It’s not just in politics where otherwise smart people consistently talk past one another. People debating whether humans have free will also have this tendency. Neuroscientist and free-will skeptic Sam Harris has dueled philosopher and free-will defender Daniel Dennett for years and once invited him onto his podcast with the express purpose of finally having a meeting of minds. Whoosh! They flew right past each other yet again.
Christian List, a philosopher at the London School of Economics who specializes in how humans make decisions, has a new book, Why Free Will Is Real, that tries to bridge the gap. List is one of a youngish generation of thinkers, such as cosmologist Sean Carroll and philosopher Jenann Ismael, who dissolve the old dichotomies on free will and think that a nuanced reading of physics poses no contradiction for it.
List accepts the skeptics’ definition of free will as a genuine openness to our decisions, and he agrees this seems to be at odds with the clockwork universe of fundamental physics and neurobiology. But he argues that fundamental physics and neurobiology are only part of the story of human behavior. You may be a big bunch of atoms governed by the mechanical laws, but you are not just any bunch of atoms. You are an intricately structured bunch of atoms, and your behavior depends not just on the laws that govern the individual atoms but on the way those atoms are assembled. At a higher level of description, your decisions can be truly open. When you walk into a store and choose between Android and Apple, the outcome is not preordained. It really is on you.
Skeptics miss this point, List argues, because they rely on loose intuitions about causation. They look for the causes of our actions in the basic laws of physics, yet the concept of cause does not even exist at that level, according to the broader theory of causation developed by computer scientist Judea Pearl and others. Causation is a higher-level concept. This theory is fully compatible with the view that humans and other agents are causal forces in the world. List’s book may not settle the debate—what could, after thousands of years?—but it will at least force skeptics to get more sophisticated in their own reasoning.
These questions are not abstract. They underlie the justice system and our day in, day out judgments about praise and blame. It is hard to think about human beings without assuming we are not the author of our own decisions. But the practical need for assuming free will is not an argument in itself.
Earlier this month, List spoke with me from his office in London, and I soon found myself wishing I could take one of his philosophy classes. He laid out the issues step by step and stayed clear of the jargon and rabbit holes into which many philosophers fall.
What really struck me about the book is you give a fair presentation of the case you’re going to argue against. I wish more academics did that. So what is the argument against free will?
A free-will skeptic argues, first, that free will requires one or more properties: intentional, goal-directed agency; alternative possibilities—that it must be possible for me to do otherwise; or causal control over our actions. Then the skeptic claims that those properties are not to be found among the fundamental physical features of our world. Different skeptics focus on different properties.
For instance, some neuroscientists and philosophers such as Patricia and Paul Churchland say we should understand human behavior not so much at the cognitive, psychological level where we invoke explicit mental states, goals, intentions, and purposes, but rather at a lower level of description, as the product of biophysical processes in the brain. Human beings have a strong tendency to ascribe intentions to all sorts of phenomena, like the weather, natural disasters, or rivers. We no longer do that. As brain science becomes more advanced, we may dispose with ascriptions of intentionality even to human beings.
If we are searching for free will at the fundamental physical level, we are simply searching in the wrong place.
The second line of argument says, if the universe is deterministic, as at least some of our best physical theories suggest, then there is no room for alternative possibilities to choose from. Determinism is the thesis that if we fix the complete state of the universe at a particular point in time, only one future trajectory of states is possible. If we fix the state of the world at the time of the Big Bang or shortly after, the entire sequence of events thereafter would be inexorably fixed. When you go to a café and you ask yourself, “Should I have a coffee or a tea?” it would be already written into the initial state of the world which of those two choices you’re going to make.
The third line of argument is that, when I raise a water glass in order to drink from it, it’s not really my intentional mental state that causes this action. Rather, it’s some underlying, subconscious, sub-intentional physical process in the brain. Benjamin Libet found that the conscious decision to press a button is not the beginning of the causal sequence that initiates the process, but there is first a certain pattern of unconscious or subconscious brain activity, and he interpreted this as a challenge for free will.
