The Use of Psychology to a Reductionist or an Identity Theorist: An Analogy

Foreword

This paper was my first paper for Philosophy 951: Larry's Favorite Papers, a graduate seminar that I was wildly unqualified for. It is nice going into this paper understanding the basics of the philosophy of mind, as well as reductionism, the mechanistic explanation and the arguments of the mechanistic hegemonist, and the mind-body supervenience problem. Many anti-reductionist positions are typically motivated by the fear that the aforementioned positions threaten psychology as a distinct science. I argue that that conclusion can not strictly be drawn from the positions, and that such motivation is therefore unjustified. I use an analogy to LLMs to draw an analogy to help shift our intuitions.

Body

Many discussions in the philosophy of mind by anti-reductionists tend to be motivated by keeping psychology as a distinct science, and attempting to prevent various physicalist positions from declassifying it as such. Reductionism threatens the validity of psychology because it claims that all psychology is fundamentally just physics. The mechanistic hegemonist threatens the validity of psychology because they claim that psychology is fundamentally just neurobiology. Mind-body supervenience yields a problematic conclusion because if mental states aren’t truly causal, psychology is not a strict science. I wish to argue that even accepting these physicalist arguments, psychology is at no risk of invalidation. Each of these positions warrants valid counterarguments, but arguing against these positions motivated solely by these grounds is unproductive. Psychology doesn’t need to have any of the aforementioned characteristics to still be useful. It may be the case that reductionism is true, or the mechanistic hegemonist is correct, or that Kim is correct and that mental states aren’t truly causal. All of these positions still must grant that psychology is at least useful, and warrants investment. Therefore, any motivation for argumentation based solely on this assumption is unwarranted.

To begin our discussion, consider laws. Compared to other strict fields like physics, in which there are many lawful claims, special sciences often don’t have many law-like claims. We note this to justify the intuition that laws should be strict. In either case, how the word law is used in fundamental sciences differs greatly from how it is used in the special sciences. Initially, note that law claims are justified using inductive reasoning. We say that gravity exists because we have observed it inductively. This means that the laws of physics could feasibly be wrong. However, that is an epistemic question. The claim that these fundamental sciences are making is an infallible one, it is either true or false that the conservation of mass is true or not. If we find a single counterexample to the law of conservation of mass, the law now fails. This is not true with laws of special science, for example, consider Gresham’s Law. The general idea is that given two physical currencies of the same face value, but with one having a greater perceived value, people will tend to prioritize the one with greater perceived value. Say we have two pennies, each worth one cent, but one is made of copper and the other of gold. Gresham’s law claims that people will prioritize gold, and that the copper pennies would phase out of circulation. Though this is an intuitive generalization, it is not strict. I can envision many scenarios in which this is not the case. The perceived value of the currencies might be so small in a society that people don’t care enough to prioritize one. Or I may gather a group of friends to make a small society, in which we all agree to prioritize currency of lower perceived value. The point is that a single counterexample of an economic law does not prove it wrong. I wish to call infallible laws “strict laws”, and non-infallible laws “general laws”. Strict laws show up in the fundamental sciences more often than the special sciences, but this is not to say that either may not show up in the other.

It could be that there are possibly strict laws of psychology, but clearly, we don’t have any now, and it is hard to think of things that could fall under this. It seems like, rather, most laws in the psychological sphere are general laws; generalizations rather than infallible rules. Does this invalidate psychology? Not at all. It hardly seems to threaten it. Just because we can’t get any strict rules from it, doesn’t mean that we can’t use the generalizations. Even if someone wants to strictly define science as “only fields that can dictate strict laws”, it is irrefutable that psychology provides us with useful generalizations.

I argue that this intuition should be the exact same idea that we have when considering Kim’s mind-body supervenience. Though it can validly be contested that mental states are not causal, the conclusion hardly threatens psychology. So what if mental states aren’t causal? One could argue that for a field to be a science, it needs to have causal power. However, the denomination of a field as a science does not automatically dictate its usefulness. If that were the case, philosophy would not be useful.

