Five years ago Paul Haley expressed his frustrations with production rule technology (to which he was one of the major contributors) because of its inability to perform deduction. He wrote that rule vendors should stop “dressing up a procedural language!” – see his Confessions of a production rule vendor Part 1. Yesterday Paul published Confessions of a production rule vendor Part 2. The bottom line of his article:
It’s time to trade rule technology dating back to the 80’s for state of the art AI.
He shows an interesting timeline of rules and AI for the last 30+ years and emphasizes the following:
- 80’s technology dominates business rules, policy automation, and decision management
- artificial intelligence (AI) and natural language processing (NLP) have improved dramatically
- logical reasoning technology has advanced while production rule technology remains stagnant.
We expect to see more discussions about integration of Business Rules and Semantic Web technologies this September in Luxembourg where DecisionCAMP and RuleML+RR will be co-located during Logic for AI Summit.
AI is essentially about identifying a high intensity of consistent correlations between data describing a context and specific, useful results.
These high intensity and consistent correlations are a) not common, b) hard to find, and c) might be there one day and gone the next. Also, it is hard to include all relevant information in this search for correlations, especially negative correlations.
Example: My AI tool says that I have a high propensity to buy X given what is in my shopping cart right now. All good, until I factor in that I bought X yesterday.
Or, my AI tool says that Fred’s insurance claim could be fraudulent. All good until I factor in that Fred has been a no claims customer for 30 years and the claim is only $500, so I will pay regardless.
So, in my view, AI is about finding a propensity leading to ‘suggested’ actions, not committing to definitive actions. This propensity is just another fact to consider in a rules based logic engine before determining definitive action.
Ergo, these two technologies are highly complementary with minimal overlap, so there are 2 winners in this debate.