The case against preferences
For economists and behavioural scientists describing human behaviour, there might be better ways to think about how we make decisions than with ‘preferences’
What are preferences?
Rational choice theory states that we have fixed preferences. But what even is a ‘preference’? Wikipedia tells us it means the following:
In economics and other social sciences, preference is the order that a person (an agent) gives to alternatives based on their relative utility, a process which results in an optimal “choice” (whether real or theoretical). Instead of the prices of goods, personal income, or availability of goods, the character of the preferences is determined purely by a person’s tastes. However, persons are still expected to act in their best (that is, rational) interest.
The idea here seems to be that humans assign, according to their tastes, an ordering of what is most-to-least liked (i.e. which object or choice has the highest to lowest expected utility) and then goes with with the option that is top of the list. This implies that each option, or object of choice, has a value on its head that is in some sense fixed. It is as though the value is somehow an intrinsic quality of the object itself rather than a characteristic of our interaction with that object in that moment. This fits with how we experience the world. We experience apples having x value to us. The idea that the value comes about through an interaction our perception and the apple itself seems alien as an idea.
Approach and avoidance behaviours — a neuroscience perspective
Neuroscience would have us think about choices a little differently, focusing on behaviour, where a behaviour is an action preceded by a choice and a choice is our subjective experience of weighing up different options before acting. The most accurate way to describe and think about behaviours, I find, is whether it is an approach behaviour or an avoidance behaviour. “I approach the object I choose, I avoid the object I don’t choose”. It can be simple: “I go towards the food, I go away from the scary tiger”. Or it can be complicated: I could have to make a choice between many things I’m attracted to (e.g. puddings on a menu), or I may be forced choose between multiple options I’m repulsed by (the least worst option). It is possible to think about each action we make in the world as an approach ‘towards’ an option or the avoidance ‘away from’ an option, or a mixture of the two.
Neuroscience tells us that, broadly-speaking, the approach behaviours system is mediated through the Dopamine neurotransmitter system and the avoidance behaviour system is mediated through the Serotonin neurotransmitter system. In mice, for example, serotonin runs through all the brain areas associated with defensive behaviours. This is examined in the case of the optimal response to a cat being in the area, where ‘optimal’ means the highest chance of survival for the mouse. Depending on how close the cat is to the mouse, the optimal response is to explore further, freeze, run or fight. The serotonin neurotransmitter system runs through each of the brain areas associated with these four behaviours, with an increased activation of the system (turning up the whole-brain serotonin ‘dial’) correlated with the closeness of the cat. Similar examples exist for the Dopamine system for approach behaviours, which is associated in particular with sex, food and potential danger.
For a more complex version of this, here are the brain areas associated and how Serotonin (aka 5HT) is thought to moderate the activity of each area associated with defensive behaviours in terms of avoidance and approach. The idea is that a global (whole-brain) increase or decrease in Serotonin moves the individual up or down the scale in line with the defensive distance, activating different brain areas according to where along the scale the level of Serotonin is and thereby changing the output behaviour.
Approach and avoidance values and System 1
The paradigm I’d like to put forward is the idea that in human decision-making System 1 (the intuitive part of our decision machinery) conjures up two numbers — two values — an approach value (positive) and an avoidance value (negative), for each option that we have at hand. It outputs these values every time we could make a decision between multiple options, including whether to act or not to act. The idea follows that System 1 recommends we make the option that, as a sum, is the most positive value of the values in front of us — the highest score. The way it comes up with these values is of utmost importance and interest and could be a complicated function with multiple common biases and multiple individual differences thrown in. Of these, we likely have background priors about how we generally make decisions which are the backdrop to additional live information. Different factors, which are ecologically or evolutionarily sensible, then bias the information towards what System 1 thinks is our optimal choice. There may also be excitation and inhibition from System 2 in this process, according to how much awareness we have over our subconscious processing. There is a key role of learning by these two systems also. For example, after a decision is made, the outcome occurs and we evaluate what happened, our priors are then updated through the mechanism of prediction errors in the Dopamine system.
For this theory to have any validity, it must incorporate existing theories of Behavioural Science into how System 1 conjures up these two values, including:
- Prospect Theory (e.g. loss aversion)
- Habits (+ approach — if we’ve done it before, we know it is safe and has a higher degree of certainty than non-habits)
- Social norms (+ approach — we want to conform)
- Previous experience (variable)
- Biases and heuristics (variable)
- Framing (variable)
- Principles and values (variable)
- Primers (variable)
- Physiological states (e.g. hunger would be + approach for food)
- The drive to fill psychological or physiological needs
System 1 is only able to make strong recommendations for certain types of ecologically sensible processing, such as those with the following characteristics. These characteristics loom larger in our (System 1) minds:
- Affect over no emotion
- Low samples / proportions over high samples / statistics
- Averages over sums
- Gains & losses over absolutes
- Representativeness / similarity over statistics within large samples
- Anchors (conformity) over other factors
Without factors such as these, System 1 either doesn’t output any approach or avoidance values from which to make decisions, or its recommendations are weak.
Where does this idea fit in the bigger picture?
This idea fits into a wider Dual-Processing Theory that then says that System 2 accepts, modifies or rejects the approach or avoidance numbers that System 1 puts forward. Whether System 2 engages is to do with working memory capacity, concentration and energy levels.
As well as choices and behaviours, the approach and avoidance paradigm is also useful when thinking about the direction of human attention and in particular vision, as it is with our eyes that we explore the environment and gather information from which to make decisions. Our eyes and attention can be attracted towards an object according to factors such as colour, sex, danger, social information, beauty and so on.
The reason I like this idea is because it replaces the idea of fixed preferences with a mechanism that can describe how preferences can be fluid, according to the context and priors. Despite likely being an oversimplification, the idea that System 1 puts forward an intuitive avoidance value and approach value is helpful because these values are unique to the specific decision at hand, with multiple variables impacting to increase or decrease each value — it is possible to plug in multiple variables and functions into how these numbers are computed, in line with multiple other findings from Behavioural Science.
This idea also has a plausible neural correlate (the Dopamine and Serotonin systems) and it helps to add detail to the very oversimplified idea of fixed preferences. The idea of these two output values per option is also something that theoretically could be used in mathematical modelling for creating a more sophisticated descriptive model of human decision making that homo economicus.
This perspective puts forward the idea that our perception of preferences is a post-rationalisation narrative to make sense of our choices and give us a stable impression of how we make decisions. This perspective does allow for having general preferences (I tend to like chocolate more than sweets)— but that in the right circumstances our choices can go against this, which feels quite accurate to how we experience the world!
Next steps
The next steps for this idea would be to explore how existing findings in the Behavioural Science literature could be plugged into the functions for each set of approach and avoidance values, to design lab experiments to test this paradigm and explore whether approach and avoidance values are compatible with economic mathematical modelling.
