediction of the probability that such a statue will exist, given the condition of (say) the solar system at some earlier date. It would also in principle describe how the statue probably got there. But such descriptions and predictions (wildly infeasible, of course) would explain nothing. They would merely describe the trajectory that each copper atom followed from the copper mine, through the smelter and the sculptor’s studio, and so on. They could also state how those trajectories were influenced by forces exerted on surrounding atoms, such as those compromising the miners’ and the sculptor’s bodies, and so predict the existence and shape of the statue. In fact such a prediction would have to refer to atoms all over the planet, engaged in the complex motion we call the Second World War, among other things. But even if you had the superhuman capacity to follow such lengthy predictions of the copper atom’s being there, you would still not be able to say, ‘Ah yes, now I understand why it is there.’ You would merely know that its arrival there in that way was inevitable (or likely, or whatever), given all the atoms’ initial configurations and the laws of physics. If you wanted to understand why, you would still have no option but to take a
further step. You would have to inquire into what it is about that configuration of atoms, and those trajectories, that gave them the propensity to deposit a copper atom at this location. Pursuing this inquiry would be a creative task, as discovering new explanations always is. You would have to discover that certain atomic configurations support emergent phenomena such as leadership and war, which are related to one another by high-level explanatory theories. Only when you knew those theories could you understand fully why that copper atom is where it is.”
One thing the lower-level physics and the higher-level “epiphysics” have in common is that they’re carriers of information. Information can be thought of as analogous to energy (it’s not mandatory to think of it that way, but it can be done). Like energy, information is a hypothetical construct knowable only through its manifestations. Like energy, information comes coded and needs a transponder in order to be communicated. (You can’t turn on an air-conditioner that runs on electrical energy by dropping it from an airplane and expecting it to operate on kinetic energy. You can’t yell at me in Nahuatl to watch out for the open manhole I’m about to step into and expect me to heed your warning. There’s no point in screaming at your laptop unless it’s loaded with the appropriate voice-recognition software and is turned on.)
The nexus between brain-physics and high-level functions is an informational transpondence. To understand how it works you need to know all the codes involved. Talk to the fermion.
Sean, thanks for a nice excursus on the validity of different levels of explanation – they can peacefully coexist without one claiming ontological priority as more real than the others.
“It’s only when you start asking ‘what effect do my feelings have on my protons and neutrons?’ that you start getting syntax errors.”
Quite right. And it applies at the level of the neural correlates of feelings (a subset of conscious phenomenal experience) as well: there’s no accepted theory about how the correlates entail feelings, or how feelings could have an effect on their correlates. And it’s difficult to identify one with the other, since after all the correlates are in the public domain, feelings in the private (no one has ever observed a pain or any other conscious experience in the way we observe their neural correlates). So when you say “You are allowed to treat mental phenomena (or whatever) as perfectly ‘real’ and causally efficacious,” it might still be a mistake to suppose that phenomenal experience can play a causal role in neuroscientific or otherwise physicalist explanations of behavior. Rather you’ve got two parallel explanatory tracks, one physicalist and one experiential, and they are mutually irreducible even though highly correlated, http://www.naturalism.org/privacy.htm This solves, or perhaps dissolves, the problem of mental causation when it comes to things like feelings.
“This parallelism view gets strong support from dualities in physics. One thing we’ve learned is that you can have completely different descriptions of exactly the same underlying “reality,” but it’s not that one is lower-level and the other is higher-level; they’re simply different.”
Different, yes, but regarding the above: The phenomenal and physical aren’t different descriptions of a single underlying reality, but (I speculate) *constitute* two parallel “realities” generated by two epistemic perspectives, a subjective phenomenal reality of qualitative conscious states entailed (somehow – the hard problem of consciousness) by being an individual cognizer modeling the world, and a physical reality of non-qualitative concepts and theories generated by the intersubjective project of science modeling the world, http://www.naturalism.org/appearance.htm#part1
Scott, I do agree with the claim that many different lower-level laws can give rise to the same higher-level laws. But I’m not sure what lesson — good or bad — you’re trying to draw from it.
