Some people are stuck between two worlds. One world of physics, where everything is predetermined and laws are rigid. The other world is a personal experience, where thoughts lead to actions and actions to observable consequences. I am one of those people, and maybe so are you. It would be very satisfying to have a theory that combines these worlds. The book that I have read attempts just this. It’s titled “How can physics underlie the mind?: Top-down causation in the human context” by George Ellis. Spoiler alert: there is a theory, but it is not satisfying.

The main thesis of the book, or, if you wish, the theory, is that complex phenomena such as the human mind can only emerge from the combination of top-down and bottom-up causation. Here, top-down and bottom-up are determined in terms of levels of hierarchy. Presumably, elementary particles form everything, but there are intermediate levels. Particles form atoms, which then form molecules, which then form cells, which then form organisms, which then form groups, which then form society. One can think about even more intermediate levels, but generally, the hierarchy progresses from elementary forms of matter to more complex ones. In this hierarchy, bottom-up causation would be from elementary particles to the society, while top-down causation would be from the society to elementary particles.

Here, the author makes two important remarks. First, he claims that there is no bottommost and topmost level. At least we don’t know everything yet, so claiming some level as bottommost is inaccurate. Second, the top-down causation may start anywhere in the hierarchy (the consequence of the fact that there are no bottommost or topmost levels). Under these assumptions, it seems it would suffice to show that the mind can change neurons, for instance. However, if I conceive of such an experiment, it would be useful to know which neurons and where.

For this question, the author gives another explanation. Such an experiment is problematic or even impossible because of the multiple realizability. It means that the same function or brain state may be activated with different combinations of lower-level entities, such as neurons. Indeed, some neurons are highly specialized for just one function, while others are more flexible. This serves as the foundation for learning, a process in which humans excel. Therefore, I would have to postpone my experiment. However, the author provides theoretical considerations about five ways how higher levels of hierarchy could exert changes in lower-level entities.

1. Deterministic top-down causation. This is about setting a context. When neurons are wired in particular networks, they are more likely to transmit information to their neighbours. This is an example of an evolutionarily defined context.

2. Non-adaptive feedback control. This assumes that information has causal powers, and the higher-level system that can use information about the difference between two higher-level states can change lower-level states.

3. Adaptive selection of outcomes. This is about selection criteria, which are not physical. As properties of lower-level entities vary, this can be used by higher-level systems to prefer entities with certain properties. Examples are complex molecules; they cannot form spontaneously, but they are the way they are through evolutionary selection (which is top-down and not physical).

4. Adaptive selection of goals. This is similar to 3 but about goals. Depending on the goal state of the higher-level system, lower-level entities may be required to perform differently.

5. Adaptive selection of selection criteria. It's similar to 3 but more complex. An example about the mind is selection based on religious or philosophical dispositions.

Can we experimentally test these considerations for the human mind? I am not entirely sure. The book is quite large, and there is more that can be covered, but I prefer to stop here.

I see the book to be organized into three main parts: the first one is quite general and abstract, laying out concepts and providing arguments; the second one contains examples from physics supporting the arguments; the third one contains examples from cognitive science supporting the arguments (presumably). When I was reading the first and second parts, it all actually sounded more or less conceivable. However, the third part sadly didn’t contain any reliable examples. For instance, the author says that placebo is a case for top-down causation, because ‘no physiologically active ingredient is present.’ I don’t see this as a plausible argument, as expectations are stored in neuronal networks, which then affect other neuronal networks that are responsible for pain. It is just causation within one level of the hierarchy. Additionally, some of the author's beliefs form the basis of his claims, such as ‘thoughts are not physical entities’ or ‘societal norms have a causal role by themselves, not via agents that reinforce them’. These are just beliefs, not proven facts. Perhaps my expectations were high, so after reading a part about the human mind, I ended up being disappointed.

However, there are some interesting ideas as well. One notable is that information, as well as its causal power, is localized on the level of interactions between entities and not on the level of the entities themselves. This presents, in my view, a sound argument for emergence, especially in the case of the brain. Another idea is that there are multiple causes: both bottom-up and top-down causation take place. Therefore, the mind is not determined by biology alone or by society alone, but by the combination of these factors. It is reassuring but maybe not extremely convincing, since there could be causation loops. For instance, society may be caused by biology, which is then affected by society; hence, hard determinism holds.

I only covered a small portion of the ideas and arguments presented in the book. If you are curious, you should read it whole.

December, 2024