These arguments have considerable force. One way to respond would be to water down the notion of free will so that one or several of the properties I’ve mentioned are not required. For instance, maybe it’s good enough for free will that I endorse my action and it’s not necessary that I could have done otherwise. Or maybe we must redefine what we mean by alternative possibilities. It might be that I was always going to choose coffee rather than tea, but if hypothetically the world had been a little bit different, I would have made a different choice.
But that’s not a strategy that I find attractive. I am quite happy to concede that free will requires intentional agency, alternative possibilities among which we can choose, and causation of our actions by our mental states. I think the mistake in the standard arguments against free will lies in a failure to distinguish between different levels of description. If we are searching for free will at the fundamental physical level, we are simply searching in the wrong place.
Let’s go through these arguments one by one. What do you say to those who consider the idea that humans are beings with goals and intentions, and that we act on them, a prescientific holdover?
If you try to make sense of human behavior, not just in ordinary life but also in the sciences, then the ascription of intentionality is indispensable. It’s infeasible and not illuminating to explain human behavior at the level of astronomically complex neural firing patterns that take place in the brain.
Suppose I ask a taxi driver to take me to Paddington Station. Twenty-five minutes later, I’m there. The next day, I tell the driver to take me to St. Pancras Station. Now the driver takes me to St. Pancras. If I look at the underlying microphysical activity, it would be very difficult to pinpoint what those two events have in common. If we switch to the intentional mode of explanation, we can very easily explain why the taxi driver takes me to Paddington on the first day, and what the difference is on the second day that leads the driver to take me to St. Pancras Station. The taxi driver understands our communication, forms the intention to take me to a particular station, and is clearly incentivized to do so because this is the way for the driver to earn a living.
The neuroscientific skeptic is absolutely right that, at the fundamental physical level, there is no such thing as intentional goal-directed agency. The mistake is to claim that there is no such thing at all. Intentional agency is an emergent higher-level property, but it is no less real for that. Whenever our best scientific explanations of a particular phenomenon commit us to postulating certain entities or properties, then it is very good scientific practice to treat those postulated entities or properties as genuinely real. We observe patterns and regularities in our social and human environment, and the best way to make sense of those patterns and regularities is by assigning intentional agency to the people involved.
What about the second argument, the challenge from determinism—that, before I walk into a café, it is preordained what I will order?
The jury is out on whether the world is fundamentally deterministic—it depends on how we interpret quantum mechanics—but suppose it is. This does not necessitate that the world is also deterministic at some higher level of description. Indeterminism at the level of psychology is required for free will and alternative possibilities. That is entirely compatible with determinism at the fundamental physical level.
Think about weather forecasting. Meteorologists are interested in higher-level patterns and regularities. In fact, the very notion of weather is a higher-level notion. At the level of individual air molecules, there is no such thing as weather. Perhaps the system at that very fine-grained level of description would indeed behave deterministically according to classical physical laws, but as you move to a more macroscopic description, you abstract away from this microphysical detail. That is not driven by ignorance on our part, but by the explanatory need to focus on the most salient regularities.
It’s infeasible to explain human behavior at the level of complex neural firing patterns in the brain.
When you consider the macroscopic weather states, the system is not deterministic, but stochastic, or random. We can attach probabilities to different scenarios, but it’s not the case that the weather state at the present time fully determines the weather state in a few days’ time. Multiple different trajectories are entirely possible.
Likewise, to describe the complete state of a human agent, we do not describe the full microphysical state of every elementary particle in the brain and body. That would be the wrong level of description. If our best theories of human agency compel us to postulate forks in the road between which agents can choose, then we’ve got very good scientific reasons to take alternative possibilities at face value. If you ask psychologists, cognitive scientists, and economists, they will give you different theories of how human choice-making works. But they all treat human beings as agents who are faced with choices between different options, so all these theories assume alternative possibilities.
Wait, I’m not sure I want indeterminism at the human level. I want my decisions to flow out of my deliberations, not to be the product of chance.
