Strong Emergence and Free Will

In this post I will discuss whether the idea of strong emergence can help us to make sense of free will and a few related issues. These ideas have been on my mind lately largely due to the above discussion, but I've been thinking about this for a while.

Background

(You may want to skim or skip this bit if you're familiar with the terms weak and strong emergence, libertarian free will, compatiblism, etc. Or read on if you want to double-check my understanding of same. There's also a couple of useful analogies about tornados and boiling points of compounds that may be thought-provoking)

Emergence is the general idea that we can describe the behaviour of a system at different levels, and we see different types of behaviour at different levels. If we think about a tornado, at a low level there are only individual air molecules bumping around. If we think at an intermediate level, we can see there are regions of high pressure and low pressure, and directions of air flow. At a very high level we can think of the tornado as a single object with a certain shape travelling at a certain velocity. We can say the high level tornado emerges from the low level collisions of air molecules.

It is often useful to use different laws of nature to describe the different levels of behaviour. At a low level, we might use Newtonian laws to model the collisions between particles, or if we want to be very precise for a small number of particles we might even go so far as to use the Schrödinger equation. At an intermediate level, we might use the Navier-Stokes equations to model fluid dynamics. At a high level we might have mathematical models of the ways tornados tend to move and behave.

For tornados at least, the assumption is usually that these higher level descriptions are instances of weak emergence. This means that the higher level behaviour is in fact fixed by the facts and the laws at a low level. We imagine that an infinite intelligence knowing the low-level laws and the low-level state would be able to use those laws to simulate the entire system and so predict not only the behaviour of individual particles but of the whole tornado. The reason we need the higher level descriptions in practice is then pragmatic and not fundamental -- it is only that high level descriptions are vastly more computationally efficient than low level descriptions, and not that they are required in principle to model high level behaviour.

However, until we actually derive the high level from the low level in a computation of some sort, this remains an assumption. I'm pretty sure, for instance, that nobody has been able to calculate the boiling point of water at standard pressure from first principles of particle physics, because it is absurdly computationally difficult to do so with current technology and analytic techniques. We have faith that it should be possible in principle, but we cannot be sure until we do it.

An alternative logical possibility is that there is just a fundamental law of nature governing boiling points. This law might give the result that water boils at 100°C at standard pressure, superceding anything we might be able to calculate from particle physics. This would be an example of strong emergence -- where high level descriptions and laws do not depend only on the laws of physics governing particles, but also on distinct high-level laws that in some way overrule or at least act in concert with the low-level laws to govern high-level behaviour.

We have little reason to believe that there are such strongly emergent laws governing the boiling points of chemical compounds. Where strong emergence perhaps has the most purchase is in discussion around free will. We all feel like we have free will and are responsible for our own decisions and behaviour. However, at least at first, this seems to be in tension with the idea that our behaviour is determined by the low-level physics of particle interactions. There are three common broad approaches to resolving this seeming conflict.

Compatibilism: We can reconcile the two pictures. Free will exists and we are responsible for our actions. Free will is a weakly emergent phenomenon ultimately arising from from particle interactions.

Hard Determinism: There is no free will. We can divide this further into fatalism, where everything is pointless and reasons are fantasies, or something more like compatibilism, where we might deny free will but we still try to make sense of something like reasons and responsibility and decision-making.

Libertarianism: Our behaviour is not governed by low-level physics. We are responsible for our own actions in a way that cannot be reduced to lower levels of description.

Personally, I'm somewhere between compatibilism and hard determinism. I think our behaviour is weakly emergent from low-level physics, but even so we can make sense of concepts analogous to decision-making, personal responsibility and free will. In such a world, I think it's a matter of convention whether we describe it by saying that free will exists or that it doesn't, and I have no strong preference. It depends only on whether we mean "compatibilist free will" or "libertarian free will".

But it's the third option I want to discuss, and in particular how strong emergence does or does not help make sense of it.

