Decision Making Under Risk and Uncertainty

Introduction

Decision-making is a crucial task in day-to-day life. Decision-makers should be able to make decisions under conditions of risk and uncertainty. It means that decision-makers should analyze and assess several options regarding their probable outcomes and likelihood. As you can see, decision-making can be classified into two categories based on the surrounding environment. The first category of decisions is those made under uncertainty, i.e., situations involving a lot of uncertainty. More simply, it involves situations where the results are not known in advance. The other category of decisions involves decisions that are made under risk, i.e., the decision taken has some risk associated with it. Risk-based decisions are made in circumstances when it is known or possible to assess the likelihood of each possible result. In these situations, decision-makers might examine the potential outcomes using statistical approaches to determine the optimal course of action.

What is Decision Making in AI?

Artificial intelligence (AI) has become widespread across various sectors, including finance, healthcare, and transportation. Integrating algorithms and data using artificial intelligence is required to make decisions. Large datasets can be evaluated by AI systems to uncover patterns and trends that may be difficult for humans to perceive. AI decision-making involves several steps.

Firstly, outlining the issue that needs to be solved, variables, and limitations that will affect the choice are determined. Next, the relevant data is gathered and organized to facilitate analysis. Data analysis is conducted in the third step of the process using statistical methods and machine learning algorithms. Patterns and trends in the data are identified at this stage, and they are used to forecast the future. Ultimately, the decision is made after data analysis, and the optimal course of action is chosen based on the likely outcomes and their probabilities.

How to Make Decisions Under Uncertainty?

Decision-making under uncertainty involves situations where the outcome of a decision is unknown. Decision-makers must consider multiple possible outcomes and their probabilities in such cases. There are several techniques that decision-makers can use to make decisions under uncertainty, including the Laplace criterion, Maximin, Maximax, Hurwicz, and Minimax regret.

Laplace Criterion

The Laplace criterion is a decision-making technique that can be utilized to make decisions under uncertainty using AI. It is used to make decisions in situations where the probability of each outcome is unknown or cannot be estimated. This technique assumes that each outcome is equally likely and assigns equal weight to each of them. In this method, we calculate the average value of each scenario irrespective of the probability of occurrence of each scenario. The decision-makers select the scenario with the highest average value. AI tools support decision-makers in putting the Laplace criterion into practice. Large datasets are analyzed, and the average result for each decision is computed. AI can be used to find trends and patterns in data. The future results of any decision can be predicted with the assistance of these patterns.

Maximin

The Maximin criterion is a decision-making technique that can be used to make decisions under uncertainty using AI. When faced with a situation where the probability of each outcome is unknown or cannot be estimated, decision-makers can employ this technique to select the best course of action. The Maximin criterion assumes that the worst possible outcome of a decision is the most important consideration. It means that decision-makers must consider the most negative result of each decision.

AI tools support decision-makers in putting the Maximin criterion into practice. Large datasets are analyzed, and the worst possible outcome for each decision is computed. AI can be used to find trends and patterns in data. We can predict the future results of any decision with the assistance of these patterns. It is worth noting that the Maximin criterion may not always lead to the optimal decision since it only considers the worst possible outcome. Nonetheless, it can be an effective technique when decision-makers prioritize avoiding the worst-case scenario.

Let's say you are playing a game with a friend and have to choose between two options: Option A gives you a guaranteed win of Rs. 5/-, while Option B provides a 50-50 chance of winning either Rs.10/- or Rs.0/-. If you use the Maximin strategy, you will choose Option A because it gives you the highest minimum payoff (i.e., Rs. 5/-) compared to Option B, which has a minimum payoff of Rs. 0/- if you lose the coin toss.

Maximax

The Maximax criterion is a decision-making technique that can be used to make decisions under uncertainty using AI. When faced with a situation where the probability of each outcome is unknown or cannot be estimated, decision-makers can employ this technique to select the best course of action. The Maximax criterion assumes that the best possible outcome of a decision is the most important consideration. The decision with the highest outcome is selected as the output.

Suppose you are a product lead manager considering launching a new product. You have estimated the potential profits for the new product under three different scenarios: low demand, moderate demand, and high demand. The estimated profits for each scenario are as follows: Low demand will give Rs. 10,000/- profit, moderate demand would provide Rs. 50,000/- profit, and a high demand would give Rs. 1,00,000/- profit. Using the Maximax strategy, you would choose the option that maximizes the potential profit in the best-case scenario. The best-case scenario is high demand, potentially generating Rs. 1,00,000/- profit. Therefore, you should launch the new product because it has the highest potential profit in the best-case scenario.

Hurwicz

The Hurwicz technique helps choose a decision that balances good and bad outcomes. This technique uses a coefficient that decides how much weight should be given to the best and worst outcomes. The best and worst outcomes are equally important when the coefficient's value equals 0.5. Only the best outcome is considered if the coefficient takes a value equal to 1. On the other hand, only the worst outcome is considered if the coefficient takes a value equal to 0.

Minimax Regret

The Minimax regret technique involves choosing the decision that minimizes the maximum regret. Regret is the difference between the best outcome and the outcome of the chosen decision. This technique aims to minimize the regret of making a suboptimal decision.

