Rule-based AI

We like to imagine artificial intelligence (AI) as mysterious. In the beginning, though, it was just rules. Lots of them. If X happens, then do Y. That’s it.

IF THEN

Picture a help desk script. Caller says “forgot password.” The system checks: If password reset, then send link. If not, move to the next step. That’s how early systems worked. The fancy name was “expert system.” Really, it was a big pile of IF–THEN rules.

Inference engines

She needed a way to decide which rules to fire first. That’s where the inference engine came in. Think of it like a traffic cop at an intersection. She looked at the facts on the table, picked the next rule, and sent things along. Sometimes she worked forward (if we know A, then infer B). Sometimes backward (if we want B, then ask what A must be). Either way, she kept the rules moving.

Where they worked

Hospitals tried them first. If a patient had symptom A and symptom B, then maybe it was disease C. Banks tested them on loan approvals. Factories used them for troubleshooting machines. She wasn’t magic. Just diligent. If we had the rules, she followed them without complaint.

What we learned

Rule-based systems didn’t scale well. Real life is messy; rules multiply. Pretty soon we had thousands of IF–THEN lines that contradicted one another. She couldn’t keep up. That’s why we turned to statistical models later.

A coder’s note

We admire the persistence. She may not have been smart, but she showed us how far stubborn rules could go.