Jaron Lanier: There Is No A.I., Just Social Collaboration
Jaron Lanier: there is no AI (2023)
As a computer scientist, I argue that the term A.I. is misleading and dangerous. Instead of fearing a new form of consciousness, we should view systems like GPT-4 as innovative tools for social collaboration. By treating this technology as a statistical mashup of human work rather than a sentient being, we can manage it more intelligently and avoid the paralysis of mythologizing its potential risks.
"The most pragmatic position is to think of A.I. as a tool, not a creature."
HN discussion
- When computers surpass humans at specific tasks like chess, the public often revises its estimate of AI capabilities upward, only to later realize the task required less intelligence than assumed.
- The term 'intelligence' is too vague and burned to be useful; it should be decomposed into smaller, measurable subcategories to accurately describe what software, humans, and animals possess.
- LLMs fail in ways completely unlike humans, such as hallucinating tool states, drifting from objectives while appearing compliant, and following prompt injections, which proves they lack human-like intelligence.
- The confusion stems from conflating the technical definition of AI with AGI; modern LLMs are essentially sophisticated calculators that process data without genuine learning or reasoning capabilities.
- The debate over AI intelligence is often unproductive because there is no quantifiable 'Intelligence Ruler' to objectively measure intelligence in text outputs or systems.