Moulding uncertainty

How to make better decisions under uncertainty (Part 1)

By Pedro G. Del Carpio.

decision making optimizacion, behavioral sciences tool — “Proelium” de Alex Hall https://www.alexhallart.com/

Life is about making decisions and living with their consequences.

But could it be possible that, until now, we haven’t really learned how to determine the actions that are most suitable for us?

This essay -divided into several chapters- presents a series of theoretical principles of decision science, with the overriding aim of providing a set of concrete tools that could be useful for the reader. Although a few technical terms are unavoidable, I’ve tried to express these ideas in the clearest and most universal way possible.

If I’ve achieved my goal, this article will establish the conceptual foundations that can improve our decision-making process. Concrete steps are presented, with which we can transform situations that are loaded with uncertainty into more manageable situations that allow us to clearly determine the best course of action.

Part 1 Moulding Uncertainty

The Basics: About Risk and Uncertainty

Aside from the allure of gambling scenarios or the aesthetic pleasure of the ambiguous, the human necessity for finding certainties in a world fundamentally made up of unknowns has led us to develop an aversion to situations of risk, and even more to the ones of uncertainty. [1]

The difference between both concepts is that in situations of risk, even though the future of an event is unknown, the probabilities and the possible outcomes are established beforehand. Let’s say you are in the terrible situation of playing a dice game in which your life is at stake (à la Russian Roulette). In this case, your fate will be determined by roll of the die, which will give you a 16.6% chance (1 in 6) of rolling a number that will end your life. In another example, the odds of dying in a plane crash [2] are 0.0000018 % (1 in 5,371,369). Although the latter scenario is certainly more complex, the basics behind the computation of risk in both cases are essentially the same. If the potential outcomes and probabilities are known, we face a situation of risk.

Conversely, uncertain situations are vastly more difficult to manage. Here, the possible outcomes are known but the probabilities cannot be determined. And under extreme cases, outcomes and their probabilities are very hard or impossible to determine.

Thus, in a general way, any situation that requires us to make a decision can be located at any point on the spectrum that we will call Clarity-Opacity (Figure 1) which expresses our level of knowledge about the information regarding the possible outcomes and probabilities in that given situation [3] (and therefore also its level of manageability). Under Risk, outcomes and probabilities are known; in a Black Swan situation we are facing an outcome with a very small probability of occurrence — even considered insignificant — but with disproportionate consequences if it happens; under Knightian Uncertainty, outcomes are known but not their probabilities; and finally under Radical Uncertainty, outcomes and probabilities are unknown.

Figure 1 : Clarity-Opacity Spectrum. Adapted from Decision making in uncertain times: what can cognitive and decision sciences say about or learn from economic crises? by Meder, Le Leec, Osma

Knowing the distinction between the concepts of risk and uncertainty is critical, since it shows us that the kind of situations we have to deal with in our modern lives are, overwhelmingly, of the uncertainty type: Should I spend now or save money? Choose a career I’m passionate about, or pick an option that is less interesting but better paid? Should I start working after college, or take some time to travel? Should I quit my job to embark on an attractive business venture? Should I continue or end the relationship with my romantic partner? Reader, feel free to expand this list to the moon and back with as many relevant personal questions as you want.

That being said, it’s undeniable that situations of risk are significantly more attractive, because knowing the possible outcomes and their probabilities lets us create a scenario where each path has a measurable, and to some extent, controllable consequence. Hence, if we want to make better decisions we must do everything possible to “mould” — even in an artificial way — the situations of uncertainty into situations of risk [4].

Introducing the Subjective Expected Utilityy

A radical proposition: We should choose the best option for our interests.

No, it’s not a joke. Although in many instances humans are sufficiently good at making decisions — particularly under familiar or simple situations — our limitations in choosing the right course of action under complex scenarios have been proven throughout history to be more than just simple anecdotes. However, leaving aside cognitive and information limitations or the influence of social interactions, our not very reliable ability to make right decisions is rooted in the lack of a set of logical procedures that enable the organisation of our knowledge of the world — both objective and subjective — when deciding on a course of action. Remember that a good decision is determined by the process followed, and not by the final result.

The Subjective Expected Utility (SEU) is one of the paradigmatic models of decision-making, which addresses head-on our limitations in correctly choosing a path to follow, by suggesting a prescriptive method [i] — i.e. it tells us how to make decisions — , based on a set of assumptions or axioms of what a “rational behavior” should be [ii]. Unlike the models that preceded it, the SEU can deal with our ignorance about the objective probabilities of any uncertain scenario through a subjective estimation of these.

The SEU establishes that we must choose a specific course of action among a series of options when the interaction between its attractiveness and its probability of occurrence is higher than the other alternatives.

The SEU can be calculated in a surprisingly easy way by identifying the following elements:

The course of actions we can take.
The outcomes that could happen in the future.
The probabilities of occurrence of these outcomes.
The consequences of each possible course of action and their level of (dis)satisfaction or (un)desirability.

