The concept of superintelligence embodies an AI that can outperform the best human brains in every field, including scientific creativity, general wisdom, and social skills. This level of intelligence involves machines that are not only adept at understanding and processing vast amounts of information far beyond human capabilities but also capable of self-improvement and recursive learning, leading to an intelligence explosion. The implications of achieving superintelligence are profound, raising questions about ethics, societal impacts, control, and safety. The transition from AGI (Artificial General Intelligence) to superintelligence is considered a critical point, often referred to as the "singularity," where predicting the behavior and outcomes of such AI becomes increasingly difficult, if not impossible.
The term "superintelligence" and its conceptual foundations have been discussed in philosophical and technical AI contexts since at least the late 20th century, gaining prominence in the early 21st century with the publication of works such as Nick Bostrom's "Superintelligence: Paths, Dangers, Strategies" (2014). The concept has roots in earlier discussions about the potential for machines to one day exceed human intelligence, a topic of speculation and analysis since the inception of computational theory and artificial intelligence research.
Nick Bostrom, a philosopher at the University of Oxford, is one of the most prominent figures associated with the discussion of superintelligence, through his detailed analysis of its potential paths, dangers, and strategies for control. Other significant contributors to the theoretical framework and ethical considerations of superintelligence include Eliezer Yudkowsky, Ray Kurzweil, and many others who have explored the implications of advanced AI systems capable of self-improvement and exceeding human intelligence.