A Seed AI is an Artificial General Intelligence (AGI) which improves itself by recursively rewriting its own source code without human intervention. Initially this program will likely have a minimal intelligence, but over the course of many iterations it will evolve to human-equivalent or even trans-human reasoning. The key for successful AI takeoff would lie in creating adequate starting conditions.
By contrast, most current approaches to AGI attempt to succeed by creating a mind immediately capable of human-equivalent intelligence. Two of the more popular efforts are IBM's Watson and Cycorp's Cyc. Both of these programs are improved through human manipulation of the source code rather than by the AGI itself.
One critical consideration in Seed AI is that its goal system must remain stable under modifications. The architecture must be proven to faithfully preserve its utility function while becoming more intelligent. If the first iteration of the Seed AI has a friendly goal, and is sufficiently able to make predictions, then it will remain safe indefinitely; if it predicted that modifying would change its goal, it would not want that according to its current goal, and it would not self-modify....