Strong Emergence and Libertarianism

Perhaps a good place to start on this is the recent Mind Chat discussion at the top of this post between Philip Goff, Keith Frankish and their guest Helen Steward, who defends a thesis something like libertarianism (though she doesn't like the term, preferring to call herself an "agency incompatibilist"), making use of a strongly emergent idea of agent causation. I confess I have not yet had an opportunity to read her book, A Metaphysics for Freedom, but I have done my best to understand her viewpoint by following the aforementioned discussion and reading a little about it on reviews such as this. As such, I'm open to correction if I'm getting some of her approach wrong.

Steward's view can be regarded as an instance of strong emergentism, in that she argues that agents such as humans and animals are the causes of their behaviour in a way which cannot be reduced to low-level physics. On this view, the behaviour of agents can only be explained with reference to concepts such as reasons and desires, and we will not get a correct picture of reality by simulating the low-level laws of physics alone. She describes a world where you could perform such a low-level simulation successfully as mechanistic and Laplacian (after Laplace's demon), and says that we no longer have any reason to think that the universe is like that since Newtonian physics has been overturned.

As far as I can tell, the main motivation for denying that the universe is mechanistic is that we are obliged to believe in agent-level concepts such as free will, reasons, etc, while pointing at some difficulties in the compatibilist program for reconciling these vital concepts with a mechanistic universe.

We are obliged to believe in free will in part due to the intuitive difficulty in dispelling such notions, but more importantly because the structure of society depends on notions such as personal responsibility and agency.

The difficulties with compatibilism seem to be that overly simplistic models of decision-making in a compatibilist framework cannot make sense of various corner cases. For example, a compatibilist might say that we perform an action intentionally if the action is caused by our beliefs and desires. However, consider the following case proposed by Donald Davidson in 1973.

A climber might want to rid himself of the weight and danger of holding another man on a rope, and he might know that by loosening his hold on the rope he could rid himself of the weight and danger. This belief and want might so unnerve him as to cause him to loosen his hold, and yet it might be the case that he never chose to loosen his hold, nor did he do it intentionally.

According to Steward, given that we have such reasons to prefer something like strongly emergent agency causation, then the burden of proof should be on those who say that the universe is mechanistic.

I don't want to get into these motivations too much but I'll briefly say that just because society depends on the belief in agent causation doesn't really weigh in favour of its existence, any more than a society's dependence on a religious framework means that God exists. We may wish it existed, but believing something because we want it to be true is the informal fallacy of wishful thinking. Secondly, just because it is difficult to make sense of compatibilism in some corner cases doesn't mean that it is impossible to do so. Even if we cannot, it may still be the case that compatibilism works fine most of the time, which is perhaps all we need to build a functioning society. All-in-all, it seems to me that the reasons to believe in agency causation are extremely weak and we should disbelieve in it on grounds of parsimony alone. We have no need to postulate agency causation to explain any of our empirical observations, so we should not.

But setting all this aside, let's entertain the idea that agency causation does exist. Some interesting issues arise.

The Issues

Three interesting and related issues in particular have been raised in recent discussions I've followed or taken part in on this topic.

1. Is strong emergence compatible with naturalism?
2. Does strong emergence imply that our current understanding of physics is incorrect?
3. Does strong emergence provide a sound basis for libertarianism?

Strong Emergence and Naturalism

Proponents of libertarianism are often taken to be denying naturalism, the proposition that supernatural phenomena do not exist. This is particularly so if they talk of a role for a soul, a spirit or a non-material mind of some kind in decision-making. Many of them are happy to accept that they are proposing a supernatural world view, particularly if arguing from a theistic perspective.

This is not the case for Helen Steward. She believes that if agent causation exists, it is a natural phenomenon, and so libertarianism is compatible with naturalism in her view.

