Is the expert system still in use today?

Question

In my country, the Expert System class is mandatory, if you want to take the AI specialization in most universities. In class, I learned how to make a rule-based system, forward chaining, backward chaining, Prolog, etc.

However, I have read somewhere on the web that expert systems are no longer used.

Is that true? If yes, why? If not, where are they being used? With the rise of machine learning, they may not be as used as before, but is there any industry or company that still uses them today?

Please, provide some references to support your claims.

Answer 1

I would say Expert Systems is still being taught. For instance, if you look at some of the open courses like MIT's, there are still lectures on it.

Also, looking at the CLIPS documentation, you will find a couple of examples of usage from 2005.

What I suspect is that Expert Systems are now embedded with "normal systems" in practice. Hence it may be difficult to distinguish from systems used on a daily basis for diagnostics, etc. and not as popular as before.

Answer 2

Are there companies that still use expert systems?

There are still some expert system inference engines available in open source form, in particular CLIPS rules

A specialization of your question could be: what companies are using CLIPS in 2020 ?

I don't have any ideas, even if I did try in https://github.com/bstarynk/clips-rules-gcc

And the RefPerSys project is right now (in November 2020) discussing the idea of incorporating such rules in it.

Read of course quickly Jacques Pitrat's blog on http://bootstrappingartificialintelligence.fr/WordPress3/ and his last book (describing the design of an ambitious symbolic artificial intelligence system -CAIA- with expert system ideas) Artificial Beings The Conscience of a Conscious Machine (ISBN 9 781848211018) - his CAIA system is on https://github.com/bstarynk/caia-pitrat but there is absolutely no documentation, since Jacques Pitrat passed away in October 2019. His CAIA system was capable of generating all the 500KLOC of its C code from some kind of expert system rules (whose design is described in Jacques Pitrat's books and papers).

I am not a native English speaker (since I am French) but I heard that expert systems are called (in 2020) business rule management systems.

I heard that major banks in France (maybe BNP or Société Générale) are using such systems to decide to give some loan or some credit to persons and companies (in particular for people buying their flat - or a brand new automobile - with a bank credit and debt during dozen of years).

The French banking system is very opaque: you won't be able to understand their internal software, and banks are not publishing any document about the design of their software. At most they would publish the name of their AI systems, but nothing public about the software design.

According to rumors, Lexifi or Yseop might use some kind of very proprietary expert system technology and sell services with them. But their software tools are closed source and very proprietary.

Regarding expert system for games, see also recent papers by Professor Tristan Cazenave. He did use some kind of expert system technology for games.

My guess is that large Internet companies like Google or Amazon are using expert system technology inside their internal software (e.g. search engines). IBM Watson is rumored to use them also.

BTW, GNU make might be considered as some very crude expert system engine driving building of software artifact from source code.

Answer 3

Yes, of course.

There is a lot of interest in externalizing domain knowledge from execution framework, but the current software engineering paradigms (well, Agile) soft of stands that approach on its head, as does having more and more domain experts who grew up assuming that some facility with software engineering was expected. E.g. medical informatics graduate programs all require at least a semester of intro to comp sci as part of the prereqs (I don't know about clinical informatics fellowships for physicians, but will check and update--but given medical school pre-clinical years is already in the 32-40 semester-hour course load fitting in anything is difficult.

In the biomedical domain, one very active project is https://www.opencds.org/ which is based on the open source Drools business rules engine and business rules management framework.

It is a great example of where it is bananas (technical neuroscience term) to try and put even a few thousand rules into even Prolog without having some sort of framework to detect rules collisions (e.g. one rule may day to give a diuretic to make people urine more and remove sodium and water from their body and another might say to give IV sodium chloride (thus adding sodium and water). It is also, maybe, a little easier for medical informaticists to pick up using Drools than Java (except that a lot of have taken at least one course taught using Java, and most of us have had some exposure to Prolog and CLIPS as part of clinical decision support/AI courses).

I think that a lot of stock trading is done using expert systems that are primarily rules-based expert systems. But, medical informatics was one of the first domains to use AI (I am thinking of early experience by IBM in the 1950s, and by the 1970s AI systems were in production in many major academic medical centers, by 1980s the US DoD Military Healthcare System and US Department of Veterans Affairs both had large enterprise-scale electronic health record systems which were created in part to facilitate delivery of decision support. Similar progress was made in the UK NHS and academic teaching hospitals around the same time.

https://confluence.hl7.org/display/CDS/WorkGroup+Home https://www.drools.org/ https://en.wikipedia.org/wiki/Drools https://en.wikipedia.org/wiki/Clinical_decision_support_system