This is subtle. There are different forms of indeterminism. In statistical physics, indeterminism is associated with randomness. But in the social sciences, we use a different kind of indeterminism based on option availability. In decision theory, we draw a distinction between the options an agent could choose and the option the agent will in fact choose, based on maximizing expected utility or some other criteria. If I’m rational, I’m going to try to systematically make choices that are in line with my beliefs and preferences and goals. But the other options don’t disappear. They are available to me right up to the moment of my choice.
A common theme I’m sensing in your arguments is that we should take the lessons of our scientific theories seriously— our theories. If physics says the microscopic world is deterministic, we should accept that, at least provisionally until some better theory comes along. If psychology says humans have genuine choices, we should accept that, too. all
Yes. It would not be very principled to say that fundamental physics is deterministic based on what our best theories of fundamental physics say, while rejecting that weather systems, for example, are indeterministic based on what our best theories of meteorology say. Provided our best theories in higher-level sciences are well confirmed, we have good scientific reasons to take them just as seriously as we’re taking our fundamental physical theories.
Let’s move on to argument number three, that we have causal control over our own acts.
If we accept my response to the first challenge that we human beings are intentional, goal-directed agents, we might still worry that our goals don’t play a causal role. We need to define what we mean by causation. Neuroskeptical arguments against free will invoke causal notions, but often they don’t spell out precisely what they mean.
In the sciences, we test for causation by looking for systematic correlations that remain in place even when we control for other factors. The way in which causal modelers now tend to think about causation—in fields such as statistics, computer science, probability theory, and the philosophy of mind—captures this aspect of scientific methodology. This approach is called the interventionist theory of causation. To say one particular variable causes another is to say that, if we were to intervene on the first variable by changing its value, we would bring about a change in the other variable.
There is no conceptual reason why sophisticated AI systems could not qualify as bearers of free will.
Let’s say I now form the intention to move my arm to lift this glass to drink some water. What should we cite as the cause of this particular arm movement? My intentional mental state—namely, my intention to drink—is very systematically associated with my actions. If my mental state changes, my resulting actions change as well. By contrast, not every change or variation in the underlying physical states would give rise to a change in the resulting act.
A leading philosopher whose work I hugely admire, Jaegwon Kim, raised an important challenge against mental causation, the so-called causal exclusion argument. If you consider a particular effect and you’ve found a cause that fully accounts for that effect, you should not simultaneously postulate yet another distinct cause for the same effect. That would be an act of causal overattribution. Let’s suppose, once again, I lift my arm to drink some water. You can fully account for the action by reference to the physical state of my brain, so there is no reason to postulate yet another cause—namely, a distinct mental cause.
My response, which Peter Menzies and I developed, is that if we accept the interventionist theory of causation, the causal exclusion argument does not generally hold. For any given system, the most systematic causal relations may not involve the lowest-level variables, but could involve higher-level variables, or there might be systematic causal relations at both levels.
Let me see whether I follow, using the taxi metaphor. You have two switches. One is a simple binary Paddington-St. Pancras switch. The other is an analog dial, and you have to fine-tune it to set the destination. It might be set to Paddington, but a slight nudge changes it to St. Pancras, and another nudge changes it back to Paddington—it’s very sensitive. The binary switch is a more effective means of controlling the outcome. It is analogous to seeking a mental cause rather than an atomic one.
What do your views on free will imply? Do machines or even corporations have free will?
In my joint work with Philip Petit, I’ve argued that organized collectives can be intentional agents in their own right, over and above their individual members. Commercial corporations would be examples, if suitably organized. Those group agents don’t have the same cognitive capacities that you and I have, but nonetheless we might say that they have a certain form of free will.
Something similar can be said about AI systems as well. One can have long debates about whether current AI systems are sufficiently advanced, but there is no conceptual reason why sophisticated AI systems could not qualify as bearers of free will. Much like corporate agents, which we also think should be held responsible for their actions, AI systems ideally should display a certain form of moral agency and not just rigid goal-seeking behavior in the interest of profit or whatever else their objective function might be. As we employ more and more AI systems in high-stakes settings, we would like those systems to make ethically acceptable decisions.
George Musser is an award-winning science writer and the author of Spooky Action at a Distance and The Compete Idiot’s Guide to String Theory. Follow him at @gmusser.