Abstract

Seventeen years ago, John McCarthy wrote the note Epistemological challenges for connectionism as a response to Paul Smolensky’s paper 'On the proper treatment of connectionism'. I will discuss the extent to which the four key challenges put forward by McCarthy have been solved, and what are the new challenges ahead. I argue that there are fewer epistemological challenges for connectionism, but progress has been slow. Nevertheless, there is now strong indication that neural-symbolic integration can provide effective systems of expressive reasoning and robust learning due to the recent developments in the field.