Simple & Intuitive Models of Decision Explanations

An explanation of a decision will provide reasons for why the chosen option was chosen over others. In this text, simple and intuitive models are presented for how to organize information into an explanation of a decision. The models are interesting only as a starting point, before we go into more elaborate ones and in particular those grounded in research.

This text is part of the series on the design of decision governance. Decision Governance refers to values, principles, practices designed to improve the quality of decisions. Find all texts on decision governance here, including “What is Decision Governance?” here.

What is the topic of this text?

What does an explanation of a decision look like? We’ll see in several texts that there are various ways to explain decisions, and various types of explanations for decisions: some emphasize motives of the decision maker, others the role of the environment, incentives, uncertainty, knowledge, and so on. None wins. Knowing various ways to explain decisions helps keep one’s distance from believing too much one explanation over others, and keeping a critical attitude towards all of them; the reason such an attitude is helpful, is that explanations of decisions of people necessarily involve speculation about what these people thought, felt, believed, could and couldn’t do, and so on, in a decision situation, and it is impossible to have unequivocal evidence for these speculations. (See the text, here, on what may qualify as evidence.)

Why is this topic relevant for decision governance?

Before designing guidelines, rules, and processes to govern decision making in a decision situation, it is necessary to have a relevant explanation of how a decision is made – without any governance changes – in that situation. Without a good explanation of a decision, you do not understand the factors which led to it, and consequently, you do not understand which factors decision governance needs to influence.

Background: What are the characteristics of good explanations of decisions?

According to dictionary definitions of “explanation”, to explain something that happened, “you give people reasons for it, especially in an attempt to justify it” (“Explanation”, Collins Dictionary); to explain is to “make plain, manifest, or intelligible; to clear of obscurity; to illustrate the meaning of” (“Explain”, Wiktionary). In another text, here, good explanations of decisions shared the following characteristics. 

1. Causal Clarity
2. Simplicity
3. Coherence
4. Specificity
5. Pragmatic Value
6. Consistency with Evidence
7. Counterfactual Plausibility

To talk about causality, and have clarity of cause and effect, requires a description of the sequence of events leading to the decision. This does not mean that events cause one another, but that even before we can claim that some events caused others, we need to order events over time. Simplicity can be achieved by removing irrelevant events. Coherence is about ensuring that the events fit knowledge and assumptions about the environment and time in which they occurred. Specificity removes irrelevant information about events and their relationships. Pragmatic value is about the usefulness of knowing how the events unfolded and what their relationships are. Consistency with evidence is achieved if the data about events and their relationships matches what we claim about the sequence and other relationships between events. Finally, counterfactual plausibility concerns what we know or can reasonably claim about alternative relationships between events, and, or alternative events.

The point of the above is that it is difficult to explain a decision without describing the sequence of events leading up to the decision.

Example: Explaining a decision in terms of steps leading to it

To make this more tangible, let’s take Achilles’ decision, described below, as an example.

In The Iliad by Homer, Achilles, the Greek hero, decides to withdraw from battle after a dispute with Agamemnon, the leader of the Greek forces during the Trojan War. Agamemnon, having been forced to return his war prize, Chryseis, to appease the gods, demands Achilles’ prize, Briseis, as compensation. In response, Achilles, feeling dishonored and enraged by Agamemnon’s actions, chooses to withdraw himself and his troops from the fighting, despite the consequences for the Greek army.

If we only want to describe this in terms of events, we have the following.

1. Agamemnon’s Loss: Agamemnon is forced to return his war prize, Briseis, to appease the gods. This action diminishes his honor, as war prizes were a key symbol of status and success for Greek warriors.
2. Agamemnon’s Demand: To restore his honor, Agamemnon demands compensation by taking Achilles’ war prize. This action asserts Agamemnon’s authority but directly challenges Achilles’ status and honor.
3. Achilles’ Reaction: Feeling deeply dishonored and insulted by Agamemnon’s demand, Achilles becomes enraged. Honor is central to Achilles’ identity as a warrior, and Agamemnon’s actions undermine his sense of self-worth.
4. Achilles’ Decision: In response, Achilles chooses to withdraw from the battle, along with his troops, as a way to protect his honor and protest Agamemnon’s actions. This decision, while personally satisfying to Achilles, weakens the Greek army and affects their chances in the war.

The events of interest are the result of actions that people take. I’ll call them agents. (This allows for more flexibility later on, when we will be looking into actions taken by systems, where AI may be behind the decisions of automated agents.) 

There is a fine line above between a description of what happened and an explanation of the same, in other words, the explanation isn’t very good. At best it consists of reasons why agents behaved as they did, and these reasons are implicit in the description of each event, for example, that Achilles decides to withdraw to protect his honor. 

(As a side note, observe that the example is not, strictly speaking, about a single decision. It is a fairly arbitrary choice to focus on Achilles’ decision. What is interesting is that we can never really talk about a single decision in isolation: Achilles’ decided something because of what other people decided, and we cannot explain his decision while ignoring theirs.)

We can make these goals more explicit, and split decisions, actions, and goals pursued through decisions. If we assume actions to be taken because of, and in order to achieve goals, we get the following revised explanation of the same decision.

We can further continue down the path of explaining decisions through motives, and to do so, we need to talk about emotions. Agents have emotions and goals that influence the actions they choose to take. Furthermore, agents use information they have at the time, to identify alternative actions, predict outcomes of these, and choose one of the actions based on how the outcomes fit goals and emotions. 

What is the explanation of a decision in this decision model?

This description of how decisions take place, and how they relate to goals, emotions, actions, outcomes, is fairly straightforward. While we will slowly break this model apart, and show its many omissions and deficiencies, we need to start somewhere – this will be that starting point to build more sophisticated models of decision explanations. 

We bundle events into steps, and consider that a decision explanation involves describing steps. The figure below shows the structure of the probable explanation of a decision when we assume that the decision is explained in terms of a sequence of steps leading to it. For the background on this diagram, see the text here.

In summary, the explanation of a decision will in this text be structured as follows.

If we apply the model to Achilles’ decision, we have the following.

In other texts, this model will be used as a simple, intuitive baseline against which more elaborate models will be discussed. The aim will remain to consider alternative ways to explain decisions, in order to inform the design of governance that should improve decision making.

To close the loop, you can revisit the characteristics of good decision explanations from the beginning of this text, and think about how it would rank – poorly, in short, as it is descriptive and speculative, relying on descriptions of hypothetical intentional states of agents. Needless to say, the example is about what dead people would do in a social environment with values and ideals that have long been discarded.