As decision governance influences how decisions are made, everyone who participates in preparing a decision, makes the decision, and lives with the consequences of it, is a stakeholder in the design and change of decision governance.
Decision Governance refers to values, principles, practices designed to improve the quality of decisions.
To override a decision, you need to know a decision was made (observability), have rights to override it (authority), and believe that doing so will lead to a better outcome, including preventing undesirable outcomes (superiority).
If the decision maker is distracted, they should pay the price of reaching their goal after more effort. A simple simulation can be done to show just how much distraction may cost relative to a case when the agent’s attention is directed to the goal.
Social distance, in the context of decision making, refers to the decision maker’s perception of similarity to others that may somehow matter for the decision at hand.
What are the parameters of a decision situation that we want to influence through decision governance?
If we want to influence preferences away from specific options, we should invest more effort to change high-level construal of their distant outcomes, rather than low-level construal of their immediate outcomes.
If we need to design governance that influences attention, then it matters if we know or not the goal of the decision maker. This text provides a simple simulation that illustrates the differences between the time it takes for the decision maker to reach the goal in both cases, all else being equal.
Three decision governance strategies are compared in terms of how they influence the ability of an agent to reach their goal in a simple problem: the first strategy involves no governance, the second complement’s agent’s memory, and the third draws their attention.
Decision governance can be designed to make decision makers aware of their and others’ emotions in a decision situation, and to help everyone move to a more neutral stance, if that can lead to a better outcome. This text outlines common strategies for doing so.
It seems obvious that it makes no sense to randomly choose between options we are presented with. In this text, I’ll set up and run a simple simulation that illustrates this. The simulation is another way to think about the impact of decision governance, even if in a very simple setting.
Ease at which memory will be accessed, accuracy of memories, association of stimuli with memories they lead to, will all influence the information that a decision maker will use. Decision governance can to some extent influence what is recalled, how that is related to the choice at hand, and where attention is drawn.
Decision governance can neutralize or amplify factors driving attention in a decision situation. The choice of strategy depends on the observed or anticipated behavior of the decision maker and the desired outcome.
We should reduce the cost of authorship and create an incentive mechanism that generates and assigns credibility to authors in a community.
If the time to decide is short, decision governance needs to improve how the decision maker identifies cues, matches them to experience, what they experience they match them to, and the quality of their prediction of action outcomes.
If a decision process is designed according to the expected utility (maximization) model, then the choice of an option is explained by it having the highest expected utility among considered options. Consequently, decision governance over such a decision process needs to help the decision maker predict and prepare for unexpected events in order to maximize utility.
If a decision process is designed according to the classical utility maximization model, then the choice of an option is explained by it having the highest utility among considered options. Consequently, decision governance over such a decision process needs to influence (i) which options are considered, (ii) how options are compared against preferences, and (iii) how preferences are formed.
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…
Explanations of decisions are central to decision governance: before changing how decisions are made, you need to explain how they are made; you need to explain why they need to be changed; and, you need to explain how changes that governance brings should lead to better decisions. So the question is: What is a good…
Let’s assume that there is a situation you observed, and you want to understand the decision that led to it – maybe there is something particularly good about the situation and you want to see how to increase the probability that this happens again, or there is something you would want to prevent from happening…
Is it one that led to the best outcome? Or one that integrates all the relevant and available information? Maybe one that is liked by a majority? If decision governance is followed to the letter, will that guarantee a high quality decision? The quality of a decision depends on the following: The reason a decision…
How many options will be identified when a decision needs to be made? How much thought will go into developing a robust rationale for each option? Doing both of these takes effort. Unless there are incentives to invest effort, a decision will be made from one or few low quality options. That is a simple…
A business process describes how something is done by highlighting mainly the actions to take, their dependencies (including their sequence), the roles in the firm who do these actions, as well as what triggers the process to start, and how we know when the process ends. Business processes implement decision governance in several ways. It…
The org chart shows much more than who reports to who. It is one of few core tools for learning about the existing decision governance setup, as well as for planning and implementing changes to how decisions are made. If you look across the org chart vertically, across reporting lines, here is what you can…
Decision rights will be exercised, and decision obligations discharged only if there are incentives to do so. If you need to make a decision and bear the consequences, i.e., exercise decision rights and discharge obligations, the only reason to do so is if you see how it makes sense with regards to what you want….
I wrote in another note (here) that AI cannot decide autonomously because it does not have self-made preferences. I argued that its preferences are always a reflection of those that its designers wanted it to exhibit, or that reflect patterns in training data. The irony with this argument is that if an AI is making…
Being entitled to make decisions carries with it the responsibility for outcomes of actions that the decisions led to. Accountability can be implemented through decision governance by defining responsibilities for outcomes of decisions. The idea that decision responsibilities are the counterpart to decision rights is easy to understand. However, defining useful decision responsibilities involves finding…
Decision rights are entitlements to act in a certain way and have access to specific information and resources required to make decisions. An executive may be asked to decide if an investment should be made or not, a manager may be deciding between candidates to hire – both are entitlements to make a decision. The…
This is the first of several notes which will introduce concepts necessary to design and do decision governance. The aim is to develop a more precise idea of what decision governance is, how it works, and what it means to design it and evaluate its benefits and costs. The focus in this first note is…
There are three ways to think about how much decision governance to do. I will call them The “overall value” approach consists of comparing an estimate of benefits of all decision governance in place, with the costs of complying with it. Benefits include: Both of the above can be attributed to decision governance only if…
How can you tell if there is a need to do anything to influence how decisions are made, that is, to govern decisions? If any of the following apply, then it is worth investing effort to improve how decisions are made.
