MatthaeusWeinhardt

Intuition in the Brain - Integration of Value

Background: Two systems

It is now broadly accepted that humans have two systems for making judgments and decisions: the explicit (or deliberative) system, which operates slowly and requires effortful consideration of all alternatives, and the implicit (or intuitive) system, which operates quickly and effortlessly.^[1]

Much research looking at implicit and explicit judgments has focused on documenting the existence of these two ways of thinking. One classic paradigm is the Stroop Task [1], in which participants are shown color words that are printed in an ink color that is different from the word itself -- for instance, the word "red" printed in blue ink. They are then asked either to read the word or to name the ink color. People have more difficulty and take longer to respond when asked to name the ink color than when asked to read the word, because reading words is a very automatic process for adults, while naming the ink color in this case requires conscious effort. For instance, if you read the word "red" printed in blue ink and have to quickly name the ink color, you have to suppress the automatic tendency to just read the word in order to consciously pay attention to the ink color and name just that. This task clearly demonstrates that people have two modes of thinking -- a faster and more automatic one, and a slower but possibly more accurate one.^[2]

Broadly speaking, the explicit (deliberative) system is slow and effortful, operates in conscious awareness, is thought to be more "cognitive," operates in conscious awareness, and can be flexibly applied. A good example would be making a list of pros and cons. In order to be useful, such a list requires time to carefully list arguments on both sides, evaluate their relative weight, and reach a conclusion. However, such a method is flexible and can be applied to almost any topic.

By contrast, the implicit (intuitive) system is fast and effortless, requires no conscious awareness, is thought to be more "affective" (involving emotions), and requires experience in a specific domain to be accurate in that domain. Consider the following quote by Green Bay Packers Quarterback Aaron Rodgers:

“When you're throwing the football, you're not thinking about your drop or your release point or the trajectory or where your feet are...you have to be quick and decisive. That's a play we've worked on for years. Years…I've thrown that ball to Greg, that same exact ball, 100 times in practice. Same exact route. So when I break the huddle, that's what's flashing in my mind. I've completed this throw in my mind 1,000 times before the ball even leaves my hand.”

This anecdotal example illustrates that in the domain of sports, athletes often train so that their split-second, intuitive decisions will be more accurate. This quote also hints at the fact that intuitive judgments have to take into account many different factors (e.g. when throwing the football, where your feet are, who to pass to, what their projected route is, what the trajectory of the ball should be to optimally reach the intended target, etc.), yet this happens outside of conscious awareness and (I will argue) only provides a signal of "go" or "no-go" -- "good" or "bad."

Advantages of Intuition

Recently, researchers have identified numerous cases in which implicit (intuitive) judgments are somewhat surprisingly superior to explicit (deliberative) judgments.

One great example is the Iowa Gambling task.^[3] In this study, subjects were given four decks of cards to draw from, two of which would be profitable in the long run and two of which would make the subject lose money. Interestingly, although participants began to draw exclusively from the profitable decks after drawing merely 50 cards, they could only explain why these decks were better by card 80. The participants’ intuition had figured out the game 30 cards before they were explicitly aware of what was going on.

These findings are not just limited to artificial lab settings: In a series of studies in the lab and among actual shoppers, it was found that choices among complex products (such as between different houses or different cars) produced better results when they were made without conscious thought.^[4] Specifically, the authors gave each of the products a high number of attributes (for instance, complex cars had 12 different attributes), some of which were positive and some of which were negative. In this context, the “better” choice or the more “desirable” product was the one with a higher proportion of positive attributes.

Another earlier study also showed that participants who chose their favorite poster from five choices after careful consideration were later less satisfied with their choice than participants who only considered the posters briefly before choosing.^[5] Furthermore, in a different study, participants were more accurate at judging the quality of paintings and poems when they relied on their intuition, instead of thinking deliberatively about their response.^[6] Although paintings and poems may be considered matters of personal preference, the authors of the study tried to circumvent the problem of subjectivity by incorporating paintings and poems that are widely recognized as low-quality. For example, the authors chose their high-quality paintings from the Museum of Modern Art (MOMA), and their low-quality paintings from the Museum of Bad Art. Pre-test ratings confirmed that this difference in quality is consistently agreed-upon.

However, despite these promising results on the adaptive nature of implicit (intuitive) judgments, the nature of implicit judgments has not been as well investigated.

Integration of value

My theory is that implicit judgments integrate separate sources of value and provide a signal of whether something is "good" or "bad." In order to be accurate, implicit judgments are slowly acquired through experience. Importantly, I believe that information about the source of value (i.e. knowledge about how the implicit judgment was formed) is lost. In other words, you may know that you like something, but not why you like it.

Behavioral Evidence

I have recently conducted studies that provide some behavioral data to support my theory that intuitive judgments integrate value outside of conscious awareness. Here, I will try to provide only a brief summary of the main findings.

