Superintelligence: Definition and Examples

Superintelligence is a theoretical concept that describes an artificial intelligence whose capabilities would surpass those of the brightest human mind in virtually every domain: scientific reasoning, artistic creativity, social skills, strategic decision-making, and even self-awareness. Unlike the narrow AI we use today, a superintelligence would not be limited to a specific task.

The concept was popularized by philosopher Nick Bostrom in his book "Superintelligence: Paths, Dangers, Strategies" published in 2014. Bostrom distinguishes several possible forms: speed superintelligence (thinks like a human but infinitely faster), collective superintelligence (a network of intelligences forming a superior whole), and qualitative superintelligence (possessing fundamentally superior modes of thought to ours).

Today, large language models like GPT-4 or Claude represent what is called generalist AI (weak AGI), capable of excelling in many tasks without reaching the level of a superintelligence. The debate revolves around the timeline: some researchers estimate that a superintelligence could emerge within a few decades, while others consider this perspective very distant or even impossible.

The question of superintelligence is central to the field of AI safety. Researchers work on the alignment problem: how to ensure that a superhuman intelligence would act in accordance with human values and interests? This challenge is considered by many to be one of the most important problems of our time.