I think the thing I find the most surprising about Expert Systems is that people expected them to work so early on, and apparently they did work in some circumstances. Some issues:
The user interfaces, from what I can tell, were often exceedingly mediocre. User interfaces are difficult to do well and difficult to specify, so are hard to guarantee quality in large and expensive projects. It was also significantly harder to make good UIs back when expert systems were more popular, than it is today.
From what I can tell, many didn’t even have notions of uncertainty! AI: A Modern Approach discusses Expert Systems that I believe used first and second-order logic, but seemed to imply that many didn’t include simple uncertainty parameters, let alone probability distributions of any kind.
Experts aren’t even that great at assigning probability densities. Many are overconfident; papers by Tetlock and others suggest that groups of forecasters are hard to beat.
My impression is that arguably Wikidata and other semantic knowledge graphs could be viewed as the database part of expert systems without attempting intense logic manipulations or inference. I know some other projects are trying to do more of the inference portions, but seem more used for data gathered from web applications and businesses instead of directly by querying experts.
I think the thing I find the most surprising about Expert Systems is that people expected them to work so early on, and apparently they did work in some circumstances. Some issues:
The user interfaces, from what I can tell, were often exceedingly mediocre. User interfaces are difficult to do well and difficult to specify, so are hard to guarantee quality in large and expensive projects. It was also significantly harder to make good UIs back when expert systems were more popular, than it is today.
From what I can tell, many didn’t even have notions of uncertainty! AI: A Modern Approach discusses Expert Systems that I believe used first and second-order logic, but seemed to imply that many didn’t include simple uncertainty parameters, let alone probability distributions of any kind.
Experts aren’t even that great at assigning probability densities. Many are overconfident; papers by Tetlock and others suggest that groups of forecasters are hard to beat.
My impression is that arguably Wikidata and other semantic knowledge graphs could be viewed as the database part of expert systems without attempting intense logic manipulations or inference. I know some other projects are trying to do more of the inference portions, but seem more used for data gathered from web applications and businesses instead of directly by querying experts.