Superintelligence refers to a hypothetical artificial intelligence that would vastly surpass human cognitive performance in virtually all tasks: scientific reasoning, creativity, strategic planning, or social skills. The concept was popularized by philosopher Nick Bostrom in his book Superintelligence (2014), and it differs from artificial general intelligence (AGI) due to its vastly superior nature, not merely being comparable to humans.
Its relevance is primarily prospective and is linked to the debate on existential risks. Those who study it warn of challenges such as:
- The control problem: ensuring that a system more capable than ourselves pursues safe goals aligned with human values.
- The intelligence explosion: the idea that an AI capable of improving itself accelerates its own progress uncontrollably.
It is worth noting that, currently, superintelligence is a theoretical scenario. Current systems, including large language models, are far from that general and autonomous capability, so the term belongs more to the field of foresight and ethics than to existing technology.