In the paradigm I have used to research this, participants play a simple, visual gambling game. In this game, visual cues predict a monetary outcome, and some cues are better than others. Different cues vary in their delay until outcome – how many seconds you wait in between seeing the cue, and seeing the outcome. They also vary in the magnitude of the non-zero outcome – so how much they pay, when they pay out. And finally, the cues vary in the probability of the non-zero outcome – how often does this cue pay out? By having participants choose between cues, I can see whether they have learned which ones are better than others.

In a first set of studies, I found that participants can confuse probability and delay when asked about these factors. Instead, what you find is that shorter delays lead to a higher reported probability. However, this is only true when participants have to rely on implicit judgments. In this task, I accomplished this by making the task difficult enough such that working memory is sufficiently taxed (using 8 cues versus 4 cues, for instance). If the task does not challenge working memory, participants could basically just memorize what happened in each trial, which would be explicit, not implicit learning.

In the next study, I found that participants also seem to merge probability and magnitude when doing this task. For instance, a certain cue might pay $0.60 on average (i.e. the magnitude of the cue) with a probability of 1/3 (the rest of the time it pays 0). Later, I tell participants that the probabilities have been changed to 100%. Through a series of new choices and questions, I am able to pinpoint what they believe the magnitude of the cue was -- and found that they treat the $0.60 cue as if it were worth $0.32. This would suggest that participants in fact learned the subjective value of cues (they were able to choose the cue with higher expected value significantly above chance) without learning the cue's magnitude!

Based on these findings, it would seem that knowing what the source of subjective value was is a distinguishing characteristic of explicit versus implicit judgments. In other words, our intuition inherently prevents us from having insight into its inner workings. When we intuitively know we like something, it seems that we cannot truly know why.

Possible Neural Correlates

These latest behavioral data on integration of value in implicit judgments have some direct correlates in neural data, though they have not always been interpreted this way. Previous research has shown that the mesolimbic dopamine system (a dopamine pathway running from the Ventral Tegmental Area of the midbrain to the medial prefrontal cortex via areas like the Nucleus Accumbens) supports tracking and updating short-term changes in expected value. Importantly, such implicit mesolimbic processes align with TD models from the reinforcement learning literature.

While it is known that the mesolimbic dopamine system can track multiple dimensions of value, such as reward magnitude, probability, and delay until receipt, it has remained a question how exactly this brain system does this. Consistent with the theory I am proposing, some have argued that the system integrates different aspects of reward into a single dimension corresponding to something like "relative reward rate." Low magnitude, low probability and long delays decrease the relative reward rate. Low probability and long delays are especially similar in experience, since both of them mean that a longer amount of time will pass until you receive the next reward (either due to the low probability of getting a reward or due to the long delay until getting the next reward).

It's hard to know where along the mesolimbic pathway value integration occurs, but based on some neural data, it is likely the integration of value may occur in cortical regions like the medial prefrontal cortex.

In one fMRI study examining the neural correlates of expected value, it was found that medial prefrontal cortex activated proportionally to anticipated gain magnitude and anticipated gain probability (see graph below). By contrast, the nucleus accumbens activated only in proportion to gain magnitude. This suggests that mesolimbic brain regions help compute expected value in a distributed manner, and that cortical regions like the mPFC may be integrating probability and magnitude.^[7]

Future research could follow up on my behavioral studies and see whether evidence suggesting that people confuse dimensions of value like probability and magnitude (or probability and delay) have neural correlates that can clarify if and in which brain regions these dimensions of value are being initially processed, and where they are integrated.

References

1. Kahneman, D. (2003). A perspective on judgment and choice: Mapping bounded rationality. American Psychologist, 58, 697-720.
2. Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643-662.
3. Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (1997). Deciding advantageously before knowing the advantageous strategy. Science, 275, 1293–1295.
4. Dijksterhuis, A., Bos, M. W., Nordgren, L. F., & Van Baaren, R. B. (2006). On Making the Right Choice: The Deliberation-Without-Attention Effect. Science, 311, 1005-1007.
5. Wilson, T. D., Lisle, D. J., Schooler, J. W., Hodges, S. D., Klaaren, K. J., & LaFleur, S. J. (1993). Introspecting About Reasons Can Reduce Post-Choice Satisfaction. Personality & Social Psychology Bulletin, 19(3), 331-339.
6. Dijkstra, K.A., van der Pligt, J., van Kleef, G.A. & Kerstholt, J.H. (2012). Deliberation versus Intuition: Global versus Local Processing in Judgment and Choice, Journal of Experimental Social Psychology, doi: 10.1016/j.jesp.2012.05.001
7. Knutson, B., Taylor, J., Kaufman, M., Peterson, R., & Glover, G. (2005). Distributed Neural Representation of Expected Value, Journal of Neuroscience, 25(19), 4806-4812.