Yesterday, we learned:

Historical data + correct outcomes → training → model

Today’s main lesson is:

A model can only learn from the data that we provide.

In normal software, incorrect data may cause a validation error or incorrect report. In machine learning, incorrect data can become part of the system’s learned behaviour.

This is commonly expressed as:

Garbage In → Garbage Out

But AI systems create an additional danger:

Biased or misleading data
            ↓
Model learns misleading patterns
            ↓
Predictions appear intelligent
            ↓
Incorrect decisions are repeated at scale

Data preparation is therefore not a minor preprocessing activity. It is one of the central responsibilities of an AI engineer.