Interview
Origin of the term AGI (Ben Goertzel) | AI Podcast Clips
- The term "AGI" (Artificial General Intelligence) was popularized by the speaker in the early 2000s during the editing of a book titled Real AI, which was eventually published in 2005.
- The initial goal was to compile serious academic papers on "thinking machines," recruiting contributors from transhumanist and singularitarian circles.
- Key contributors to the project included Shane Legg (later co-founder of Google DeepMind), Peter Vose (of AGI Incorporated), Marcus Hutter, and Pei Wang.
- The term "AGI" was selected over alternatives like "Narrow AI," "Broad AI," "General AI," or "GAI" following discussions on a private email list.
- Marcus Hutter's book Universally Intelligent provided a mathematical foundation for the concept, arguing that a truly general intelligence is theoretically possible within computable environments.
- Hutter's theoretical model, "AIXI" (Artificial Intelligence), requires infinite computing power to achieve true generality, rendering it unachievable under current physical constraints.
- The speaker critiques the term "Artificial" because it implies a tool-based nature rather than an autonomous agent, and argues against the artificial/natural distinction since machines are part of nature.
- An alternative term preferred by the speaker and his friend Joshua Bach is "Synthetic Intelligence," though it has failed to gain traction.
- The speaker notes that the word "AI" is widely understood, and "AGI" has persisted primarily due to its sociological function despite technical inaccuracies regarding the definitions of "artificial" and "intelligence."
- The speaker's own child's name, "Exiphanes" (meaning "the former principal underlying AIXI"), incorporates the acronym for "AICSI," while his eldest child is named "Qorxi" (an acronym for "Quantum Organized Rational Expanding Intelligence").
- The speaker contrasts the standard AI definition of maximizing expected future reward with David Weinbaum's "open-ended intelligence" theory, which views intelligence as a complex, self-organizing system that improves without a single objective function.
- Future naming conventions for AI systems are anticipated to potentially replace current hardware terms; the speaker jokes that AGI systems might eventually refer to humans as "Biological Grandpa Processing Units" (GPUs).
- The speaker acknowledges that the origin of "AGI" is not unique, citing an essay by Mark Gubrid published in 1997, and suggests the term is an obvious, retroactive invention rather than a groundbreaking innovation.