Application: Making Better Decisions

Let’s apply the Subjective Expected Utility method for decision-making using one of the earlier examples: Should I quit my job to embark on an attractive business venture?

1.First, establish all the possible courses of action [iii].

Quit your job.
Stay at your job.

2. Then, establish the outcomes that could happen in the future and estimate[iv] the probability of occurrence of each outcome.

The business venture fails = 35%
The business venture succeeds = 65%

3. Next, estimate a value for each of the consequences arising as a result of the interaction between each course of action and the possible outcomes. From -100 which is absolutely undesirable or unsatisfactory, to +100 which is absolutely desirable or satisfactory[v]. This value is called utility.

You quit your job and the business venture fails = -60
You quit your job and the business venture succeeds = +80
You stay at your job and the business venture fails = +30
You stay at your job and the business venture succeeds = -30

4. Finally, apply the following formula to each course of action. Choose the course of action with the highest SEU value.

SEU of each course of action = (probability of outcome 1 utility of consequence X)+(probability of outcome 2 utility of consequence Y)

SEU of quitting your job = (0.35*-60)+(0.65*+80)= +31
SEU of staying at your job = (0.35*30)+(0.65*-30)= -9

From the matrix we can infer that quitting your job is the decision that, based on our subjective estimation of the inputs that form this scenario, is the course of action that will give you more utility or satisfaction. Clearly, if different values are estimated, the SEU will also be different and therefore the right course of action could be the opposite one.

Taming the Beast (And The Limitations Of SEU)

The huge merit of a method such as the Subjective Expected Utility is that it allows us to identify what is best for our interests, a very elusive challenge when the situation under analysis is new, complex or when the stakes are high. Since the SEU expresses utility in numerical terms, we can objectively compare alternatives and thus explicitly determine our preferences and the actions we should take; something exceptionally useful when our hunches, emotions, or guesses are not trustworthy enough.

Through the subjective estimation of potential outcomes and their probabilities, this tool is capable of taking a situation from the world of uncertainty and bringing it closer to the one of risk. Uncertainty isn’t measurable nor actionable, risk is. You can’t kill the beast, but you can tame it. How much better would some aspects of our lives be if the most important decisions had been made through a logical process like the SEU?

Uncertainty isn’t measurable nor actionable, risk is. You can’t kill the beast, but you can tame it.

However, the decisions made with the SEU stand on a very ambitious assumption: all the elements that shape them — actions, outcomes, probabilities, and consequences — have been correctly identified and estimated. Thus, decision-making’s “Holy Grail” seems to be in obtaining the highest quality and quantity of information, with the least noise and bias possible, and then using it in a method with logical rules that identifies the path of greatest utility [vi].

Then, one might ask, do we know enough about the world to assign probabilities to its events with certainty? Can we distinguish between what could give us more or less satisfaction? Can we trust that our tastes and preferences won’t significantly change in the future? Knowing that time will inevitably bring unpredictable events to our lives, how fragile are the decisions we make today? How can we be sure that our decisions won’t lead to negative second and third order effects [5]? In general terms, how confident can we be that the information we have is adequate and sufficient? Can our cognitive boundaries be ever able to handle such amount of information load ? As you might have guessed, even if we follow the right process, there are no guarantees about the ending. If supplies are defective, so is the product. Shit in, shit out.

As we will see throughout the following series of articles, the aim of obtaining perfect information is an unfeasible utopian target.

In the next chapters of How to Make Better Decisions Under Uncertainty, we will use concepts from Psychology, Risk Management, Behavioral Economics, Probability Theory and Philosophy to explore the different ways in which we can deal with our cognitive limitations, with the difficulty of accessing information, and with ontological uncertainty.

Finally, I invite the reader to directly experiment with the Subjective Expected Utility method, using it as an aid to determine which are the courses of action, big or small, that will give you the greatest satisfaction. Even though the SEU method is far from infallible, to know it and be able to apply its principles is a first and critical step in the journey towards the improvement of our decision-making process.

Main Takeaway

Making decisions in a situation of uncertainty requires us to transform or mould it into a situation of risk. To do so, both the outcomes and their probabilities of occurrence must be determined. The Subjective Expected Utility allows us to make logical decisions based on a subjective estimation of its component elements. With the SEU we can objectively compare alternatives and thus explicitly determine our preferences and the actions we should take; something exceptionally useful when our hunches, emotions, or guesses are not trustworthy enough.

Notas

[i] Its validity as a descriptive model is a matter of academic debate.

[ii] What rationality really means for decision-making will be developed in the next chapters.

[iii] It may be possible to establish more courses of action, such as staying at your job and working on the business project during your spare time.

[iv] Note that these probabilities are estimated from the information we have, either objective, subjective or the combination of both.

[v] A value of zero is equivalent to being indifferent to the consequence.

[vi] As we will see throughout the following series of articles, the aim of obtaining perfect information is an unfeasible target.