As an alternative to piling generalities on generalities, not that there’s anything wrong with that, it also helps to look at issues as they arise in concrete applications.
One of the most concrete applications I ever attempted was the program I worked on all through the 80s that sought to integrate a basic form of inductive (data-driven) learning with a fundamental form of deductive (concept-driven) reasoning. Having recently begun a fresh attempt to essay all that on my blog I think it might serve our ends to share that here.