This is the third post in a series discussing some key ideas from Why Free Will is Real (Harvard University Press, 2019). Many thanks to John Schwenkler and the Brains Blog for giving me this opportunity.
I have argued that realism about free will – treating it as real, not as an illusion – is justified because the picture of humans as agents with alternative possibilities and causal control over their actions is not just compatible with science but indispensable in some of our best explanations of human behavior. My argument is analogous to the one often given for realism about other properties or entities in science. Physicists are realists about particles, fields, and forces because postulating them is indispensable in our best physical theories. Similarly, biologists tend to be realists about cells, organisms, or eco-systems because postulating them is indispensable in the best theories within their domains. The principle underlying these arguments is the “naturalistic ontological attitude”: if postulating certain properties or entities is indispensable in our best explanations of a given phenomenon and compatible with the rest of science, then we have good reasons for taking those properties or entities to be real. I suggest that, from the perspective of this principle, free will and its prerequisites are no different in their reality than other emergent, higher-level phenomena whose reality we seldom doubt: the weather, markets, economies, and so on.
Yet, readers might wonder whether the picture of humans as indeterministic, choice-making agents is genuinely compatible with a worldview in which the laws of physics could, for all we know, be deterministic. Recall that determinism means that the state of the world at any point in time fully determines the future course of events, like in a mechanical clockwork. If the world is physically deterministic, then only one sequence of events will be physically possible, given the past. There does not seem to be any room for the kind of agential indeterminism that free will requires. Everything that will happen in the future, including all human actions, will be inevitable consequences of the past. We would therefore have to be skeptical towards any theory that implicitly or explicitly postulates indeterminism in human agency.
In this post, I want to explain why this line of reasoning is mistaken. Physical determinism, I argue, does not rule out the kind of agential indeterminism to which our theories of human behavior are committed. Agential indeterminism can co-exist with physical determinism.
Before I explain why this is so, I need to make a few remarks about why all this is relevant. One might think that quantum mechanics, one of our best physical theories, shows that the world is indeterministic. To give a simple example, when a photon, a light particle, hits a semi-transparent mirror with a very sensitive light detector attached, there is a 50% chance that the photon will be transmitted and a 50% chance that it will be reflected. Even the entire past history of the universe appears to be insufficient to determine which of these two outcomes will occur. If this is right, then the debate about whether there could be alternative possibilities in a deterministic universe is of no practical relevance, as our universe is indeterministic from the bottom up. But this conclusion would be too quick. First of all, quantum mechanics is not the final word on physics. Notoriously, it has not yet been reconciled with general relativity theory, which explains phenomena such as gravity, and that theory does not share the apparent indeterminism of quantum mechanics. The jury is still out on whether a future unified theory of physics will vindicate determinism or indeterminism. Secondly, the interpretation of quantum mechanics itself is controversial, and while some interpretations, such as the standard “Copenhagen” one, take it to imply indeterminism, others do not. Rival interpretations include ones according to which some hidden variables determine which trajectory the world is on. In the case of the photon, these hidden variables would have predetermined the photon’s path. This post is not the place to discuss the interpretation of quantum mechanics. I simply want to note that the question of whether agential indeterminism is compatible with physical determinism is of more than hypothetical interest.
So, let me turn to the main question itself. Suppose, for the sake of argument, the world is deterministic at the fundamental physical level. How, then, could there be any indeterminism in human agency? Would this not require a breach of the laws of nature? My answer begins with the observation that the physical level is just one among many different levels at which we may describe and explain the world, and other levels, such as the chemical, biological, psychological, and social ones, are no less important from a scientific perspective. Different such levels give us different windows into reality, and it would be a mistake to consider what we see from some of those windows as less real than what see from others, especially when those windows correspond to well-confirmed scientific perspectives. When we are interested in what humans can and cannot do, the right level at which to ask this question is the level of the human and social sciences, not the fundamental physical one. This point should already be clear from what I have said so far. But now comes a crucial insight. Contrary to what is often assumed, the distinction between determinism and indeterminism cannot be drawn once and for all in a way that applies to all levels simultaneously. Rather, it is a level-specific distinction. The world may be deterministic at some levels and indeterministic at others — a point that may initially sound surprising.
