Logic and Induction: an Old Debate

For more than 2500 years, philosophers tried to identify induction as the way to derive general knowledge from individual observations of the outside world. Some of those philosophers wanted to formalize induction with the help of a few elementary inferences which appear to be similar to some of the elementary steps involved today in machine learning programs. Over and above tracking the foundations of induction, there are basically three reasons for studying such similarities. The first involves rereading the classics in order to seek inspiration from the ancient philosophers, to compare the procedures both ancient and new, to simulate them and thus test their limits. The second is related to the contemporary development of a new approach to machine learning, namely Inductive Logic Programming which explicitly states that induction is fundamental to machine learning. The third and last reason concerns the limitations inherent to machine learning and induction which were pointed out by some philosophers. Does this philosophical indictment of induction doom the automation of induction through machine learning to failure?