However I think we need some sort of criterion to distinguish what we mean by natural or supernatural. A believer in ghosts, prayer or telepathy could claim to be a naturalist by saying that these are natural phenomena. If we can call anything "natural", then "naturalism" seems to be an empty position. I've delved into this topic earlier and concluded that natural phenomena are those that can be said to  be governed by, supervene on, or weakly emerge from laws of physics that are expressible mathematically. This means that, in my view, strong emergence is only compatible with naturalism if the laws on which it ultimately depends can be expressed mathematically. Even ghosts might be natural, but if so, they must be understood to arise somehow out of the laws of physics, whatever they may be. The hallmark of the supernatural is to suggest that some things are irreducibly mysterious and cannot be understood by the reductive methods of naturalistic science.

Note that I posit weak emergence as part of a naturalistic world view, but this doesn't rule out strong emergence. As long as the strongly emergent laws are expressible mathematically, then they can act as a basis for further higher level phenomena to emerge from weakly.

I think it is plausible for a mathematically robust strongly emergent law to govern the boiling of water. It's not too hard to imagine some absolutely objective criterion for what constitutes a molecule of water, and strongly emergent laws that describe how it interacts with other similar molecules and so fix the boiling point of water.

It's a little less easy to see how strongly emergent laws would apply to agents. A water molecule is pretty clear cut. There is relatively little ambiguity about what counts as a water molecule and what does not (though there is some, e.g. whether heavy water counts). But mathematically rigorous strongly emergent laws to describe the behaviour of agents would necessitate that there be objective criteria by which we could unambiguously say what counts as an agent and what does not. This needs to be absolutely clear, as the universe needs to "know" when to apply the laws for particles and when to apply the laws describing agents. But this seems implausible -- even Steward admits that it's not clear what should count as a mechanism and what should count as an agent, e.g. which category appluies to bacteria.

Once you've identified what counts as an agent, you then need to have some mathematically precise laws to describe how the agent behaves. This might take the form of some sort of algorithm, for instance. This is perhaps less of a problem, but it does lead to the rather unintuitive picture where the high-level algorithm a system is running directly influences the low-level "hardware" running that same algorithm. This kind of top-down causation should seem thoroughly alien to a computer scientist such as myself, but it's hard to rule it out as a logical possibility.

So strong emergence itself does seem to be compatible with naturalism, but agent causation seems a bit of a stretch.

Strong Emergence and the Core Theory

This topic came up in an earlier Mind Chat with Sean Carroll. It was a long and interesting discussion, but much of it was taken up with a somewhat unproductive impasse between Goff and Carroll on whether strong emergence would be in conflict with the core theory of modern physics, composed of the standard model of particle physics taken together with general relativity. They are due for a rematch on this topic in a few months, which I look forward to greatly. Briefly, Goff thinks that strongly emergent laws could co-exist with the core theory without overturning it, but Carroll disagrees. Carroll says that if strongly emergent behaviour exists in the universe, then physics as we know it is simply wrong. That doesn't mean of course that strongly emergent behaviour is ruled out, but he does want to be clear about what that would mean for our understanding of physics.

I think this is perhaps a semantic disagreement on what it means for a theory of physics to be incorrect. We might want to say that a theory is correct but incomplete if it has a limited domain of applicability. Nancy Cartwright (if I understand her correctly) has proposed that nature may be a patchwork of different domains wherein different laws apply, each law being correct within its own domain. This picture might help to reconcile issues such as the famous incompatibility of quantum mechanics with general relativity, were it not for the fact that there seem to be situations where the two domains overlap (such as inside black holes or near the Big Bang). Agent causation would fit well with this patchwork view, as it proposes a new domain wherein the core theory no longer quite applies and we need a theory that accounts for agent causation instead. The point being that the core theory would still be correct in its own domain.

The problem with this view is that the core theory purports to completely describe what all particles do everywhere in the universe. It doesn't recognise any limits to its domain. If the world is a patchwork, then the core theory is incorrect for this reason alone.

Another tack may be that the core theory and strongly emergent laws work together somehow. On this view, the core theory is indeed acting on the particles in our brain, it's just that there are also influences from strongly emergent laws. On this view, again the core theory is correct, just insufficient. But again, this amounts to refuting the core theory, because the core theory purports to be an exhaustive account of the influences on particles.

