Introduction: Cognitive autonomy in machine discovery

This special issue of Machine Learning on discovery raises the question: How is discovery different from learning? In machine learning we typically distinguish between (1) learning as acquiring new knowledge in the form of concepts, taxonomies, regularities, and the like, and (2) learning as performance improvement and skill acquisition. "Discovery" applies to things that exist, such as the moons of Jupiter or the laws of nature, so in this section we confront discovery with learning in the first sense, of acquisition of objective knowledge. In the second section we discuss the distinction between discovery and invention. The third section reviews the ways in which the articles in this issue contribute to the growing autonomy and integration of machine discoverers, and it mentions recent research in machine discovery not represented in this issue.