Steps toward robust artificial intelligence

Abstract

Let me begin by acknowledging the recent death of Marvin Minsky. Professor Minsky was, of course, one of the four authors of the original Dartmouth Summer School proposal to develop artificial intelligence (McCarthy et al. 1955). In addition to his many contributions to the intellectual foundations of artificial intelligence, I remember him most for his iconoclastic and playful attitude to research ideas. No established idea could long withstand his critical assaults, and up to his death, he continually urged us all to be more ambitious, to think more deeply, and to keep our eyes focused on the fundamental questions. In 1959, Minsky wrote an influential essay titled Steps Toward Artificial Intelligence (Minsky 1961), in which he summarized the state of AI research and sketched a path forward. In his honor, I have extended his title to incorporate the topic that I want to discuss today: How can we make artificial intelligence systems that are robust in the face of lack of knowledge about the world?.