Towards learning 2.0
Etienne Barnard, Brigitte Palensky, Peter Pálenský, Dietmar Bruckner · 2009
Learning certainly qualifies as one of the core issues of artificial intelligence (AI). During the years, it has gained - and subsequently lost - popularity in the research community. After a historical perspective on the rise and fall of learning research in AI, some of the limitations of current learning systems are reviewed, followed by a presentation of various responses about how to overcome them. A special focus is given on one of the responses, the attempt to draw lessons from a detailed study of evolutionary and developmental processes and stages of learning in nature, in particular in human beings. From this, a number of principles for machine learning are inferred. A key aspect seems to be that learning should be cumulative to compensate for the exponential growth in learning complexity.