What do psychologists do?

[This post is a summary of ‘What is Psychological Explanation?’ by William Bechtel and Cory Wright] 

Science is in the business of offering explanations. Why do certain things happen the way that they do? For example, Galileo showed that you can explain the period of a pendulum (T) in terms of the length of the pendulum (l) and the gravitational force (g).

Like physics, psychology is an explanatory science. But neat equations like this are actually quite rare in psychology. So what kinds of explanations does psychology give? What are psychologists up to with their theories? 

Consider some examples. One thing psychology does is theorize about what mental processes underlie certain behaviors. Suppose I ask you to study a list of numbers, e.g. [5,6,2,9], take it away, and then ask you whether 2 is in it. Here are three hypotheses for how you search through your short-term memory of the list:

1. you examine the whole list at once, in parallel 
2. you examine the list one at a time, and stop as soon as you have found the two (i.e. you examine only three items)
3. you examine the list one at a time, and keep going after you have found the two (i.e. you examine all four items)

As you can probably work out for yourself, each of these hypotheses predicts a different relationships between the features of the list and how long it takes you to respond. Hypothesis a) predicts that list length and search item position will not matter; hypothesis b) predicts that length until the search item will matter; hypothesis c) predicts that list length, but not the position of the search item, will matter. Do you have a guess which hypothesis is true? [1]

Psychologists also look for physiological findings. For example, you can link certain reward behaviors to brain regions, such as (so Bechtel and Wright tell us) “several highly-interconnected anatomical structures constituted, in part, by the long and fast myelinated dopamine-selective axons projecting afferently from the ventral tegmental area to the nucleus accumbens”. 

Lastly, sometimes psychology does actually give you nice law-like things that you can put on a graph. If stone A is 10 pounds, how much heavier does stone B need for you to be able to reliably say which one is heavier? 10.05 pounds clearly won’t do it. Call the smallest amount that will allow you to do this is called the ‘just noticeable difference’. Let’s say that at 10 pounds, the just noticeable difference is 2 pounds – a 12 stone needs to be able be able to tell the difference. From this, you can predict what the just noticeable difference at 50 pounds is going to be. (Do you have a guess?) It’s 10 pounds. The JND for 100 pounds is 20 pounds. And so forth. Weber’s Law says that the just noticeable difference is a constant ratio of the weight (In this example, the constant is 0.2). Weber’s Law holds for a wide variety of discrimination tasks, not just weight. This is a non-obvious and regular law, one that links physical quantities to psychological phenomena, and you can think of this law as ‘explaining’ why the JND at a given weight is what it is. 

From laws to information processing mechanisms

Weber’s Law notwithstanding, Bechtel and Wright argue that genuine laws are rare in psychology. For one thing, there are few things like Weber’s Law in psychology. For another, Weber’s law itself stands in need of further explanation. Why does Weber’s law hold? What goes on in the brain that explains why the JND is a constant ratio of weight, instead of (for example) some fixed amount? According to Bechtel and Wright, the reason for Weber’s Law is still unknown! In any case, because observed relationships in psychology typically call for further explanation, they are generally more often and more properly called effects instead of laws [2]. Or, you could think of them as what philosophers dub empirical laws – regular observed relationship between quantities (like weight and discriminability). The pendulum example above is an empirical law. Empirical laws can themselves be explained by theoretical laws – in the pendulum case, the relevant theoretical laws would be Newton’s laws. But in the case of psychology, it is not clear what the theoretical laws behind the empirical laws would even be. Biological laws, perhaps? The problem is that biology itself has very few laws. Ultimately at the bottom things, the theoretical laws are probably the laws of physics, but clearly psychology’s explanations are very far from bottoming out in physics. Psychologists do not, and indeed need not, make reference to quantum mechanics when they construct their theories.

Bechtel and Wright claim that when psychologists offer theories that go beyond the empirical laws that they have established, they do so by proposing information-processing mechanisms. A mechanism is “a composite system organized in such a way that the coordinated operations of the component parts constitute the mechanistic activity identified with the explanandum”. (6) These component parts can themselves be mechanisms that can be broken down into further parts, in a hierarchy of mechanisms.

Since the invention of the digital computer, cognitive scientists have proposed that what the mechanisms of the mind do is process representations of the world, for example taking the input of light on the retina and transforming this information into a coherent perception of the environmental scene before you. The computer model has in general inspired psychologists to think of the mind as doing things like storing, retrieving, and operating on representations. This is in marked contrast with the behaviorists of the mid-20th century, who thought that psychology should eschew talk of internal representations.

One major task of psychology is to identify these mental operations with certain brain regions. Before good brain imaging techniques, the main way of doing this was to wait for people with localized brain damage and see what that damage affected. A surgery on the famous Patient H.M., for example, left him able to learn new skills like playing guitar, but not able to remember that he had learned these new skills, providing powerful evidence that there is a distinction between knowledge-how and knowledge-that, and that different brain regions underlie them. Research on non-human animals allowed people to use more invasive techniques to map things; and in the last few decades the increased power of fMRI and other imaging techniques has led to a better understanding of what regions are responsible for what. But it’s not knowing where something is that is explanatory; rather, knowledge of the geography of the brain can help us discover mental mechanisms.

Thus we should not think of psychology as like physics, where specific laws like the pendulum are eventually subsumed into more general laws like Newton’s mechanics, which is in turn subsumed into general relativity. More often psychological explanations come from proposing information-processing mechanisms to explain patterns of behavior.

[1] Hypothesis C is true.

[2] In this parlance, scaling laws in machine learning should instead be called scaling effects, since why these relationships between data size and test loss hold, is not really understood.