· ·

Simulation of Attention, Memory, Competence, Expectations, Reputation in Decision Making

Let’s say we can design decision governance that influences specific parameters of how the decision maker behaves. Specifically, let these parameters be called the decision maker’s attention, memory, competence, expectations, and reputation. How do alternative decision governance designs impact these parameters, and how do they in turn affect the steps a decision maker takes to reach a goal?

In this text, I provide and discuss a simulation of the impact of alternative ways to govern these decision parameters on the number of steps the decision maker takes to reach a goal.

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.

The simulation is set up as follows. The decision maker starts from a specific cell in a grid, and there is a cell that it needs to reach, called the goal. At each step, it can decide to stay in the same cell or move to an adjacent one. Decision governance consists of rules which influence the perception the decision maker has of the alternative options it has at each step.

Five parameters influence which option the decision maker takes at each step. The parameters are:

  • Attention, or which option is given more or less prominence,
  • Memory, influenced by the presence or absence of an accurate log of past steps,
  • Competence, or whether the decision maker was given training on how to reach the goal,
  • Expectations, which are influenced by framing the goal as being aligned or not with the decision maker’s expectations, and
  • Reputation, which is influenced by framing how the decision maker perceives its proximity to the goal being aligned or not with its perception of its own reputation.

It is assumed that the design of decision governance can impact each parameter in different ways. For example, we can design governance which helps focus attention in the right way, but at the same time incorrectly frame the goal in relation to the decision maker’s reputation – loosely speaking, we might paint the wrong picture of what the goal is, and so the decision maker, even if decision governance directs their attention to the goal, do not perceive it as something that they value pursuing, as they perceive it as harming its reputation.

To simulate the effect of decision governance in the way described above, we need to make assumptions about how decision governance impacted each of the five parameters. While there are potentially many ways to do this, the idea below is to keep things simple, and assume that decision governance will lead to one of three possible outcomes over each parameter. These are as follows.

  • Decision governance influenced the decision maker’s attention in one of the following ways:
    • Attention is directed to the goal, in which case it will prefer the options that get it closer to the goal at each step.
    • Attention is directed to a random option, that is, decision governance ends up confusing the decision maker.
    • Attention is directed away from the goal, a case of decision governance which is steering attention in a way that is harmful to the decision maker’s ability to reach the goal.
  • Decision governance influenced the decision maker’s memory in one of the following ways:
    • The decision maker’s path is correctly logged, and the decision maker reads that log accurately, so it would only go across the same parts of its path if it has no other choice.
    • The decision maker’s path is not logged at all.
    • The decision maker’s path is incorrectly logged, or it may be logged correctly, but the decision maker cannot accurately read and understand it. 
  • Decision governance influenced the decision maker’s competence in one of the following ways:
    • The decision maker received training on how to reach the goal.
    • The decision maker received no training on how to reach the goal.
    • The decision maker received training, but it is counterproductive.
  • Decision governance influenced the decision maker’s expectations in one of the following ways:
    • The decision maker formed expectations aligned with the goal.
    • The decision maker formed expectations that are independent from the goal.
    • The decision maker formed expectations that are competing with the goal.
  • Decision governance influenced the decision maker’s reputation in one of the following ways:
    • The decision maker perceives its proximity to the goal as improving reputation.
    • The decision maker perceives its proximity to the goal as independent from its reputation.
    • The decision maker perceives its proximity to the goal as harming its reputation.

There are 243 combinations of the potential impacts of decision governance on the decision maker, across the five parameters. One simulation involves taking one of the combinations, and computing the number of steps the agent would take to reach the goal, and the standard deviation of that number of steps.

The following chart shows the average number of steps on the vertical axis, over the 243 different combinations on the horizontal axis. Combinations which lead to the lower number of steps are more desirable, and are towards the right hand side of the chart. The combinations are not labeled in this chart, we will go into this below.

The reason the maximal number of steps is set to 10,000, is to stop simulation runs at a point where, for the scale of the grid (30×30, or 900 positions) it makes little sense to want to have the agent take further steps. 10,000 is arbitrary, but looks large enough, as it means that the agent visited many positions many times, without ending up at in the goal one. 

The chart above shows that there are combinations of parameters that lead the agent to be, so to speak, confused to the extent of being unable to reach the goal at all. All combinations which yield fewer than 10,000 steps are combinations that lead the agent to reach the goal.

Having seen above that not all combinations are equally effective, we need to look into them in detail. To do so, we can visualize each as follows. The combination is shown as a line connecting the outcomes of decision governance on each of the five parameters. In this configuration, attention is directed away from the goal, memory results in agent’s path being correctly logged, training was provided and is effective, expectations are aligned with the goal, and reputation improves the closer the agent is to the goal. With this configuration, the agent takes always the same number of steps, 59, as standard deviation is 0.

Remember that these parameters are simplifications. In actual decision situations, we can never isolate one parameter, such as attention, and influence it independently from others. Moreover, it is hard to say which parameters there are in general – I wrote about the design space for decision governance in another text, here, and that was a simplified view of the topic, given the variety of research on decision making, across philosophy, psychology, economics, management, and then across various specific domains, such as, e.g., medical decision making.

Despite these limitations, which will apply across various approaches to simulating decision making behavior (not only the one I am describing here), the reason the 243 combinations are practically interesting, is that they capture some intuitive ideas about what needs to be done, in actual decision making, to help the decision maker through governance. 

Below is another combination. When governance is set up as described in the diagram, the decision maker does not reach the goal in the first 10,000 steps. In this case, attention is not directed, or is not governed, the decision maker does have a log of the steps it took, it received counterproductive training, it has expectations which are opposed to the goal, and it does see its reputation as increasing with proximity to the goal. While this may seem like an abstract case, if you have experience with decision making in firms, you might have encountered a situation which this case describes quite well: there is little management direction and mentoring when the decisions are being made (hence no governance of attention), there are tools that record logs of decisions (meeting minutes, documents, emails, chats, and the like), training was of low quality, not much effort was invested to align expectations, and at the same time, pursuing the goal is perceived as helping, say, career advancement.

The following table shows all 243 simulated cases, and the corresponding average number of steps and standard deviation of the number of steps.

A practical way to use the table is to make assumptions about what you can do in terms of governance in a decision situation, and see what potential impacts it can have relative to other cases. In some cases, you can, for practical reasons such as available time and other resources, govern only specific parameters, and need to ignore others. For example, you can influence attention but not competence, or attention and memory and nothing else. The table shows what kind of damage you can expect, relative to a situation in which you would be able to do better.