Lernzettel: AI in Economics and Data Science

📋 Course Outline

1. AI for Economic Decision-Making
2. History and Evolution of AI
3. AI Definitions and Applications
4. Supervised Learning and Naive Bayes
5. Unsupervised Learning and K-Means
6. Hybrid Recommendation Systems
7. Data Collection Methods
8. Data Analysis and Visualization
9. Advanced Data Analysis Techniques

📖 1. AI for Economic Decision-Making

🔑 Key Concepts & Definitions

• Economic decision-making : economic decision-making is the process of choosing actions based on information to improve outcomes in markets, businesses, or public administration.
• Artificial intelligence : Artificial intelligence is the ability of computers to perform tasks resembling human cognition and behavior using software and hardware methods.
• Machine learning : Machine learning is an AI approach that uses large datasets to make predictions or decisions rather than relying on fixed instructions.
• Predictive economic modeling : Predictive economic modeling uses AI methods to forecast market trends, consumer behavior, and economic impacts.

📝 Essential Points

• AI education targets applying AI to solve economic problems, improve decisions, and optimize economic processes.
• Students practice analyzing large economic datasets with AI tools to support decision-making.
• AI is presented as enabling forecasts of market trends, consumer behavior, and economic impacts.
• In economics, AI supports strategic scenarios and demand prediction through predictive analytics and simulation.
• AI systems are linked to practical dashboards that support performance management using analyzed data.

💡 Memory Hook

AI for economics: Data → predictions (markets/consumers) → decisions (scenarios/dashboards).

📖 2. History and Evolution of AI

🔑 Key Concepts & Definitions

• Artificial intelligence as a discipline : Artificial intelligence is treated as an academic discipline whose origins are traced back to the 1950s.
• Turing Test : The Turing Test is a proposal by Alan Turing that assesses whether a machine shows intelligence through interaction with a human.
• Artificial intelligence term (1956) : In 1956, John McCarthy coined the phrase artificial intelligence at a conference he organized.
• Logic Theorist : Logic Theorist is an early 1956 AI program focused on proving mathematical theorems.
• ELIZA : ELIZA is a 1967 program created by Joseph Weizenbaum to stimulate therapist–patient style interactions.

📝 Essential Points

• AI research between 1956 and 1966 was mostly theoretical, aiming to model human thought and imitate human behavior in machines.
• Sad Sam (1960) was an AI program by Robert K. Lindsay that generated conclusions in English from simple sentences.
• ELIZA was developed in 1967 by Joseph Weizenbaum to simulate interactions with patients in a therapeutic setting.

💡 Memory Hook

Turing (test), McCarthy (name), then programs: Logic Theorist (prove), Sad Sam (say), ELIZA (therapy chat).

📖 3. AI Definitions and Applications

🔑 Key Concepts & Definitions

• Contract classification : Contract classification is an AI justice application that identifies and organizes contract content for automated legal analysis tasks.
• Adaptive traffic lights : Adaptive traffic lights are a transport AI use where signal control adjusts to reduce road congestion.
• HR chatbot : An HR chatbot is an AI assistant that answers employee questions within recruitment and human-resources management workflows.

📝 Essential Points

• Machine learning differs from traditional programming because it relies on learning from data to produce predictions or decisions rather than executing explicit instructions.
• AI is used to predict the spread of epidemics and support medical chatbots that answer health questions.
• In transport, AI supports autonomous vehicles, adaptive traffic-light control to reduce congestion, and route/time planning based on actual demand.
• In public administrations, AI analyzes large public databases for effective decision-making and helps monitor transactions to detect fraud and anomalies.
• In business management, AI supports automatic CV filtering, candidate–role matching, KPI analysis for team productivity, and predictive analytics such as market-demand forecasting plus scenario simulation.

💡 Memory Hook

AI vs traditional programming: rules-first instructions vs data-first learning for predictions and decisions.

📖 4. Supervised Learning and Naive Bayes

🔑 Key Concepts & Definitions

• Supervised learning : A supervised learning approach trains models on labeled data so the algorithm can learn to map inputs to the correct class or output.
• Support Vector Machine : A Support Vector Machine is a supervised classifier that separates classes by finding a hyperplane using support vectors near the decision boundary.
• Support vectors : Support vectors are the data points closest to the decision limit that determine the position of the SVM separating hyperplane.
• Naive Bayes : Naive Bayes is a supervised probabilistic classifier built on Bayes’ theorem and a simplifying assumption that features are independent given the class.

📝 Essential Points

• A Support Vector Machine searches for a hyperplane that separates classes while maximizing separation using support vectors closest to the decision limit.
• Naive Bayes assumes a class y makes the observed characteristics in X independent of one another.
• Naive Bayes computes the posterior class probability using $P(y\mid X)=\dfrac{P(X\mid y)\,P(y)}{P(X)}$.

💡 Memory Hook

SVM: “support vectors set the border”; Naive Bayes: “Bayes + feature independence” leads to $P(y\mid X)$.

📖 5. Unsupervised Learning and K-Means

🔑 Key Concepts & Definitions

• Unsupervised learning : Unsupervised learning is a machine learning approach that uses algorithms to find patterns in data without using labelled answers.
• K-means : K-means is a machine learning algorithm used within unsupervised learning to group data into clusters.

📖 6. Hybrid Recommendation Systems

🔑 Key Concepts & Definitions

• Purchasing behaviour : Purchasing behaviour refers to how customers buy over time, producing signals that can be used to infer preferences or group them.
• Customer category : A customer category is a grouping assigned to a customer based on their purchasing behaviour patterns.