A decision is a commitment to a course of action. There is no way to see commitment, which is a problem if you want to spot decisions. Instead, you can see actions people take, and infer from that what they may have committed to – note that you are inferring what they may think or…
The corporate function or group will usually have in its scope to decide how decision governance works in a firm. Key design decisions they will make involve who can design how decisions are made, how they can do so, including how to handle changes to, exceptions, and incidents related to how decisions are made. When…
Decision governance consists of defining how decisions should be made, and auditing that the processes for doing so are in fact applied. While it often seems like decisions are simply made, and there isn’t much to govern, this is incorrect. It is possible to completely change how to think about decisions, in particular in terms…
Decision governance puts constraints on how decisions are made: e.g., assess impacts of decision options before picking one, estimate probabilities of outcomes of options, elicit preferences of decision makers, and so on. In other words, explain the reasons for a decision before deciding. If these constraints are a source of competitive advantage, then this means…
In the context of human decision making, a decision is a commitment to a course of action (see the note here); it involves mental states that lead to specific actions. An AI system, as long as it is a combination of statistical learning algorithms and/or logic, and data, cannot have mental states in the same…
It has been successfully argued in research on organizations that a decision is an abstraction [1], if it is defined as a commitment to a course of action. If I say that I decided to go skiing tomorrow, is this abstract or concrete? The only thing you can observe will be my behavior and the…
Let’s start with the optimistic “yes”, and see if it remains acceptable. Before we get carried away, a few reminders. For an LLM to be a source of competitive advantage, it needs to be a resource that enables products or services of a firm “to perform at a higher level than others in the same…
I use “depth of expertise” as a data quality dimension of AI training datasets. It describes how much a dataset reflects of expertise in a knowledge domain. This is not a common data quality dimension used in other contexts, and I haven’t seen it as such in discussions of, say, quality of data used for…
Just like l’art pour l’art, or art for the sake of art was the bohemian creed in the 19th century, it looks like there’s an “AI for the sake of AI” creed now when building general-purpose AI systems based on Large Language Models. Let’s say that the aim for a sustainable business are happy, paying,…
As currently drafted (2024), the Algorithmic Accountability Act does not require the algorithms and training data used in an AI System to be available for audit. (See my notes on the Act, starting with the one here.) The way that an auditor learns about the AI System is from documented impact assessments, which involve descriptions…
Artificial Intelligence, if incorrectly defined, is even more confusing than it can be. Sometimes, it is considered a technology, which itself is problematic: is it a technology on par with database management systems, for example, which are neutral with respect to the data they are implemented to manage in their specific instances? Or, is it…
The less data there is, or the lower quality the data that is available, the more difficult it is to build AI based on statistical learning. For scarce data domains, the only way to design AI is to elicit knowledge from experts, design rules that represent that knowledge, parameterize them so that they apply to…
The short answer: careers that reward creative problem solving in domains with scarce knowledge. Let’s unpack that.
If AI is made for profit, then should its design be confidential? This choice is part of AI product strategy. The decision on this depends on the following at least. What is the relationship of each of these to AI confidentiality? Correctness: The more likely the AI / algorithm is to make errors, the more…
The short answer is “No”, and the reasons for it are interesting. An AI system is opaque if it is impossible or costly for it (or people auditing it) to explain why it gave some specific outputs. Opacity is undesirable in general – see my note here. So this question applies for both those outputs…
Opacity, complexity, bias, and unpredictability are key negative nonfunctional requirements to address when designing AI systems. Negative means that if you have a design that reduces opacity, for example, relative to another design, the former is preferred, all else being equal. The first thing is to understand what each term refers to in general, that…
Should the explanations that an Artificial Intelligence system provides for its recommendations, or decisions, meet a higher standard than explanations for the same, that a human expert would be able to provide? I wrote separately, here, about conditions that good explanations need to satisfy. These conditions are very hard to satisfy, and in particular the…
How good of an explanation can be provided by Artificial Intelligence built using statistical learning methods? This note is slightly more complicated than my usual ones. In logic, conclusions are computed from premises by applying well defined rules. When a conclusion is the appropriate one, given the premises and the rules, then it is said…
If there is a market for AI training datasets, then the price will be determined by supply and demand. How does the supplier set the price, and how does the buyer evaluate if the price is right? The question behind both of these is this: how to estimate the value of a training dataset? We…