The media (and movies) have made AI conversations confusing and too emotional. I propose a simple exercise for middle and high school students (and others who might be interested) to provide hands-on training to understand how an AI model works. That exercise will focus on defining, designing, and implementing the AI Utility Function that drives the AI model’s recommendation and decision-making processes.
The exercise will be centered around something we do nearly every day: deciding where to dine. We will follow this four-step process in defining, creating, and then executing our “Dining Recommendation” AI Utility Function:
- Step 1: Define AI Utility Function Variables and Metrics. In a collaborative setting, students will brainstorm and identify the variables that “might” impact their dining decisions. Examples could include affordability, food quality, and ambiance. This step establishes what factors are considered in the AI model and introduces students to AI model variable selection based on intent, relevance, and impact. We will narrow down that list to the top 10 variables for the rest of the exercise.
- Step 2: Create Entity Models (Analytic Profiles). Utilizing the prioritized variables, we will employ ChatGPT to construct entity models or analytic profiles for each selected restaurant. We will rate each variable on a scale from 1 to 100 for each restaurant. A high rating (80 – 100) indicates that the restaurant strongly represents the variable, while a low ranking (1 – 20) suggests that the restaurant does not represent that variable well. For example, if ‘ambiance’ is a variable, a fine dining restaurant might score 90 on this scale, signifying a solid correlation with the variable ‘ambiance.’
- Step 3: Personalize Variable Weights. Students will assign weights to each variable based on their dining intent and preferences. For example, if a student is looking for a quick and affordable meal, they might assign higher weights to ‘convenience’ and ‘affordability.’ This refinement teaches them the importance of weighing the different variables that might impact the decision-making processes and tailoring the model to individual intent and desired outcomes.
- Step 4: Calculate AI Model Restaurant Recommendations. Finally, using the weighted variables, the AI Utility Function calculates and ranks each restaurant by generating a score that aligns with the student’s preferences. This step synthesizes the variables and their associated weights into a practical outcome, demonstrating how AI models can provide personalized recommendations based on user-defined criteria.
Hopefully, this exercise offers an excellent way to introduce middle and high school students (as well as curious seniors) to concepts of decision-making processes using AI methodologies and the importance of defining a healthy, representative AI Utility Function.
Refresher: What is an AI Utility Function
The AI Utility Function is a mathematical model that quantifies an AI system’s preferences, guiding it to make decisions that maximize the desired outcomes based on specified criteria and constraints.
The AI utility function is crucial in guiding AI models to generate more relevant, meaningful, responsible, and ethical outcomes. It defines what success looks like by quantifying preferences and trade-offs among different potential outcomes. For instance, an AI model might use an AI utility function that considers treatment effectiveness, side effects, and patient quality of life in a healthcare setting. This allows the AI model to recommend treatments that balance these factors in a way that aligns with the defined healthcare goals, ensuring the relevance and meaningfulness of the outcomes (Figure 1).
Figure 1: The AI Utility Function
Furthermore, by integrating ethical guidelines and responsibilities into the AI Utility Function, AI models can be directed to uphold human values and societal norms. For example, an AI model assigned to credit scoring can be programmed with an AI Utility Function that “penalizes” decisions, leading to unfair biases against certain groups. Moreover, the AI Utility Function can be adjusted in real time to reflect evolving human insights and regulatory requirements, ensuring that AI systems remain relevant and aligned with current ethical and societal expectations. As a result, AI utility functions not only provide a technical framework for decision-making but also serve as a means to ensure that AI technologies contribute positively and ethically to society.
Let’s go through an exercise to help every high school student (and hopefully everyone else) understand our role in defining and crafting an AI Utility Function to ensure more relevant, meaningful, responsible, and ethical outcomes.
Step 1: Define AI Utility Function Variables and Metrics
First, we will ask students to identify and brainstorm the different variables that might impact their decision on where to eat. These variables could include cost, distance, food type, and more. Encourage the students to think critically about why each factor is essential and how it might impact the recommendation process of the AI model. This will introduce them to the concept of variables in programming and data science and teach them to recognize how various factors can affect outcomes in everyday decisions.
