The Expert Systems Era

After a collapse, it's natural to narrow your ambitions.

General intelligence had proved elusive. So researchers tried something more constrained: instead of teaching machines to think, they would give machines knowledge. Specific, careful, expert knowledge, encoded into rules that could be applied to real problems.

It worked. Within the right boundaries, these systems performed impressively. They solved problems that previously required years of training to handle. They found their way into hospitals, factories, and boardrooms. For a time, they were AI's most visible success.

But a system built on explicit rules has a particular kind of fragility. It knows what it knows, and nothing tells it where that ends. Step outside the boundaries, and it doesn't hesitate. It just gets things wrong, with the same confidence it brings to everything else.

Maintaining those boundaries turned out to be its own endless problem. The world kept changing. The rules couldn't keep up.

This chapter is about what happens when a solution that works beautifully in the right conditions meets the conditions it wasn't built for.