Recursive self-improvement

Seed AI (English for "KI - seed ( grain ) " ) is a technology developed by Eliezer Yudkowsky theory of a self-learning artificial intelligence (AI), which is improved by recursion itself and expanded. The AI ​​must be able to adapt its own program code as and to improve, that the next generation can make further improvements, which would have been impossible, according to the previous versions of the theory. The "seed grain " so to speak, sown by the first generation, the subsequent reaping the rewards for these improve again and new implant. This cycle is continued until eventually reaching the human intelligence and surpassed in the next cycle.

A highly simplified version of this intelligence make today's programming compiler dar. These optimize its own program code in terms of efficiency and speed, but extends this kind of improvement is by far not sufficient to realize the unlimited recursive self-improvement a seed AI. Although the previous compiler examine their own code and improve it, but this only once and not recursive. Although The resulting compiler work faster, but not better than the predecessor. Here ends the previous development, the new version never discovered new optimization methods or can improve their own program code. A proper Seed AI needs to understand the purpose and the design of your own program code to create another generation with improved intelligence.

This self-optimizing artificial intelligence currently does not exist yet, but some research groups have to the goal of the creation of Seed AI. The Singularity Institute for Artificial Intelligence or Artificial General Intelligence Research Institute are the best known among them.

In futurology, the view is taken that seed AI can lead to a state in which the technical development is sufficiently rapid that its a average intelligent person can no longer follow.