Key Concepts
“We speak of understanding a sentence in the sense in which it can be replaced by another which says the same; but also in the sense in which it cannot be replaced by any other.” ― Ludwig Wittgenstein, Philosophical Investigations
“Words are the source of misunderstandings.” ― Antoine de Saint-Exupéry, The Little Prince
“If you wish to converse with me, define your terms.” — Voltaire
To have a conversation you needs to agree on what the words mean. The goal of this section is to provide a frame of references for the key terms used in the notebook. As many terms have slightly diverging meanings depending on who is using them, a clear definition is essential to avoid misunderstanding.
While the use of certain terms in this notebook might deviate from the most common practices in the literature, I do not aim to suggest that it is “correct” use. It is only a way to increase the clarity of thought in this particular context.
Decision and Decision-making
Decision-making literature typically takes the term “decision” as self-explanatory and typically says nothing about the key term of analysis. Most of the dictionaries define decision as a choice made between alternative courses of action often in a situation of uncertainty.
For this notebook, I will define decision as a cognitive task of choosing an alternative course of action to achieve certain goals based on the perception of an environment and available information. This hefty definition incorporates key elements of the decision:
- Environment or context in which decision is being made. Environment constraints available courses of action and determines the outcomes of the decision.
- Each decision includes at least two options or alternative courses of action, inaction being an acceptable alternative. It is important to note that decision-makers are not necessarily aware of all the options allowed by the environment.
- Information or factual statements about the environment that are external to the decision-maker. The correctness of the statements is not assumed.
- Perceptions, mental representations or models (the terms will be used interchangeably depending on the context) are symbolic structures that represent external reality. In a broad sense, they can be treated as interpretations of external reality used to predict outcomes of the decision alternatives.
- Preferences or goals are sets of attitudes and believes used to evaluate the attractiveness of the decision alternatives. Unless stated otherwise, I will not assume any properties of preferences, such as transitivity or consistency over time. Furthermore, I will not even assume that the decision-maker is fully aware of the preferences guiding its decision.
Uncertainty and Risk
“Risk means more things can happen than will happen.” ― Elroy Dimson
In his 1921 paper “Risk, Uncertainty and Profit” Frank Knight identified two distinct types of uncertainty- the measurable uncertainty which he called “risk” and unmeasurable one for which he used the term “uncertainty”. Despite being of limited use, this distinction became the most widely used way to talk about uncertainty or risk.
To better recognize certain nuances of the term, in this notebook I’ll use more detailed classification (mostly borrowing from Douglas Hubbard):
- Uncertainty is an absence of certainty or the existence of more than one possibility. Defined like this uncertainty can be measured using probabilities, however, the ability of an observer to measure is not guaranteed.
Two special cases of uncertainty can be identified that refer to different states of knowledge:
- Strict uncertainty, where all possibilities are identified but probabilities are not known. This is unmeasured or not measurable uncertainty. While Hubbard argues that “this should never have to be the case”, some evidence suggests that some decision making is done in the environments resembling strictly uncertain.
- Deep uncertainty or what Hubbard calls ignorance is a state where possibilities are unknown. Although most of the real-world environments have a certain level of deep uncertainty, as at least some of the possibilities are not known.
Finally, after defining uncertainty we can move to risk.
- Risk is a state of uncertainty where some possibilities could lead to any undesirable outcome such as financial loss, physical or psychological discomfort, etc. Possibility of harm implies that “risk” unlike “uncertainty” is always relevant.
Decision risk
“The risk of a wrong decision is preferable to the terror of indecision.” – Maimonides
People make millions of decisions every day. What socks to wear, where to eat dinner, how to deal with a misbehaving child, whether to suppress or embrace the urge to punch an annoying colleague, what part of the family fortune to invest in hot cryptocurrency. Most of our decisions are unconscious and automatic. The few that reach our conscious thought are usually resolved instinctively and without much deliberations. And then there are those decisions we invest hours of meticulous thought and build complex infrastructure just to make sure we make them right.
While most of the time we are happy with the choices we make, occasional we are not. Our discontent hides two distinct (though sometimes overlapping) groups of the “bad” decision - ones that lead to undesirable outcomes and ones that are made using poor processes. It is the second group that is the object of this notebook.
While we often care about decision-making processes only after we experience undesirable outcomes, favourable outcomes do not always indicate a sound decision-making process. What is a sound decision process?
- Sound Decision-making process is a process that works well over a wide range of plausible futures when compared to the alternatives.
This allows us to define decision risk:
- Decision risk is the risk of loss resulting from using insufficiently sound decision-making practices and tools.
The concept of decision risk includes a more widespread concept of model risk, as models are one of the tools used in decision-making.
- Model error is the risk of loss resulting from wrong, poorly implemented or improperly used models.