The philosophy of free will is fascinating.
You have a clash between an inherent feeling most of us have – that we act freely in the world – and the fruits of our collective knowledge – science and deep thinking over millennia.
It is central to political discussions and questions of moral responsibility.
Many of the challenges we experience with artificial intelligence – explainability and predictability – overlap with the ongoing questions of free will.
Yet there are many who feel that it does not exist or argue that our common-sense knowledge is wrong. Just what is going on?
This post follows up a previous deep dive on the topic – I’d start there before coming back here for a revisit.
In the previous post, we looked at the terms “free will” and “determinism”. We found that they are bed-fellows. Much of the debate in free will philosophy concerns whether these two things are compatible.
Determinism is often the easier of the two concepts to unpack and define but is possibly more obscure to most people. A good primer can be found on the Stanford Encyclopedia of Philosophy. My preferred definition is: a reality where causes lead to effects, and a number of natural (and possibly local) rules of physics are met. When you “determine” something, you (solely) fix or control its state. The old Latin root is informative – it comes from “de” – completely + “terminare” – terminate, with terminate coming from terminus meaning an end or boundary and coming to mean fixing or limiting, i.e. bringing to an end. From this, determinism came to mean that the “end”, in the form of the present moment of local reality, was completely fixed by what came before, in the form of causes.
Now apart from those that believe in the supernatural, many will not have a problem in agreeing that the present state of things is influenced by the past. The more questionable part of determinism has thus tended to be the “de” – that the present state is completely fixed by the past. This can be looked at as removing personal choice and responsibility – if you are completely set by the past, where is the freedom to act otherwise?
Lately, determinism has been contrasted with indeterminism. Indeterminism is more problematic than determinism, because it has multiple somewhat contradictory interpretations.
Indeterminism is often used to refer to one or more of the following:
- the concept that not all events are wholly dependent on antecedent causes;
- an opposite or the negation of determinism; and/or
- uncertainty or unpredictability.
Here, you can see how the first two meaning can sometimes overlap: the concept that not all events are wholly dependent on antecedent causes may be considered as the negation of determinism, where all events are wholly dependent on antecedent causes. However, you can also see that the last meaning need not entail the other two – you could have a world where events are wholly dependent on antecedent causes, which may be a determinist world, but where a future state of that world is still uncertain or unpredictable. In this case, indeterminism and determinism are not necessarily opposed; a world may be indeterminist yet determined.
Listening to the philosophical debates, I often find philosophers using the term but relying on the third meaning. I think this third meaning is the right one to pick for useful analysis, but it’s also slightly problematic with the surface syntax of the term – “in” is normally prefixed to indicate the negation of something. Hence, I think it’s better to refer to uncertainty or unpredictability and determinism without using the term indeterminism.
You could also use the term probabilistic determinism to refer to a view where the past completely determines the present and the future, but where states within the past, present and future are only knowable probabilistically, i.e. in terms of different possible states with different possible weights. Some feel that this is an oxymoron, but this is only so if you interpret determinism to mean that a perfect knowledge of the past is available (and knowable), and that the future is a fixed function of this knowledge.
Determinism and Quantum Physics
Floating around the edge of the philosophy of free will since the early twentieth century is quantum physics. The philosophers’ interpretation of quantum physics is that the universe is indeterministic – you cannot predict when an atom would decay, or the position of a photon. But all quantum physics really says is that a quantum state is representable using a superposition of eigenvector bases, wherein on measurement by an observer the quantum state collapses to just one of the eigenvector bases.
The determinism of physics specifies that if a present state of an object and the laws of physics are known, the future can be predicted. The indeterminism of quantum physics specifies that the present state of an object can only be probabilistically known, which means the future cannot be perfectly predicted. Or sometimes the indeterminism is put as, even if you know the (classical) state of an object, you cannot know for sure the future state.
However, quantum physics doesn’t actually say that anything goes. It does say a photon can suddenly turn into an elephant, or that the speed of light magically changes with the seasons. The actual range of potential states (the eigenvector bases) are actually pretty specific and restricted – what quantum physics is actually mainly saying is that we cannot know for sure which of those states is the current one – indeed, particles can be said to exist in all the states at the same time.
Going back to our definitions of indeterminism, we see how the first meaning – that not all events are wholly dependent on antecedent causes – fits with quantum physics – even if we measure all previous states (the antecedent or earlier causes), and have perfect knowledge of all the variables we thing we are measurable, we still cannot predict a future state (in the form of a future event). Put another way, a future state cannot be wholly dependent on previous states as even though we know previous state “perfectly”, we cannot with 100% accuracy predict the future state.
There is also quantum non-locality or quantum entanglement. I prefer the term entanglement as even though two quantum particles may be measured and have correlated properties, they are generated or interact locally at one point in time to allow the correlated properties at separate locations at a later point in time. Unfortunately, this doesn’t at present allow us to perform faster than light communication, rather the two later measurements are correlated due to the entanglement regardless of the distance.