I'm more in sympathy with Carroll here. If strong emergence exists, then the core theory is wrong, just because the core theory as interpreted by physicists purports to be complete. Even so, if we accept Goff's position that a rejection of the core theory's pretentions to completeness does not amount to a refutation of the core theory itself, then there is a sense in which the core theory could still be correct in certain domains even if strongly emergent phenomena do exist. I suspect Goff (and Cartwright) may be underestimating the difficulty of making such a patchwork coherent though.

Strong Emergence as a basis for Libertarianism

I think there's a clear problem with naturalistic and strongly emergent agent causation as a basis for libertarianism, which is that if it is naturalistic, then it is necessarily mechanistic. We may not be able to explain our behaviour by working up from the low-level physics of particles, but we will be able to do so by working up from the higher-level laws of agent causation, whatever they turn out to be. As far as I can see, all the same issues arise here as arise here for agents weakly emerging from low-level physics. Both visions are ultimately equally mechanistic.

If behaviour weakly emerges from low-level physics, we can tell a story where this agent made a choice not because she was "bad" and "deserves to be punished" but because these particles interacted with those particles. If it emerges instead from laws of agent causation, then we can tell a story where the choice was because of the algorithm the agent was running. The exact same issues arise. Without agent causation, we have the problem that an agent is not ultimately responsible for the state or interactions of the particles driving her behaviour. With agent causation we can similarly say that the agent is not ultimately responsible for whatever the state is in the agent-causation picture. Just as with particle-level physics, the agent's behaviour is ultimately caused by mechanistic prior causes going back all the way to the Big Bang. Everything is still explained by mechanisms built upon the laws of physics. The only difference is what form the true description of the laws of physics take, and whether those laws of physics recognise her as an agent.

In the Mind Chat episode (1:55:22), Steward is posed a question from an audience member who may or may not be yours truly.

Can we simulate or model down top-down causation? If so how is this not deterministic? If not then it seems the rules or principles of top-down causation are not well defined.

Steward's response:

I'm not the sort of person who could write the program that's for sure! But I would imagine it could be simulated, and I'm a bit puzzled by the suggestion that that would make it deterministic because as I understand it simulation is often most valuable and useful in connection with systems that are thought to be indeterministic, or at least full of massive chaos

First of all, this is an unprepared off-the-cuff response, so we should be charitable. It looks like she has not considered this suggestion before. I assume that had I more time to explain the issues I perceive, and had she more time to explain her position, then we might do better to get to the heart of the issue.

In particular, we should not take her to task too much for conflating chaos and indeterminism. I completely agree with Keith's follow up explanation that chaos is not at all incompatible with determinism.

But I think the issue here is not really with determinism per se but with whether we can capture the behaviour of agents mechanistically. If we can indeed faithfully simulate it, even if we do so by introducing a little indeterminism in the form of pseudorandomness (or even real randomness coming from some sort of quantum measurement), then what we are doing is implementing agent behaviour mechanistically.

Perhaps Steward would want to say that though we could simulate agent causation, such a simulation would not in fact give rise to a true agent, but just a simulacrum of one which may be good enough for many purposes even if not perfect. If so, we're back to the same dilemma. Either we can make an absolutely perfect simulation of an agent, in which case agent causation is mechanistic, or we cannot, in which agent causation is not well-defined (and starts to look supernatural by my lights).

Perhaps Steward would say that a simulated agent would have behaviour indistinguishable from a true agent, but would nevertheless not be a true agent. In that case, it is possible to get perfectly agent-like behaviour without true agents. This means that we have no reason to suppose that true agents exist -- because any evidence we could find to support the existence of true agents would also support the existence of pseudo-agents. In particular, there would be no way for evolution to select for true agency in favour of pseudo-agency. Rather as in the philosophical zombie argument for phenomenal consciousness, the only reason we have for supposing that true agents exist at all is unevidenced personal conviction that we ourselves are true agents, which is rather unsatisfactory.

As such, it seems to me that we are forced to conclude that agent causation is either supernatural or redundant. Strongly emergent naturalistic agent causation does not work as a basis for libertarian free will.