Data Science Foundations Chapter 14: Machine Learning and AI Explained Simply

Everyone has an opinion about AI. Your coworker worries robots will take his job. Your cousin swears ChatGPT wrote his college essay. Chapter 14 of Data Science Foundations by Stephen Mariadas and Ian Huke explains what machine learning and artificial intelligence actually are. And how data science connects to all of it.

What Is AI, Really?

Here’s the thing. Nobody fully agrees on a definition. NASA keeps it simple: AI is computer systems doing complex tasks that normally need human reasoning, decision making, or creativity. The UK and EU definitions add two key qualities: adaptiveness and autonomy. The system adjusts on its own. It does not just follow a fixed script.

Machine Learning Is How AI Learns

Machine learning is software that gets better at a task the more it does that task. You do not program every rule by hand. Instead you give it data and let it figure out patterns.

The authors use a pothole example. Say you want software that spots potholes in road photos. Writing manual rules would be a nightmare. Every road looks different. Shadows and puddles exist. Instead, you show the model thousands of road images, some with potholes, some without. It learns the patterns. Then you give it a new photo and it predicts. That is machine learning.

And here is where data science fits in. The algorithms behind ML are built on algebra, statistics, and probability. Data science methods build ML. ML solves data science problems and creates AI. They all feed into each other.

Supervised vs. Unsupervised Learning

Two main approaches to training.

Supervised learning means you give the model labelled data. Every pothole photo is tagged “pothole.” Every clean road is tagged “no pothole.” The model learns which patterns match which labels. More data, better predictions. Great when you know what you are looking for.

Unsupervised learning means the data has no labels. The model looks for patterns on its own and groups similar things together. But it might cluster by time of day or sidewalk presence instead of potholes. It finds patterns, but not always the ones you wanted.

From my experience, the real power comes when you combine both. Unsupervised to discover the unexpected. Supervised to get precise answers.

Neural Networks: Imitating Your Brain

Neural networks try to copy how brains work. You have layers of nodes, similar to neurons. Input layer takes in data. Hidden layers process it. Output layer gives a result.

Each connection has a weight. Data flows through, gets multiplied by weights, adjusted by biases at each node, and passes forward. During training, the model tweaks weights and biases until the output is accurate enough. This is called back propagation.

Stack many layers and you get “deep learning.” This powers generative AI. I remember when neural networks were academic curiosities. Now they run half the internet.

Narrow AI, General AI, and Generative AI

The chapter makes important distinctions here.

Narrow AI does one specific thing well. Your pothole detector is narrow AI. It cannot book you a flight. Most AI today is narrow.

General AI would handle everything a human can. It does not exist yet.

Generative AI sounds similar to general AI but is different. It generates new content: text, images, code, music. Large language models are one type. Still narrow in many ways, but applicable to many language tasks.

Foundation Models

Foundation models are generative AI trained on massive, diverse datasets. Because they are pre-trained on so much data, you can adapt them to different tasks without starting from scratch. This makes building new applications much faster and cheaper.

They work by predicting the next item in a sequence. Next word, next token, next pixel. One prediction after another builds up into something that looks meaningful. Simple idea. Powerful result.

The Risks Nobody Should Ignore

The authors list four main risks. Bias: the AI produces skewed outputs because training data was skewed. Hallucination: the AI confidently makes things up. Transparency: you cannot explain how it reached its answer. Privacy: your data gets used in ways you did not expect.

I have seen all four in real projects. Bias is the sneakiest. You might not notice until the damage is done.

The chapter also covers regulation. The UK and USA take a pro-innovation approach. The EU went further with the AI Act, putting controls based on risk levels. Different philosophies, same concern: benefits without the harm.

My Take

This chapter connects everything from earlier in the book to the AI world. Statistics, probability, classification, clustering. You already know the building blocks of machine learning if you followed along.

The practical tip I liked most: before building something complex and expensive, ask if a simpler solution would work. I have seen too many teams reach for deep learning when a basic regression would have been fine.

Good chapter. Short and honest about both the potential and the risks.

This post is part of a chapter-by-chapter retelling of “Data Science Foundations: Navigating Digital Insight” by Stephen Mariadas and Ian Huke. My thoughts and interpretations, not a copy of the book.

Previous: Chapter 13: Communication Next: Chapter 15: Case Studies