📖 7. Data Collection Methods

🔑 Key Concepts & Definitions

• Data collection : Data collection is the process of gathering relevant information so you can identify economic behaviour patterns and train an AI predictive model.
• Web scraping extraction : Web scraping extraction is an automated way to collect data from websites for later use in training predictive models.
• Consumer surveys and forms : Consumer surveys and forms are questionnaires used to collect responses that reveal patterns in consumer behaviour.
• Real-time sensor data : Real-time sensor data are measurements collected from sensors to capture conditions as they change over time.
• Interaction logs : Interaction logs are records of user actions from website or mobile app activity that support analysis for predictive modelling.

📝 Essential Points

• Data collection methods include manual or automated extraction via web scraping, surveys/forms on consumer behaviour, sensors for real-time data, and website/mobile app interaction logs.
• The goal of data collection is to obtain data patterns of economic behaviour that can be used to train a predictive model.
• Employee-role prediction can use data collection items such as questionnaire, HR interview, CV, LinkedIn profile, tests, and direct observation.
• Product-selection prediction can use data collection items such as online questionnaires, ratings, and analysis of comments using NLP plus product standards.

💡 Memory Hook

Web→Talk→Sensors→Click logs (web scraping, surveys, sensors, interaction logs) to get training data for predictions.

📖 8. Data Analysis and Visualization

🔑 Key Concepts & Definitions

• Model accuracy : Model accuracy measures the fraction of predictions a model gets correct out of all predictions it makes.
• True positives (TP) : True positives are positive cases that the model predicts as positive.
• False negatives (FN) : False negatives are positive cases that the model predicts as negative.

📝 Essential Points

• Accuracy can be computed as $\frac{\text{NB of correct predictions}}{\text{total number of predictions}}$.
• Accuracy can also be computed as $\frac{TP+TN}{TP+TN+FP+FN}$.
• True negatives (TN) are negative cases predicted as negative, while false positives (FP) are negative cases predicted as positive errors.
• In this setup, false negatives (FN) are positive cases predicted as negative errors.

💡 Memory Hook

Accuracy = (TP + TN) ÷ all outcomes; keep TP/TN as the “right calls.”

📖 9. Advanced Data Analysis Techniques

🔑 Key Concepts & Definitions

• Predictive analytics : Predictive analytics is an approach that uses data patterns to anticipate future events, such as machine failures, before they occur.
• Sensor data monitoring : Sensor data monitoring is the analysis of readings from equipment to detect signals indicating abnormal behaviour and likely breakdowns.
• Real-time trend detection : Real-time trend detection is the analysis of incoming data streams to spot trends or anomalies as they happen.
• Algorithmic auditing : Algorithmic auditing is a structured examination of AI systems to check compliance with requirements and best practices.

📝 Essential Points

• Predictive analytics based on sensor data can anticipate failures and trigger preventive maintenance planning to improve machine reliability.
• Real-time analysis of large economic or financial datasets helps identify trends or anomalies.
• Public policy can require algorithmic auditing mechanisms for AI systems.
• Public policy can introduce labels or certifications for AI systems that comply with best practice.
• Public policy can monitor societal, economic, and environmental impacts of AI to limit risks.

💡 Memory Hook

Failures: sensors → predictive analytics → preventive maintenance.

📅 Key Dates

📊 Synthesis Tables

Data collection methods: comparisons

Deep learning vs machine learning

⚠️ Common Pitfalls & Confusions

1. Confusing supervised learning (labelled data with a learned mapping to outputs) with unsupervised learning (unlabelled data to discover structure/clusters).
2. Assuming Naive Bayes removes the need for probabilities; it still computes P(y|X) using Bayes’ theorem and the independence assumption.
3. Mixing up SVM’s goal (separating classes with a hyperplane using support vectors near the decision limit) with k-means clustering (partitioning data into K clusters by distance).
4. Treating data collection as the model itself; the course frames predictive modelling as a pipeline: data collection → analysis/visualisation → prediction → testing/evaluation.
5. Using only accuracy = (NB correct predictions)/total without knowing the alternative TP+TN formula and what TP/TN/FN/FP represent in this setup.
6. Believing collaborative filtering recommendations use item content alone; it predicts user preferences by similarity between users and their ratings.
7. Thinking “algorithmic auditing” is just general model testing; the course presents it as a structured examination to check compliance with requirements and best practices.

✅ Exam Checklist

1. Define economic decision-making and explain how AI supports it in economic contexts (forecasts, scenarios, dashboards).
2. List core AI fundamentals introduced in the course: AI as an academic discipline (1950s), Turing Test, and the 1956 McCarthy naming context.
3. Describe the early AI programs and their focus: Logic Theorist (1956), Sad Sam (1960), and ELIZA (1967).
4. Differentiate machine learning from traditional programming by stating data-first learning vs fixed instructions.
5. Explain supervised learning and what “labelled data” enables, including classification vs regression.
6. For SVM: state the hyperplane objective and the role of support vectors closest to the decision limit.
7. For Naive Bayes: state the Bayes theorem formula P(y|X)=P(X|y)P(y)/P(X) and the independence assumption.
8. Explain unsupervised learning goals and the clustering idea, then describe k-means (non-hierarchical partitioning) including the meaning of K and the centroid-distance assignments.
9. For reinforcement learning: state the reward/penalty mechanism and that it learns an agent’s best decisions autonomously in an environment.
10. Explain deep learning vs machine learning using the course distinctions (depth of neural networks; dependence on structured/labelled vs raw/unstructured data).
11. Reproduce the k-means procedure steps (initialise K centroids → assign to nearest centre → recompute centroids → check validity/iterate).
12. For predictive models: describe the full pipeline (data collection methods; analysis/visualisation methods such as PCA/MCA/time series/discriminant; prediction; testing/evaluation with accuracy formulas and TP/TN/FP/FN).