To generalize is to be an idiot - or a machine:
Machine learning and the problem of induction
Machine learning is concerned with algorithmic generalisation from a given set of seen examples to unseen examples. Obviously, this can only be done by tackling the problem of induction in some way. Nowadays, researchers in machine learning, who naturally have a more practical attitude towards induction, become more aware of the underlying philosophical problems. On the other hand, philosophers start to become interested in machine learning as some sort of new and/or experimental methodology of science. However, there still seems to be too little mutual knowledge, and "interdisciplinary" research is often not much more than an enumeration of superficial similarities. This talk would like to spark interest in digging deeper into the foundations of machine learning and theoretical computer science in general in order to stimulate fruitful cooperation between these fields and the philosophy of science. The questions raised (and hopefully at least partially answered) are: How does machine learning deal with the problem of induction? What are the common interests of machine learning and the philosophy of science? Finally, is there anything philosophy may learn from the approaches taken by machine learning?