An expert systems history



Artificial intelligence (AI) hasn’t always been a lucrative field. But the technology that first made AI commercially successful – the ‘expert system’ – is commonly forgotten.

Before our modern AI-powered tools, expert systems were a notable component of computerised automation. They were the innovative trend; the ‘big thing’ in the tech of their time.

Today, though, the history of expert systems has taken a step away from the limelight. But that doesn’t mean the technology wasn’t a key milestone in the development of AI technology.

A knowledge of expert systems history, then, is crucial to truly understand the history of artificial intelligence (AI). So, let’s dive in.


What are expert systems?

Before exploring the expert systems history, it’s helpful to know what, exactly, expert systems are.

Expert systems are the first example of ‘knowledge-based systems’. They work using rules and rely on two components: a knowledge base and an inference engine.

Let’s break those terms down:

  • Knowledge base: An organised collection of facts about the world and/or the task the system is designed to carry out.
  • Inference engine: Applies logical rules to the knowledge base to deduce new information.

In the simplest terms, expert systems are artificial intelligence systems that specialise in one task. In other words, they’re computerised systems that act as experts in one given field.


Before it all

The expert systems history starts almost alongside the dawn of the modern computer in the 1940s, when the first digital programmable computers began to emerge.

It wasn’t long before researchers started to think about the potential of these new machines. What if they could emulate human decision-making? What if they could “think” as humans do?

And so it was that researchers started looking into artificial intelligence — and began on the path to creating expert systems.


The first expert systems

Officially, the expert systems history starts in 1965. This is when the technology saw its formal introduction by the Stanford Heuristic Programming Project. Edward Feigenbaum – the ‘father of expert systems’ – led the inaugral project.

Edward Feigenbaum was involved with both MYCIN and Dendral — two separate early expert systems.

Dendral was an expert system that specialised in analysing and identifying chemical compounds. It’s widely considered the first expert system.

MYCIN was derived from Dendral. It was another expert system — one that focused on identifying bacteria that caused infections and recommending antibiotics.

These systems didn’t try to be general intelligence. They weren’t general problem solvers. Rather, they focused on a limited (but in-depth) foundation of knowledge. And this made them one of the first successful attempts at AI software. That is, machines that appeared to analyse and ‘think’.


SUMEX, a computer designed to encourage the application of artificial intelligence in medicine. Public Domain Via National Library of Medicine 

Reaching the heyday

The heyday of expert systems came in the 80s. During this time, two-thirds of Fortune 500 companies used expert systems.

Interest in expert systems was international. They saw increased research funding in Europe, and the Fifth Generation Computer Systems Project in Japan, which saw researchers focus (in part) on inference technology and knowledge bases.

A Symbolics Lisp Machine: an early platform for expert systems. Source

But expert systems were not without their problems. There were difficulties managing and maintaining the knowledge base. There were difficulties writing the rules that reflected the knowledge of experts. And the hype around expert systems was spiralling faster than the technology could keep up.

To paraphrase a common idiom, hype comes before a fall. And this was true for expert systems. The AI winter was coming.


Fading to obscurity?

In the 1990s and onward, the expert systems history involves the decline in the popularity and hype of the technology. As the tech world saw an AI winter, the excitement around expert systems faded.

The apparent decline of expert systems at this time has two reasons behind it.

The first, simply, is that expert systems failed to live up to the hype. They couldn’t perform the over-egged functionality that had been promised. They didn’t expand to a more general form of AI fast enough, and so they were discarded.

The other explanation is that they were absorbed by other technology tools. As expert systems became better known, programmers and developers could use the technology behind them as part of other offerings.

In short, rule-based systems became useful for more than expert systems, and so the standalone expert system stepped out of the spotlight.  


The present and future

You probably won’t hear much mention of expert systems these days. Indeed, it would seem that there are very few in use.

However, the basic tools and premises that stem from the expert systems of the past are present in modern software.

For instance, think of the rules-based systems found in automation tools. Or, consider the understanding of the need for data and knowledge in machine learning and other modern AI-powered tools. Consider the different types of database for different types and formats of data and knowledge.

The advancements in these technologies all have roots in expert systems.


An expert systems history

Expert systems are a key player in the history of automation and AI. While they’re not in the spotlight today, there was once a time where they were the height of artificial intelligence.


Useful links


Download