Chapter Five - The Revival: 1980s Expert Systems

CHAPTER FIVE

The Revival: 1980s Expert Systems

AI DIDN’T STAY dead for long.

By the 1980s, a new breed of machine intelligence was making headlines again — not because it could think like a human, but because it could act like one. In narrow domains. With strict rules. For companies that were willing to pay.

Enter the Expert System: the duct-taped ghost of symbolic AI, back from the dead and wearing a tie.

This wasn’t thinking. This was faking it — just well enough to sell.

At its core, an expert system is just a giant if-then tree on steroids.

It asks you questions. It matches your answers to a decision tree. It spits out advice. If the problem fits the system’s knowledge base, it works. If it doesn’t, the whole thing collapses.

No learning. No adapting. Just pre-programmed expertise.

And in the ‘80s, that was enough to get tech people paid.

One of the first expert systems, MYCIN, was built in the 1970s to diagnose blood infections. It actually performed surprisingly well — better than junior doctors in some cases — by walking through a series of symptoms and rules crafted by actual physicians.

Then came XCON, built for Digital Equipment Corporation (DEC). It configured computer orders for customers, saving DEC millions in misorders and manufacturing costs.

For a hot minute, it looked like AI was finally useful.

And in a sense, it was. But only within strict cages.

These systems didn’t understand anything. They couldn’t generalize. They broke instantly when problems fell outside their carefully built domain. You couldn’t feed MYCIN a dermatology case. You couldn’t ask XCON to fix a printer.

They were like savants with amnesia. Powerful but dumb.

Still, companies were sold.

From banking to aerospace to logistics, everyone wanted an “AI consultant” who never slept. Big firms started deploying expert systems to diagnose machinery, evaluate loans, detect fraud, and do paperwork that would’ve required a dozen mid-tier analysts.

This wasn’t artificial general intelligence.

This was artificial bureaucracy.

And it kinda worked.

For a while.

Expert systems died for the same reasons AI always stumbles:

1. Brittleness — One edge case and the whole thing broke.
2. No learning — Updating rules required human coders, constantly.
3. Cost — Building and maintaining these beasts was expensive.
4. Maintenance hell — As businesses evolved, the rule sets became impossible to update without breaking everything.
5. Hype fatigue — Once again, AI overpromised and underdelivered.

By the early 1990s, the second AI Winter rolled in.

People were tired of being sold intelligence and getting spreadsheets with opinions.

This cycle made one thing painfully clear:

If AI was ever going to work, it needed to learn.

Not just follow rules. Not just imitate experts. Not just fake logic.

Learn. From data. On its own.

That meant throwing out decades of symbolic logic — and resurrecting something far messier, more organic, and way more powerful.

Something that sounded a lot like a brain.