To illustrate this point — as a “proof of concept” — let me introduce a toy model in which a system behaves deterministically at a micro-level and indeterministically at a macro-level (List 2014). Consider a system which, at each point in time, is in a particular state, and where that state evolves over time in accordance with certain laws governing the system. Let’s call the set of all possible momentary states in which the system could be its “state space”. A “history” of the system is a possible sequence of states across time. We can think of the system’s laws as constraints specifying which histories are possible and which not. For example, the possible histories could be as shown in Figure 1, reproduced from my book. In this example, there are six time periods, labelled t = 1, 2, 3, 4, 5, 6. Little dots represent states of the system, and lines from bottom to top represent histories. We can think of the state in the bottom row as the system’s initial state, and we can think of the states along the upward-moving lines as the subsequent states. In this figure, all the possible histories are deterministic. That is to say: the initial state of each history fully determines all subsequent states; there is never any branching in any of the possible histories. We can interpret the states in Figure 1 as micro-states of the system, for instance states that specify the complete configuration of all the physical particles, fields, and forces making up the system at the relevant time. Possible histories then represent the system’s behavior at a micro-level.
Now, let’s suppose that we are interested in the system’s behavior at some macro-level, where the focus is not on particles, fields, and forces, but on certain macro-states. These “supervene” on the system’s micro-states, but are are more coarse-grained, in the sense that the same macro-state can be instantiated by different micro-states – a phenomenon called “multiple realizability”. An example of such a macro-state in physics is a system’s temperature. Different configurations of molecules can have the same mean kinetic energy and thereby instantiate the same temperature. An example of a macro-state in psychology is a mental state such as desiring to eat chocolate and believing there is chocolate available in the kitchen. Plausibly, different neuronal configurations in the human brain could realize that same macro-state. Formally, we can think of each macro-state as an equivalence class of micro-states, consisting of all its different possible “micro-realizers”. In our example, suppose that whenever two or more different micro-states lie in the same cell of the rectangular grid in Figure 1, they instantiate the same macro-state. The relevant equivalence classes are thus given by the cells. While in this toy example there are no more than three possible micro-states for each macro-state, the real systems we study in the special sciences typically admit more complex forms of multiple realizability. In principle, each macro-state could have infinitely many possible micro-realizers, and it might be infeasible to describe what they all have in common from a micro-level perspective alone. Figure 2, also from my book, shows what our toy system looks like at the macro-level. Thick dots represent macro-states, and thick lines from bottom to top represent macro-histories.
It is easy to see that, unlike the micro-histories, the macro-histories are not deterministic here. Regardless of the system’s macro-state at time t = 1, several sequences of subsequent macro-states are possible: the macro-histories exhibit branching. This illustrates that macro-level indeterminism, such as the indeterminism we find in the human and social sciences, can be an emergent byproduct of micro-level determinism. More technically, the property of determinism is not preserved under changes in the level of description, such as when we move from a lower, more fine-grained level to a higher, more coarse-grained one. Crucially, all of this is entirely consistent with the higher level supervening on the lower one. Jeremy Butterfield (2012) has expressed the same point by saying that, in a system that admits multiple levels of description, the system’s micro- and macro-dynamics need not “mesh”. Furthermore, one can not only go from determinism at a lower level to indeterminism at a higher one, but the reverse is also possible. The bottom line is that indeterminism at the lower level is neither necessary, nor even sufficient, for indeterminism at the higher level. (Related results were obtained by Jeffrey Yoshimi, 2012, and, with a slightly different interpretational angle, by Charlotte Werndl, 2009.)
