The Rise, Fall, and Weird Afterlife of a Would-Be AI God

In 2011, a computer named Watson went on Jeopardy! — the famously tricky American quiz show — and won. It didn’t just win. It obliterated two of the show’s greatest human champions, including Ken Jennings, who joked afterward:

“I, for one, welcome our new computer overlords.”

At the time, it felt like a turning point. This wasn’t just another AI gimmick. IBM promised Watson was the future: it would transform medicine, law, business, and everything in between.

And then… nothing.

Today, Watson is a ghost. No one talks about it. While OpenAI, Google, and Anthropic are defining the modern AI landscape, IBM is barely part of the conversation.

So what happened? Let’s rewind.

Deep Blue: The Prequel

Before Watson, there was Deep Blue — IBM’s first brush with global AI fame.

In 1997, Deep Blue defeated the reigning world chess champion, Garry Kasparov, in a match watched around the world. At the time, this felt like AI’s coming-of-age moment. A machine had beaten a genius.

But Deep Blue wasn’t intelligent in any real sense. It was a brute-force symbolic system. It evaluated 200 million chess positions per second, guided by handcrafted evaluation functions and a team of human chess experts. It didn’t “learn.” It didn’t generalise. It was powerful, but narrow.

Kasparov was furious. He even accused IBM of cheating — of slipping human moves into the machine’s playbook. Years later, he admitted he was wrong, calling Deep Blue “intelligent the way an alarm clock is intelligent.”

Still, the win made headlines — and it set IBM up as the company leading the charge into AI.

Jeopardy: Watson’s Debut

Fast forward to 2011. IBM unveiled Watson, a new kind of machine — one that could understand language. It was designed to play Jeopardy!, a long-running American game show known for its quirky format:

• Players are given clues in the form of answers.

• They must respond with questions.

“This U.S. state’s name means ‘great river’ in Algonquian.”

“What is Mississippi?”

It’s fast, nuanced, full of puns and obscure references. And Watson crushed it.

The world was amazed. If a computer could understand questions, parse language, and beat the best humans at their own game — what couldn’t it do?

IBM wasted no time. Watson wasn’t just a game show gimmick — it was marketed as the future of intelligent work. A universal AI assistant for doctors, lawyers, CEOs, and researchers.

The Big Promise: Watson for Healthcare

The flagship project was Watson for Oncology. The idea was dazzling: feed Watson every medical paper, every case study, every treatment guide. Then let it assist doctors by recommending personalised treatment plans based on the latest research.

It sounded like science fiction. But behind the curtain?

Watson didn’t learn. It wasn’t ingesting global cancer data. It was mostly regurgitating advice from a narrow set of doctors at a partner hospital (Memorial Sloan Kettering). And in some cases, it gave dangerously flawed recommendations.

In internal IBM documents, even staff were worried about the system’s reliability.

Eventually, hospitals pulled back. Projects were shelved. And in 2022, Watson Health was quietly sold off.

The dream had died. But the world had already moved on.

Meanwhile, AI Was Changing

While IBM was busy rebranding everything as “Watson” — Watson Analytics, Watson Assistant, Watson for Advertising — the real AI revolution was happening elsewhere.

Neural networks were getting better. Then came deep learning, and in 2017, the transformer architecture — a Google invention that changed everything.

Transformers power modern AI. They made it possible for language models like GPT-3, GPT-4, Claude, and Bard to emerge. These models could:
• Write code
• Translate language
• Summarise research
• Pass law exams
• Simulate human conversation

They could scale across domains instantly, without handcrafted rule sets.
Watson, by contrast, needed a team of consultants to babysit it into every new application.

By the time OpenAI released ChatGPT in 2022, IBM’s once-iconic Watson had become… irrelevant.

What Is Symbolic AI, Anyway?

o understand Watson’s failure, you have to understand symbolic AI — also known as GOFAI (Good Old-Fashioned Artificial Intelligence).

Symbolic AI works by defining knowledge explicitly. You write rules. You create ontologies. You tell the system:

“If an animal has fur and purrs and chases mice, it might be a cat.”

It’s logical. It’s human-readable. And in the early days, it was the only game in town.

It worked well in narrow, structured domains — like medical diagnosis or chess. But symbolic AI doesn’t scale. It breaks under ambiguity. It doesn’t handle messy data well. And it can’t learn on its own.

Watson was basically symbolic AI with a bit of machine learning sprinkled on top. And that just wasn’t enough.

Here’s the Twist: Symbolic AI Isn’t Dead — It’s Being Rebuilt

Ironically, the very LLMs that killed symbolic AI… are now rebuilding it.

Large language models (LLMs) like GPT-4 can now write symbolic systems:
• Expert systems
• Rule-based engines
• Finite state machines
• Logic-based tutoring systems
• Compliance tools that turn law into code

You can literally ask ChatGPT to build you a rule-based advisor for GDPR compliance, or to explain a legal framework as a Prolog program.

We’re entering a hybrid era:
• Use LLMs for creative, messy language tasks
• Use symbolic systems for reasoning, auditing, and logic

The brute force meets the rulebook.
The black box generates the white box.

The Real Legacy of Watson

It had the Jeopardy! win. It had a media halo. It could’ve led the AI revolution.

Instead, it became a case study in overpromising, underdelivering, and failing to pivot.

IBM built Watson to win a quiz show.
But it wasn’t built to survive the future.

Watson won Jeopardy!
But it never made it to Final Jeopardy.

Thought for the day: What does the combination of LLM and GOFAI look like for the future of your industry?