What AI and machine learning mean

Three terms, one family

• Artificial intelligence (AI) is the broad goal of making computers perform tasks that normally need human intelligence, such as understanding language, recognising images or planning.
• Machine learning (ML) is the most successful approach to AI today. Instead of programmers writing every rule, the system learns patterns from data.
• Deep learning is a kind of machine learning that uses large neural networks with many layers. It powers image recognition, speech recognition and large language models.

Rules versus learning

Suppose you want to flag spam emails. A traditional program needs hand-written rules:

def is_spam(subject):
    return "free money" in subject.lower() or "winner" in subject.lower()

Spammers quickly change their wording and the rules break. A machine learning approach shows the computer thousands of emails already labelled "spam" or "not spam" and lets it discover which patterns matter.

Kinds of machine learning

Supervised learning is the most common. It solves two kinds of problem: classification (choosing a category, such as spam or not spam) and regression (predicting a number, such as a price).

How a model learns: a tiny example

Imagine data on hours studied and exam scores:

A model is a mathematical function with adjustable settings called parameters. Here a straight line fits well: score = slope × hours + intercept. Training means finding the parameters that make the line's predictions closest to the real scores.

With the popular scikit-learn library (pip install scikit-learn), that takes a few lines:

from sklearn.linear_model import LinearRegression

hours = [[1], [2], [3], [4], [5]]   # inputs (features)
scores = [52, 58, 65, 71, 78]        # labels

model = LinearRegression()
model.fit(hours, scores)             # training

print(round(model.coef_[0], 2), round(model.intercept_, 2))   # 6.5 45.3
print(round(model.predict([[6]])[0], 1))                      # 84.3

The model learned that each extra hour adds about 6.5 marks, and it predicts 84.3 for someone who studies 6 hours.

Training and testing

A model can perform well on the data it learned from yet fail on new data, a problem called overfitting. To get an honest measure, data scientists split the data: most of it for training, and a held-out portion for testing. Only performance on the test data tells you how the model is likely to behave in the real world.