In step 1, the students will work in small teams to brainstorm the variables and factors that could impact their choice of where to eat. This exercise aims to gather as many factors as possible without getting stuck in over-analysis. Below is a partial list of the factors the students might have identified (Figure 2).
Figure 2: Variables That Might Impact Deciding Where to Go Eat
I have selected ten variables to simplify the rest of the exercise. Although we could have used more variables to allow the AI model to provide even more accurate recommendations, I decided to stick with 10 to make it more manageable for our human minds, which can easily get overwhelmed.
The ten variables we will be using for the rest of the exercise are:
- Affordability
- Convenience
- Menu
- Ambiance
- Food quality
- Organic food
- Social Media reviews
- Value for Money
- Parking / Accessibility
- Hygiene / Cleanliness
Step 2: Create Entity Models (Analytic Profiles)
Analytic profiles (or entity models) are structured data frameworks that define and quantify the characteristics and behaviors of specific entities, such as customers or products, to support targeted analyses and predictive modeling.
Step 2 involves creating entity models, or analytic profiles, for a select number of restaurants. We can use ChatGPT to generate the variables’ weights that comprise the entity models and then tweak the variables’ weights based on our experience and further research.
Using the list of 10 variables, ChatGPT can generate an entity model for each selected restaurant, scoring the variables that comprise the entity models on a scale from 1 to 100 on how well they meet each variable. For example, a budget-friendly restaurant like McDonald’s might score high on affordability (80 – 90) but lower on organic food options (1 – 10). This step illustrates how data can be quantified and standardized to enable comparisons, reflecting how the AI Utility Function evaluates options based on predefined criteria.
We will create analytic profiles for each of these restaurants:
- McDonald’s is a global fast-food chain known for its affordable burgers, fries, and quick service.
- Chipotle is a fast-casual restaurant specializing in Mexican-inspired burritos and bowls. Its focus is on fresh, organic, and locally sourced ingredients.
- Applebee’s is a popular chain that offers a casual dining experience. The menu includes burgers, chicken, pasta, and bar-style appetizers.
- Flemings Steak House is a high-end steakhouse known for its prime meats, fine wines, and elegant dining atmosphere.
- Legal Seafood is a reputable seafood chain renowned for its commitment to freshness and quality. It serves various fish and shellfish dishes.
- LeTour is a sophisticated dining establishment that offers gourmet cuisine emphasizing artistic presentation and exceptional service.
Figure 3 contains the Analytic Profiles that ChatGPT built for our six selected restaurants, with my tweaks based on my personal experiences.
Figure 3: Restaurant Analytic Profiles or Entity Models
AI Utility Function Exercise Summary – Part 1
The “Becoming an AI Utility Function” exercise provides a practical and enlightening way for students to explore the inner workings of AI decision-making processes using a familiar daily activity—choosing where to dine. This exercise consists of four critical steps designed to improve understanding of how AI models incorporate different data points to make informed, personalized recommendations.
In Part 1, we covered the first two steps:
- In Step 1, students were introduced to the foundational concept of defining relevant variables and metrics, such as affordability, food quality, and ambiance. This stage is essential for setting up the criteria the AI will use later to evaluate options. Students learn to think critically about the factors that most influence their decisions, mirroring how AI systems must be programmed to consider various elements in their algorithms.
- In Step 2, we create entity models for each restaurant. Each establishment is scored on a scale of 1 to 100 based on defined variables in this step. This quantification process demonstrates to students how raw data and subjective assessments are standardized for fair comparisons across different options. It clearly illustrates how AI systems process and evaluate diverse inputs.
In Part 2, we will complete the exercise with steps 3 and 4. Hopefully, this hands-on exercise demystifies AI for students and highlights the importance of defining AI Utility Functions to ensure ethical and responsible AI outcomes. By understanding these processes, students can appreciate how AI models are designed to reflect human values and societal norms.