In particle physics, this feature is not only true but obvious and occasionally frustrating. It’s a consequence of the renormalization group; no matter what happens at high energies, low energy physics can be essentially unchanged. (We say that high-energy particles “decouple” from the low-energy phenomena.) That makes us sad, because we can’t simply do arbitrarily precise experiments at low energies to discover new massive particles; we have to build giant particle accelerators that actually reach high energies, and that costs money.
There is a slightly different flavor to things when you move to more slippery realms like computer programs or for that matter texts like blog posts. (I was addressing this a bit in comment #2.) It’s certainly true that very different substrates can represent the same higher-level processes. I think that’s an extremely interesting feature of the world, but don’t necessarily see that it counts as an “anti-reductionist” argument.
Reductionism is the most powerful tool in the history of science, and it seems like almost every “emergent” victory contains some implicit reduction. Still, I could see how it comes to an end.
The Godel proof shows how in mathematics. In physics it may turn out that particle theory is determined by geometry. I don’t know if these examples vindicate holism, per se, but they don’t seem to be examples of classic reductionism either. Probably, in the end, the debate becomes meaningless.
Sean: I don’t think the sensible type of reductionism that you defended in this post (and that I also believe) is challenged in the slightest by “decoupling” phenomena—sorry if I wasn’t clear about that!
Since (like essentially all reductionism-debates among sensible people 🙂 ) this one just boils down to the meanings of words, let’s define “super-reductionism” to be the following theory:
Higher-level phenomena ARE just aggregates of lower-level phenomena. In other words, not only are they causally determined by lower-level phenomena, but they can’t even be properly defined without reference to the lower levels.
My point was just that, if someone was tempted to believe super-reductionism—or if they thought that super-reductionism was the inevitable consequence of ordinary reductionism—then they could be cured by thinking about decoupling phenomena. Of course, this fits in perfectly well with what you said yourself:
just because we can think of something macroscopic in terms of its microscopic parts, doesn’t mean that macroscopic thing becomes any less real. Baseball is real, temperature is real, free will is real — all in the sense that they are useful categories for organizing the macroscopic world, whether or not these concepts are nowhere to be found in the vocabulary of fundamental physics.
I’d go maybe just a hair further: not only is a baseball game as real as the subatomic particles involved in it, but if we “swapped out” the subatomic particles and replaced them by a different physical substrate, it would still be the same game!
I think prime numbers are a good example of emergence. As such they are the intersection of two different numeric ordering systems; division/multiplication and sequential addition. Both are mathematically foundational, but not identical. What this suggests is there many not be a singular bottom level.
Yes, there are smaller levels, subatomic particles, etc, but does size equate to foundational? What if there is no Higgs? Could the effect of mass be generated by field effect of some equilibrium state? What if there is some spacetime geometry affecting the motions of the macrocosm? Wouldn’t it be as foundational as subatomic quantum behavior? If the entire universe originates/exists in a singularity, wouldn’t it be macrocosmically foundational and potentially exert a downward influence equal to the bottom up influence of QM?
The laws governing outcomes may well be deterministic(or they wouldn’t be laws), but is the input a given?
@Scott Aaronson: You know, you have got me thinking now. Could we have such a thing as a “perfect” high-level theory?
Initially I thought not, but actually I think there are such things.
Any logical tautology with a well-defined mapping from predicates (etc) onto properties of physical reality would count.
To the extent that high-level theories are imperfect, the imperfection hides in the details of the mapping.
So you could make a perfect high-level-theory by choosing an artificially simple mapping, e.g. if you have five objects and then create another three, you have eight. For suitably strict definitions of “object”, you now have a perfect high-level theory.
Does this help a believer in causally efficacious free will? Perhaps if you are prepared to say that a law of mathematics or computing causes effects in any physical system which satisfies its axioms… …you could argue that there is some program which constitutes your decision algorithm with the property that your brain is an instantiation of it. But I don’t think that the free will supporter would really want to think of herself as a program any more than she would want to think of herself as a collection of atoms. Certainly, it is an interesting train of thought, and from my reading of Dennett, Gary Drescher, Drew McDerrmot, etc it’s basically the route that some modern determinists/reductionists go down anyway. And I guess it answers the original question: is there downward causation? Well to the extent that you think that mathematics can cause things, there is.