Consider the field of linguistics. It is hardly easy to say that anything about it is clearly causal, at least not in the way one would argue psychology is. It is the study of language and its structure. Completely ignoring the fact that many people still consider linguistics a science, it still provides us with useful information about how our language works. This can further provide us with information about how language tends to affect the way we think. Whether strict laws exist or not in this field, or if the generalizations it makes are causal or not, does not change that fact.

Finally, assume that the mechanistic hegemonist is correct, and all functional explanations are simply incomplete mechanistic ones. Think again of the discovery of genes. They were initially understood purely functionally (i.e., how parents' genetic makeup functionally affected the genotype of their offspring), yet were later discovered to have physical mechanisms that explain heredity. Though Shapiro would say that the hegemonist would need to say that the functional explanation is “wrong”, this is not the case. It being incomplete doesn’t mean that it is wrong. The opposite was true; the partial understanding of how genes functioned was irrefutably useful. The mechanistic hegemonist would only need to claim that at that time, there was a deeper way to understand how genes worked. One may be worried that this conclusion necessarily reduces all psychology to neurobiology. I wish to promote an analogy about how, even if this is the case, we still have great reason to keep the fields distinct.

Consider Large Language Models (LLMs). For reference, I have a machine learning background, so I have a good understanding of how these work. Strictly speaking, LLMs are computationally reducible to mathematical functions. Your input of text is tokenized into vectors, which are lists of numbers that store the semantic information of the tokenized word. All of your words are put into a matrix, a list of the vectors. This matrix then goes through the transformer, which is just a lot of matrix multiplication combined with other mathematical functions. These operations then result in an output vector, which itself is a token. The model then untokenizes this word, appends it to your original matrix, and repeats the process until one of the predicted tokens is an end-of-sequence token, which indicates to the model to stop the process. All of those predicted words are now the response it gives. So, strictly speaking, all of the functions of the LLM are reducible to mathematics. However, Anthropic is doing research into mechanistic interpretability. Their work focuses on trying to locate those overarching patterns between input, output, and the internal states of the transformer. For example, what does one value in the matrix being more active indicate about what the output will be? This research has already had great success, finding general trends that help understand how the model is “thinking”.

There are two cases: either LLMs have something similar to mental states, or they do not. The hegemonist would likely assume the (more likely) latter. If there are no mental states to discover, is their research in vain? Should they really just be doing research into how the connecting nodes work together? Clearly not. We are at a point where exclusively understanding the low-level details does not give us a good understanding of what the model is doing at higher levels. Sure, these higher levels may not be causal, but the overarching patterns we get from them are indispensable. We need to understand what is happening at a more functional level to make those generalizations in the future.

Assuming any of these physicalist positions, and ignoring the complexities of quantum theory, one can envision that we may understand physics so much that we can make correct causal predictions about one's actions given the physical states of one's brain: something like Laplace's Demon, except a real device that can make these predictions. But, until then, research into the special sciences can only help us reach this point; the reductionists cannot invalidate the special sciences.

Rounding back to our initial discussion, why would non-reductionists have any motivation to use the invalidation of psychology as motivation to argue against these positions? Through our discussion, it seems like the physicalist cannot invalidate the special sciences. I think the fear is that accepting these conclusions may cause a harsh outlook on the special sciences like psychology. If psychology is theoretically nothing more than neuroscience, it is worth less. This may be enough to motivate counterarguments; however, the intuitions behind these physicalist positions are stronger than those behind keeping psychology distinct. Thus, I believe that intuition does not take precedence. However, let us grant the opposite of my intuition: that keeping distinct psychology is more important. I argue that the prior conclusion is not true, that people won’t think it is any worse of an explanation. I think people generally consider broad special sciences like economics or sociology to be quite respected. Both of these fields seem to be at least largely reducible to psychology. I think this would be the same with psychology. Sure, it may be reducible, but people will still respect it because of the practical conclusions it yields.

Thus, it seems like the anti-reductionists are misguided by motivating their arguments by the worry that granting the opposing view is detrimental to the field of psychology. Again, there are other sound reasons to argue against these positions, and all skepticism of positions is how work in philosophy gets done. However, the idea that psychology is at any risk is misguided.