Brain Activation during Hypothesis Generation

The scientific process is composed of generating hypotheses and testing those hypotheses through experiment. Yet we don't know a whole lot about how about hypothesis generation happens on the level of the brain.

Recognizing that I am dealing with a loaded term -- scientists have strong opinions on the meaning of the term hypothesis -- I would like to talk about a study that looked hypothesis generation in the brain. Kwon et al. used fMRI to look at the brain activation associated with hypothesis generation (as opposed to being just told a hypothesis), with and without training.

Before we talk about the study, there is the little matter of operationalizing hypothesis generation. I have discussed before how the real rub in psychology and behavioral neuroscience is that you have to create a test that assesses ability for a particular cognitive process. This is called operationalizing a task. But how do you do this for hypothesis generation?

From the cognitive science perspective, hypothesis generation is a process of causal inference. We see phenomena in the natural world, and we form an hypothesis to explain why those phenomena are the way they are. For example, I see that normally white moths are a lot grayer than usual, and I see that they are right next to a coal-burning smoke stack. I hypothesize that the reason that they are gray is that they need to hide from predators and that they have become gray is because all the white ones were eaten. I have inferred the cause of the gray moths. (Whether this is a correct hypothesis is another matter.)
Kwon et al. operationalize hypothesis generation by testing the subjects ability to form a causal inference from a described phenomena.

Kwon et al. looked at the brain activation in 18 right-handed, female undergraduates divided into two groups. The first group -- called the hypothesis-generating group -- was given training over two months. Training included being presented with situations like the moths up above. They were then asked to generate hypotheses from these observations. The second group -- called the hypothesis-understanding group -- was presented with similar observations over two months, but they were merely told acceptable hypotheses rather than having to generate them on their own.

At the beginning and end of the two months, all the participants were imaged using fMRI while they performed a task that required hypothesis generation. Remember that fMRI uses the relative ratio of oxygenated to deoxygenated blood in a particular part of the brain to detect changes in brain activity.

The researchers asked a couple of questions. 1) What areas of the brain are activated during hypothesis generation? 2) Did those areas of the brain activate more in the group that was trained to perform the task? This second question is critical because it is hard to tell whether a part of the brain is required for a particular task using just the brain activation. Lots of parts of the brain are activated that may not be strictly necessary for good performance. Showing that training increases the activity -- and that performance correlates with changes in activity -- is one way to infer that a region is required. (It is not the only way. Another would be to show that people with lesions to a particular part of the brain lack the ability to perform a task.)

Here is the researchers results.

First, you want to look at the behavior. Did subjects with training in hypothesis-generation perform the task better? The researchers graded the subject's responses on the sophistication of the hypothesis. For example, a simple hypothesis is that a gray moth has more pigment. A more sophisticated hypothesis might include that more pigment was selected for as a trait because it conferred an fitness benefit. The researchers constructed a score for this sophistication, and as shown below (Figure 1 in the paper). Subjects with training did better than those who were just told hypotheses.

Second, you want to identify brain regions that fulfill three criteria. First, they need to be activated or deactivated during the hypothesis generation. Second, this activation or deactivation needs to increase (or decrease) in proportion to task performance. Three, they need to change their activity in response to training.

The researchers identify a couple of regions that fulfill these criterion including the superior frontal and inferior frontal gyri (mostly in the left hemisphere). These are depicted in the diagram below (click to enlarge):

A shows the hypothesis-generating group. B shows the hypothesis-understanding group. Red indicates areas that increased activation with training. Green indicates decreased activation. The red bits on the top part of A are the superior frontal and inferior frontal gyri.

What is the significance of this paper?

First, it is significant that the authors identified brain regions that may be required for hypothesis-generation. Let me say, though, that I am not floored by the fact that the superior frontal and inferior frontal are involved. These regions -- part of the brain called the dorsolateral prefrontal cortex (dlPFC) -- have been implicated in all sorts of functions such as planning and organizing complex behavior. These functions are lumped into the term executive function (a term that has problems...but I am not getting into that now). I use the word may because while I think they present convincing evidence, I will only believe these regions are absolutely required when they show me someone with a lesion to these areas that cannot form new hypotheses.

It is also significant that the authors correlated this activation with training and with performance. That is important because you don't always see that in fMRI papers. You could pick a variety of activated brain regions out of a hat, but it is unlikely that they facilitate a cognitive process unless they meet these criterion. These guys did due diligence.

Finally, I think it is interesting to know that you can train people in hypothesis-generation. From the results of their behavior experiment, you could conclude that a science teacher just telling the students the hypothesis isn't going to cut it. You really need to have the students practice explaining results themselves, or they won't get any good at it.

I do have a couple caveats. 1) Increased activation in this case correlated with increased performance, but you have to be skeptical of that as a general finding. It is not necessarily true that a highly activated region will perform better. It could be compensating for a deficit. 2) These results show more left hemisphere lateralization. That isn't surprising because the subjects were all right-handed, but that could be different in another population. Further, the authors note that men and women often have different overall patterns of activation when performing tasks. The results could also show gender differences. 3) My skepticism with all of these studies is pretty high just because the group sizes are so small. This is an understandable methodological problem; it is just too time-consuming and expensive to have more. But you still should be skeptical until the results are repeated.

All in all, a very interesting study though. Here is the authors summarizing the significance of their work:

Throughout the experiments for this study, the participants showed neural correlate in their progress of biological hypotheses generation. The results show that their brain can be changed differently in terms of their functions by constantly experiencing different kinds of classes (hypothesis-generating and hypothesis-understanding group). The changes in their brain activation related to hypothesis generation occurred for a period of several weeks, and learning-dependent cortical plasticity may have played a role in such changes. In terms of learning-dependent brain plasticity, the results of this study have such cognitive implications as the following.

This study deals with higher inquiry skills than prior ones in many of which only simple skills like typing were studied. It also shows that changes in brain activation patterns can be possibly made not only by fast training but also by slow training. Our findings will provide a more scientifically specific and supportive foundation for biology classroom instructors who are willing to improve students' hypothesis generation abilities.

However, more issues remain to be researched in future studies. We could see training-induced increases and decreases in brain activity through this study, but it is still unclear whether or not the participants' brain activation patterns tend to go back to old ones unless sustainable trainings are given. And, if they are reversible, the critical period will remain to be revealed by further researches.

Y KWON, J LEE, D SHIN, J JEONG (2008). Changes in brain activation induced by the training of hypothesis generation skills: An fMRI study Brain and Cognition DOI: 10.1016/j.bandc.2008.08.032