Two Ways to Build a Spam Filter | AI Learning Guide

Think about the last spam email you received. Someone could have written a rule to catch it, but the sender only needs to change one word and the rule fails. That is the weakness of hand-written rules: they only cover situations the programmer already imagined. When the computer learns from labeled examples instead, it is not following anyone's instructions about what makes an email suspicious. It is comparing thousands of examples and finding whatever pattern actually separates the spam from the legitimate mail. The knowledge lives in the data, not in a person's head.

0:00 / 0:00

Reprodução automática

Where the decision logic comes from

Rule-based programming

A person writes explicit conditions the computer must follow.
The computer applies the rules exactly as written.
New spam tricks require new rules written by hand.
The logic is readable and inspectable by a human.

Learning from examples

The computer is given many labeled emails and infers the pattern.
The computer derives its own internal decision logic.
New tricks are absorbed when new labeled examples are added.
The logic may be too complex for a person to state directly.

The practical consequence is about who does the work of capturing knowledge. With rules, a human must anticipate every situation in advance. With learning, the human supplies examples and the computer finds the regularities. Neither approach is universally better: rules are transparent and easy to audit, while learned models adapt to patterns that are too subtle or too numerous to write down by hand.