Expert systems proved that a carefully built rulebook could do useful work inside a narrow field.

But their success depended on clear boundaries. When an important case or factor was missing, people had to add knowledge. Every change needed testing to make sure earlier answers still worked.

The systems had not become useless. They had become expensive to build, extend, and maintain. Ordinary computers were also taking away the advantage of specialized AI machines.

The second winter did not show that rules had no value. It showed how much human work was required to keep a rule-based system useful.

The next chapter asks what parts of human expertise may never fit neatly into a rulebook at all.