"General Intelligence" refers to AI systems seriously intended
to match the fundamental capabilities of human intelligence;
systems with architectures large enough to compare
to a human brain composed of 104 distinct areas in the cerebral
cortex alone.
"Seed AI" refers to an Artificial Intelligence designed
for self-modification, self-understanding, and recursive
self-improvement. A seed AI is an AI designed with
the intention that it should eventually read, rewrite, and
enhance its own source code and design, then become capable
of further improvements at that higher level of intelligence;
an AI designed to use its overall intelligence to improve
its components, or whose component processes are designed
to use its overall intelligence.
The
principles, goals, and expected advantages of seed AI; a
short introductory article. (Singularity dynamics,
future AIs, and what an AI does once it's built.)
A paper by SIAI Research
Fellow Eliezer Yudkowsky, to appear as a chapter in "Real
AI: New Approaches to Artificial General Intelligence"
(Goertzel and Pennachin, eds., forthcoming). Presents
a full exposition of the foundations of the Singularity
Institute's intended AI design, including the decomposition
of intelligence into interdependent, internally specialized
subsystems; cognitive processes existing within five levels
of organization; grounding in neuroscience, cognitive psychology,
and evolutionary theory; and analysis of the prerequisite
competencies for recursive self-improvement and the resulting
developmental trajectories. Printable and single-page
version available. (380K).
Although we are not ready to start building the
Friendly seed AI, we are close enough to begin forming the
development team. At present, we have two confirmed team
members. We are now actively searching for
Singularitarians with software engineering and
cognitive science expertise to join the development team.
The predecessor to "Levels of Organization",
written in informal language; recommended as a user-friendly,
more easily readable exposition of AI theory. Printable
and single-page versions are available. (360K).