The funny thing is that efficient RETE engines only became widely available towards the end of "the golden age of AI" so that most of the seminal works of that period do not even use them.
I think though the intuition that practical knowledge has an if-then-else structure was right and I think people talk about "explainable AI" too much and not enough about "I just want to tell it specificially what to do"
It looks a lot like the kind of systems I envisioned 10 years ago there would have been some kind of rules and knowledge-graph system that could be lined up with classifiers and scorers and fact extractors and other machine learning models and there would be a system to collect training and test data, handle exceptions, etc. Think Palantir but for operations instead of analytics and a much bigger TAM.
its hard to say right now from the outset till they release something concrete about there architecture . Feels more like state + question/options → learned probabilistic judgment → decision.
Somewhat.
The funny thing is that efficient RETE engines only became widely available towards the end of "the golden age of AI" so that most of the seminal works of that period do not even use them.
I think though the intuition that practical knowledge has an if-then-else structure was right and I think people talk about "explainable AI" too much and not enough about "I just want to tell it specificially what to do"
It looks a lot like the kind of systems I envisioned 10 years ago there would have been some kind of rules and knowledge-graph system that could be lined up with classifiers and scorers and fact extractors and other machine learning models and there would be a system to collect training and test data, handle exceptions, etc. Think Palantir but for operations instead of analytics and a much bigger TAM.
its hard to say right now from the outset till they release something concrete about there architecture . Feels more like state + question/options → learned probabilistic judgment → decision.