2.1.2. AI, Machine Learning, and Generative AI
💡 First Principle: The terms describe nested circles, each a more specialized way of producing intelligent behavior. AI is the goal (machines performing tasks that normally need human intelligence). Machine learning is one route to that goal (learning patterns from data instead of following written rules). Deep learning is a powerful kind of ML (multi-layer neural networks that handle images, audio, and text). Generative AI is deep learning that creates new content rather than only classifying or predicting. Knowing which circle a problem lives in tells you what kind of solution, data, and risk to expect.
The distinction that matters most in business scenarios is between predictive (discriminative) AI and generative AI:
- Predictive ML answers a question about the input: Is this transaction fraudulent? How many units will we sell next week? Which customers will churn? Outputs are labels, scores, or numbers. It needs historical examples with known outcomes, and its quality can be measured with clear accuracy metrics.
- Generative AI produces new content: a draft email, a summary of a 60-page contract, an answer to a policy question, an image, a block of code. It is powered by foundation models, very large models pretrained on broad data that can be adapted to many tasks through prompting. Its outputs are open-ended, harder to score, and can be fluent but wrong (hallucination).
Some business problems need one, some the other, some both. A bank's credit-risk score is predictive ML; the letter explaining a decision to the customer could be drafted by generative AI; the decision itself should stay with the predictive model and human policy. A common strategic error is reaching for generative AI because it is fashionable when the problem is a well-defined prediction with good historical data, where classic ML is cheaper, more accurate, and easier to explain.
⚠️ Exam Trap: If a scenario needs a numeric forecast, a risk score, or a yes/no classification from structured historical data, the best fit is usually traditional ML, not generative AI. If it needs to draft, summarize, converse, or extract meaning from documents, generative AI fits. Answer options that use "AI" and "generative AI" interchangeably are often the distractors.
Reflection Question: A logistics firm wants (a) to predict late deliveries from shipment data and (b) to write proactive apology messages to affected customers. Which circle of the diagram does each need, and why would one solution not serve both well?