There are, of course, a number of salient objections that one might raise, which I cannot all discuss here. The most common objection is that the appearance of indeterminism at a higher level is merely “epistemic” – the result of our incomplete information about the system’s micro-state. Even if the macro-state at time t = 1 is indeed insufficient to determine the history of subsequent macro-states, the micro-state at time t = 1 would certainly fix all subsequent states, micro as well as macro. And so, what I have called “higher-level indeterminism” is an illusion due to our epistemic limitations. However, as I explain in my book and in some related papers, this objection is mistaken. There are good reasons for adopting an “ontic” and not merely “epistemic” interpretation of higher-level indeterminism – one which treats it as a real phenomenon.
Let me sketch just a few of these reasons. First of all, good scientific practice supports a form of pluralism about levels under which it is appropriate to take a realist attitude towards the properties at each level, provided they are explanatorily indispensable for some relevant special-science purposes. Realism about higher-level indeterminism is arguably supported by this principle.
Secondly, the claim that the system’s micro-state would be enough to fix all subsequent macro-states does not contradict macro-level indeterminism at all. It merely reasserts the already known fact that the system is deterministic at the micro-level. As an objection to macro-level indeterminism, it fails, because the definition of macro-level indeterminism does not — and should not — refer to the system’s micro-states. Macro-level indeterminism means that the system’s macro-state at a particular time does not determine the subsequent sequence of macro-states. This definition is unambiguously satisfied in Figure 2, and it is the right definition in light of the pluralistic case for considering each level on its own terms.
Finally, we cannot assume that there is always a most fundamental level at all, which could then somehow be treated as the privileged level for distinguishing between determinism and indeterminism “simpliciter”. As Marcus Pivato and I have shown (2015), a scenario in which there is a bottomless hierarchy of levels, with determinism at even-numbered levels and indeterminism at odd-numbered ones, is entirely coherent, albeit hypothetical. In such a scenario, it would make no sense to speak of determinism or indeterminism “simpliciter”, or to tie the distinction to any particular privileged level; after all, there is no fundamental level here. The system’s indeterminism at odd-numbered levels is just as real as its determinism at even-numbered ones. This scenario supports the idea that the distinction between determinism and indeterminism is generally best understood as a level-specific one, and it thereby renders an ontic (as opposed to epistemic) interpretation of level-specific determinism or indeterminism plausible.
Another common objection to my argument is that, even if there is agential indeterminism, this only establishes a form of randomness (or pseudo-randomness); but surely, the objection goes, such randomness is not enough for free will. I agree that free will requires more than randomness. But we must not forget that agential indeterminism is only one of three requirements for free will; intentional agency and mental causation are needed too. And more to the point, it would be a mistake to associate agential indeterminism with randomness. Generally, randomness and indeterminism are not the same thing. As my analysis shows, there are different kinds of indeterminism. Some are associated with randomness, for instance the kinds of indeterminism we find in statistical physics. In the human and social sciences, however, there is another kind of indeterminism, which is associated with option availability. In intentional explanations, we draw a crucial distinction between the options that an agent could possibly choose and those that the agent will actually choose (often for intelligible reasons). Agential indeterminism means that the set of possible options is non-singleton (meaning different courses of action are possible for the agent), not that the choice is random. And this is the kind of indeterminism required for free will, as well as the one supported by our theories of agency.
It is time to wrap up. I have argued in this post that the agential indeterminism involved in free will does not conflict with physical determinism. Indeed, the question of whether the fundamental laws of physics are deterministic or indeterministic is completely irrelevant to the question of whether there are alternative possibilities at the level of agency. The latter question is adjudicated, not by fundamental physics, but by our best theories of human behavior, and these — I already argued in my previous post — support the notion of choice between alternative possibilities as central.
There is still one important point on which incompatibilists about free will – those who believe that free will is incompatible with determinism – are right. Free will is not compatible with determinism at the level of agency. If our best theories of human behavior were to give us a deterministic picture of human psychology, thereby refuting the sort of agential indeterminism I have defended, then this would also amount to a refutation of free will of the kind I have discussed. For the time being, however, we have good grounds for thinking that our best theories of human behavior are not like this. They support agency, choice, and mental causation as real phenomena.