How to Make Decisions Under Risk?

Decision-making under risk involves situations where the probability of each outcome is known or can be estimated. This helps decision-makers to use statistical methods to analyze the options and make the best decision. To make decisions under risk, decision-makers might employ a variety of strategies as discussed below:

Maximum Expected Value

The maximum expected value method is the first way that decision-makers might use to assess and contrast various options. Finding the option with the highest expected value is the key to employing this strategy. The expected value is calculated as the product of all possibilities and their corresponding probabilities. The choice that has the best probability of rewarding them or offering benefits can subsequently be made by decision-makers. This method is beneficial when decision-makers are judged to be risk-neutral, that is, neither risk-avoiders nor risk-seeking, and are therefore only concerned with maximizing their expected gains. Using the maximum expected value technique, decision-makers can decide the optimal course of action to assist them in reaching their intended goals by basing their decisions on the probability associated with each possible outcome.

Suppose you are a sales manager and must decide whether to launch a new product line. You have estimated the potential profits for the new product line under two different scenarios: a 90% chance of moderate demand, which would result in an Rs. 50,000/- profit, and a 10% chance of low demand, which would result in an Rs. 10,000/- loss. To use the Maximum Expected Value strategy, you would multiply each scenario's potential profit or loss by its probability and add them together. In this case, the expected value of launching the new product line would be:

The expected value of launching the new product line is Rs. 46,000/-, which is the sum of the potential profits and losses weighted by their probabilities. Since the expected value is positive, you can launch the new product line since it has the highest expected profit. However, it is important to note that the Maximum Expected Value strategy does not consider each scenario's potential risks or uncertainties.

Maximum Utility

The maximum utility technique is a popular approach to decision-making that involves selecting the decision with the highest expected utility. Utility refers to the value or satisfaction a decision can provide an individual. In contrast to the maximum expected value technique, the maximum utility technique considers risk-avoiding decision-makers and assigns more significant value to outcomes with a lower probability. This technique is often used when decision-makers want to maximize the overall satisfaction of a decision rather than merely its expected gains. This technique allows decision-makers to account for the potential trade-offs between risks and rewards, which can help them make more informed decisions. Additionally, the maximum utility technique can help decision-makers to prioritize their preferences and determine which outcomes are most desirable, given their risk tolerance and personal preferences. Overall, the maximum utility technique is a powerful tool for decision-makers who want to optimize their decision-making process under conditions of uncertainty and risk.

Most Probable Outcome

Decision-makers may opt to use the most probable outcome technique when a risky option is present. This approach decides with a high chance of success. This method seeks to reduce the level of uncertainty while making decisions. It helps in giving a thorough understanding of the possible outcomes. This approach can be especially useful when decision-makers seek to avoid risks and want some degree of certainty in their choice. By selecting the most probable outcome, decision-makers can make well-informed decisions with high confidence.

Composite Criteria

The composite criteria technique includes integrating several factors to produce a judgment. This method gives several criteria with varied weights while considering the preferences and priorities of the decision-maker.

Analysis Methods

Decision-makers can examine the results of their decisions in a variety of ways. These techniques include Monte Carlo simulations, sensitivity analysis, and decision trees. Decision trees visually represent a decision's potential outcomes and associated probabilities. They help figure out the best course of action and visualize complicated decisions. Sensitivity analysis comprises altering the inputs and assumptions used in the study to see how robust a choice is. This method can assist decision-makers in locating the crucial elements that influence a decision's outcome. In Monte Carlo simulations, the variables employed in the analysis are given random values, and the potential outcomes of a choice are simulated. With the help of this technique, decision-makers can better comprehend the range of potential outcomes and the likelihood of each one.

Compare Making Decisions Under Uncertainty vs Making Decisions Under Risk

The following table shows a comparison between making decisions under uncertainty and making decisions under risk:

Risk and Risk Management in AI

Risk management is crucial in AI because bad decisions made by AI systems could have serious repercussions. AI hazards include, among others, ethical, legal, and social risks. Decision-makers should consider the following elements to reduce these risks:

• Identifying the risks: Identifying and evaluating the hazards related to the AI system is the first stage in risk management. This step involves evaluating each outcome that the AI system might produce and any potential repercussions.
• Creating the AI system's design: The hazards should be considered in the AI system's design to minimize them. This step entails choosing the right analytical and decision-making processes as well as developing an understandable and transparent AI system.
• Testing and validation: The AI system needs to be tested and verified to ensure it functions as planned and doesn't present any hazards. At this step, the AI system is put through a variety of scenarios to test its performance in comparison to the risks that have been identified.
• Monitoring and maintenance:  To ensure the AI system stays secure and functional, it must be regularly checked and maintained. At this step, the AI system's performance is monitored, and it is updated to handle any dangers or problems that are found.
• Accountability and transparency:  With detailed descriptions of its decision-making procedures and the variables impacting those judgments, the AI system should be open and accountable. Making sure the AI system can be explained and that decision-makers can comprehend the reasoning behind it is part of this stage.