Indeterminism is closely linked to randomness. Again, we often see problems in the use of the term “randomness” that in turn cause problems for free will.
One misinterpretation that is often made is that “randomness” means completely unpredictable, in a similar way to how determinism means that an event is completely caused. But a random variable is a variable whose value falls within a known event space (a set of predefined outcomes), where the selection of the value, within a group of values, follows a particular pattern (a probability distribution). In this sense, like quantum physics, there are degrees of not knowing and in actual fact, in many cases, we do know something: we know (or can measure beforehand) the event space and we can often know or measure the probability distributions that govern the behaviour of multiple measurements.
Now the statistical version of “randomness” does seem a little out of line with our lay interpretation of “randomness”. The latter is often used to mean “unexpected” – we think an event is random when the event has an outcome that was not predicted beforehand. This doesn’t sit as well with the concept of a known event space. But the statistical version of “randomness” does sit fairly well with “free will” – we feel that when we act freely, we are “choosing” from a know set of predicted outcomes. This feeling is retained despite the fact that often the result of our “free” actions varies from the predicted outcomes – what seems important is that, at the time of action, we have a set of more than one outcome and that we can “select” an action that aims for one outcome rather than another.
Why Does It Matter?
Let’s assume the universe is composed of local interconnected events that play out in time a set way but that cannot be completely predicted beforehand.
This immediately kills some of the baggage of the “multiverse” and “counterfactuals” – they become intellectual tools rather than aspects of reality. It also leaves room for freedom via a lack of complete knowledge – most of our fears over a lack of free will are actually fears that someone can know what we can do and that we have a lack of control over our lives. This assumed form of the universe provides a freedom of action in that we can never know what someone will do beforehand; they can always surprise us and there can always be the hope of redemption. It also doesn’t do away with control. Although, our local area of spacetime will play out in the way it will play out, there is room for the way it will play out being a way that is influenced by the actions that we take. Control returns via the route of practical semantics, in the way we talk about events and act based on them and knowledge.
This view also fits better with certain aspects of physics, namely the bi-directional symmetry of many models of the world. It can be true that the present influences the past as much as the past influences the present; it simply depends on your framing. The past is as imperfectibly knowable as the future. Hence, as the present unfurls it changes our knowledge of the past, making certain states more likely and certain states unlikely. As time and distance from the past increase, our knowledge of it also decreases, until we just have simplistic models (stories) that allow for about the same range of possibilities as our stories of the future.
The view also fits with Gödel’s theories of completeness. We can never set out a set of axioms that allow us to divine, machine-like, the future. There will always be facts that don’t fit the models, things we cannot describe, surprises.
The risks of this assumed form is a descent into fatalism – that our lives will unfurl a particular way, whether good or bad, and we are simply inert passengers on this ride. However, the lack of knowledge means we can never know the way, so we can never know how our inaction or action interacts with the world that will become.
The lack of knowledge is not a complete lack of knowledge. We can still be aware of patterns in the world, just not how they interact to entirely and accurately predict the future. A ball thrown in the air still traces a parabola, but can still be blown off a perfect curve by an unexpected gust of wind, or be steered in its curve by the Brownian motion of air molecules and other particulates. We can still make good guesses. We can still know that undertaking a particular course of action will more likely unveil a particular future world. That future world has always been there but we also choose the form it takes.
I used to believe in a block universe. However, a better understanding of relativity, quantum physics and perceptual neuroscience shows that this conception doesn’t fit our broader knowledge. Relativity teaches us that events and objects are linked by threads that are constrained by the speed of light, and that mass manipulates time such that space and time cannot be cleanly separately. Quantum physics suggests a chaotic, messy, unknowable, and discrete fabric of low-scale reality. Perceptual neuroscience shows that our perceptive equipment is shaped by local aspects of a geography that extends only in relation to our own scale: the dance of the solar system is as unintuitive as the notion that solid objects are the interaction of electron shells. A better model for the universe is an evolving network that can be viewed at multiple scales.
Without the baggage of a human-like deity, east Asian understanding possibly comes closest to an accurate model of the world. But the core message of redemption in Christianity also applies – there is always hope that our most negative predictions of the world are wrong, and that people and events surprise us on the up side. Of course, there is nothing inherently positive about how the future will unfold, pessimists may equally contend that our most positive predictions of the world may also be wrong, and that people and events may surprise us on the down side as well. However, we can communicate with the future from the past, playing a game of documentation against the winds of generational forgetting. Our models of patterns in the world improve (without tipping into omniscience). This means we can more act in a way that our models suggest provides a beneficial outcome; even if we don’t always receive this individually, we receive this statistically and the world improves. We just have to choose what outcome we think is beneficial.
This view also leaves just the present with possibilities. It rules out complete control. Bad things may happen. Good things may happen. The most knowledge we have is now (yet even this is incomplete).