This concludes my series of three posts on the naturalistic case for free will. Free will, I suggest, is no less real than many other phenomena that we readily postulate in the special sciences. The mistake in the various forms of free-will skepticism that have recently gained prominence lies in their failure to recognize free will as a higher-level phenomenon, and in their tendency to search for free will at a lower level than the one at which it can be realistically found.
Butterfield, J. (2012). “Laws, Causation and Dynamics at Different Levels.” Interface Focus 2(1): 101–114.
List, C. (2014). “Free Will, Determinism, and the Possibility of Doing Otherwise.” Noûs 48(1): 156–178.
List, C., and M. Pivato (2015). “Emergent Chance.” Philosophical Review 124(1): 119–152.
Werndl, C. (2009). “Are Deterministic Descriptions and Indeterministic Descriptions Observationally Equivalent?” Studies in History and Philosophy of Science Part B 40(3): 232–242.
Yoshimi, J. (2012). “Supervenience, Dynamical Systems Theory, and Non-reductive Physicalism.” British Journal for the Philosophy of Science 63(2): 373–398.
Lovely clear exposition, thank you.
There may be interesting parallels between the physical-determinism versus agential-indeterminism apparent paradox you highlight, and a different apparent paradox discussed in:
The observers’ paradox: apparent computational complexity in physical systems
John F. Kolen, Jordan B. Pollack (1995)
Journal of Experimental & Theoretical Artificial Intelligence 7 (3), 253-269.
In both cases different levels of description support different statements that are apparently in contradiction. But once it is appreciated that such statements are relative-to-the-framework, or to the observer’s current level of description, the anomaly disappears. Paris is to the South (viewed from London). Paris is to the North (viewed from Nice).
I take it that the moral significance of agential indeterminism is that the agent could have done otherwise in the exact situation as it played out but did *not* do otherwise; therefore the agent, should she intentionally behave badly, deserves blame of a sort (deep, ultimate) that would not accrue to a merely deterministic agent. Indeterminism serves to create genuine, metaphysically open alternative possibilities, as opposed to the (mere) epistemic openness of not knowing which of several considered options will play out deterministically. Under agential indeterminism (if I understand it correctly), the agent chooses the particular option in a way that can’t be traced to a sufficient set of determining factors, for if it could the choice would be deterministic and could not have been otherwise given the actual conditions.
What isn’t clear to me however is how, under agential indeterminism, the choice is any more the agent’s, more up to the agent, than under determinism. For me to have control and responsibility, I need to determine a choice according to my desires, intentions and deliberations, so I’m not sure how inserting indeterminism into the choice process would confer more causal control or responsibility. Rather the opposite: it would tend to make the choice process less reliable in reflecting who I am and what I want. I can see how, given indeterminism, a choice might have been otherwise given the genuinely open possibilities, but not how I could bear more responsibility for it, and thus more blame for a bad choice, than under determinism.
List’s argument here only proves that the DESCRIPTION of a state of a system at a higher level may be insufficient to determine its next state; but if we have two exact copies of the same system at the lower (assumed to be deterministic) level, a description of them at a particular state DO totally determine their next states (and determines, obviously, that these two next states are exactly equal, both at the lower and at the higher level). Hence, the causal description at the higher level is just INCOMPLETE.
For example, a full description of the world weather in a particular moment, indicating the temperature, atmospheric pressure, wind speed, etc., in each square meter of the Earth surface, will not be sufficient to determine whether it will rain tomorrow in a particular place; but this does not entail that the atmosphere is NOT a deterministic system, and that the atmosphere has an ontological capacity of determining alternative, really existing possible futures, only that its ‘meteorological’ description is incomplete, no matter how detailed it were (if it remains at the level where ‘meteorological’ concepts apply).
Is quantum theory deterministic?