Working Up

by Dwayne Phillips

Expert systems live on after 40 years. Well, sort of.

I worked in artificial intelligence research in the mid-1980s. We had an AI boom in that decade. Then a winter and a boom or two since then.

Back in 1985, we had “expert systems.” (I am old enough to remember this. I was there.) Have one of your human experts talk to an AI researcher about their expertise. The AI researcher would write software that captured the expertise. That software would run 24/7, never become tired, never take a day off, and never make mistakes from fatigue.

Believe it or not, the idea worked. Some businesses actually used the idea to their great benefit. Most folks bought into the marketing hype, were disappointed in reality (reality always seems to be less than market hype), and gave up. Hence, an AI winter.

Today? Here we go again. Sure, companies are hiring people in the developing world (Is that what we call it today? I am not sure.) for a dollar a day to say, “Yes, that is right. No, that is wrong,” when shown pictures of cats, dogs, giraffes, etc.

Not reported as much is the trend in hiring actual experts and paying them real salaries to check the answers from systems that only actual experts could check. Is that a fracture or a bad x-ray? Is that a correct proof to a theorem? Is that argument logical?

Experts training AI systems. Hmm, sounds like expert-based systems. Sounds like expert systems.

We’ve come a long ways in 40 years. Right? Have we? Maybe.