Since currently the focus on intelligent agents to solve tasks specific to AI is well studied, how important is to study the foundations of the Expert Systems in an Artificial Intelligence undergraduate course? In this course the classic search techniques are analyzed, the most common schemes of knowledge representation and several approaches such as the use of artificial neural networks for various problems are also explained.
It depends how you define expert systems, Watson could be one. I teach about expert systems as the introduction to AI, when I talk about the history. I don't think skipping any "older" systems is a good idea, just look at deep learning - there was a time when neural nets were considered dead and many courses only mentioned them. If you plan to do some exercises, let students think about how to combine old ideas with new methods. Expert systems can be a good basis for introducing stochastic fuzziness into stiffer GOFAI representatives.
Well, how relevant is it to teach calculation by hand in a math class, if there are computers around ;-) ? Of course, it is a question of basic education. Students should understand the basics before higher concepts can be understood.
Coming back to your question. I think it is important for students to learn what are the pros and cons of symbolic and subsymbolic AI. Each of them has difficulties with tasks the other can perform more easily. From my point of view students need to be aware of such limitations. Take for example an application where you just have a small number of example cases which are not sufficient for machine learning. In such a domain clearly a GOFAI approach may help.
Take as another example medical diagnoses, which needs to rely on a huge amount of background knowledge in order to perform differential diagnosis. This background knoweldge is available in symbolic ontologies. Applying it requires knowledge about its representation and logical inference.
Take as another example linked open data or semantic web approaches, besides the knowledge of how to query RDF repositories, some knowledge of classification, symbolic knowledge representation and inferencing is of course needed.
Have you ever seen a deep neural network or ML approach solving a planing tasks?
So foundations of symbolic AI are beneficial, even though GOFAI is not very hip at the moment. But we work on that ...
I agree with Rafals idea of combining both approaches is a path we should explore further.
Thanks, Rafal Rzepka and Thomas Hoppe, for your answers!
It is for me more evident that an IA course should describe the broad vision of all topics. In the IA official program at my school, machine learning is only a sub-topic, but Expert Systems are covered in two of the five thematic units. The others are Introduction of IA, search methods, and recent topics.