Personally I like my cause and effect to be between concrete objects like electrons.
@ Scott Aaronson,
“I’d go maybe just a hair further: not only is a baseball game as real as the subatomic particles involved in it, but if we “swapped out” the subatomic particles and replaced them by a different physical substrate, it would still be the same game!”
I agree with the point of your posts, but with respect to the above quote, aren’t you just assuming that a particular change to the physical substrate ultimately has no impact on the game? At least in theory, particular changes (or a complex of changes) could “change” the game, couldn’t they?
Of course, as far as we know in most all cases it would be impossible to compute and predict what such changes would consist of, but in a sense that just reinforces the conclusion that higher level explanations do have a certain autonomy.
@Rationalist:
Could we have such a thing as a “perfect” high-level theory?
Awesome question! As I see it, if we have a theory that describes (say) the behavior of some computer program, then not only is that theory not necessarily “approximate” by virtue of computers being complicated, composite objects; it might well be more accurate than our theory of the subatomic particles that the computer is made of! Of course a meteorite could always hit a computer and make it stop working. But crucially, most people wouldn’t call that a defect in a theory about computer programs—they’d just say that the (broken) computer is now a system to which the theory no longer applies.
@Mike:
with respect to the above quote, aren’t you just assuming that a particular change to the physical substrate ultimately has no impact on the game? At least in theory, particular changes (or a complex of changes) could “change” the game, couldn’t they?
You’re right: in this sort of Kripkean thought experiment, the ground rule is that we assume no changes to the observable, macroscopic features of the game. In principle, that might be hard to arrange, if (for example) the decision of which team went to bat first had to be made by observing the behavior of a particle accelerator! So I guess it’s important for this thought experiment that baseball has no such rule.
@ Scott Aaronson,
“In principle, that might be hard to arrange, if (for example) the decision of which team went to bat first had to be made by observing the behavior of a particle accelerator! So I guess it’s important for this thought experiment that baseball has no such rule.”
Can’t disagree — that would be a lousy rule. I was thinking more along the lines that it sure would be hard to specify the physical substrate in such a way as to result in Abner Doubleday (or Alexander Cartwright, depending on your preference) deciding that the base paths would be an oval rather than diamond — it would cut down on infield hits and stolen bases by a large measure and it would certainly be much harder to score from first on a double. Unless, of course, those changes also led him to shorten the distance between bases by some “corresponding” length. 😉
“The Game of Life rules are analogous to the fundamental laws of physics. They determine everything that happens on a Game of Life grid. Nevertheless there are higher level laws that are not derivable from them.
“Certain Game of Life configurations create patterns. The most famous is the glider, a pattern of on and off cells that moves diagonally across the grid. It is possible to implement an arbitrary Turing machine by arranging Game of Life patterns. Computability theory applies to such Turing machines. Thus while not eluding the Game of Life rules, new laws (computability theory) that are independent of the Game of Life rules apply at the Turing machine level of abstraction — just as Schrödinger said.
“Furthermore, conclusions about Turing machines apply to the Game of Life itself. Because the halting problem is unsolvable, it is unsolvable whether an arbitrary Game of Life configuration will reach a stable state.
“Not only are there independent higher level laws, those laws have implications for the fundamental elements of the Game of Life. I call this downward entailment, a scientifically acceptable alternative to downward causation.”
— Russ Abbott, “Reductionism, emergence, and levels of abstractions”
http://arxiv.org/ftp/arxiv/papers/0707/0707.4198.pdf
“Just because we can think of something macroscopic in terms of its microscopic parts, doesn’t mean that macroscopic thing becomes any less real. Baseball is real, temperature is real, free will is real.” – Agreed. Just because I know I am using my prefrontal cortex to make a decision, that doesn’t mean I’m not deciding.
“It’s only when you start asking “what effect do my feelings have on my protons and neutrons?” that you start getting syntax errors.” – Disagree. “Feelings” cause measurable changes in body chemistry, and “feelings” can be altered by conscious choice; otherwise, biofeedback wouldn’t work. Reference the idea of “mind-body integration” as taught by yoga.
Pingback: Downward Causation | Evolving Thoughts