AGI vs. Narrow AI vs. Superintelligence: The Capability Ladder
This page is a structured working draft — real analysis, not yet expanded with the full expert sourcing given to the flagship pages. Safe to build on; treat specifics as provisional until sourced.
Three terms people use loosely
Artificial Narrow Intelligence (ANI) is every AI system in production use today, including the large language models behind ChatGPT, Claude, and Gemini. Narrow does not mean weak — a narrow system can outperform any human at its specific task — it means the competence doesn’t transfer. A world-class chess engine cannot fold a protein.
Artificial General Intelligence (AGI) is the threshold where a system’s competence transfers across domains the way a skilled human’s does: reason about a new problem, form a plan, and execute it, without being retrained for that specific task.
Artificial Superintelligence (ASI) describes a system that exceeds the best human performance across effectively all cognitively valuable domains simultaneously, likely including the domain of AI research itself.
Why the ladder matters more than any single rung
Policy and household decisions differ enormously depending on which rung a system occupies. Regulation aimed at narrow-AI harms (bias, job displacement, deepfakes) looks nothing like regulation aimed at a system that could recursively improve itself. Conflating the three terms lets both AI boosters and AI doomers stretch narrow-AI headlines into AGI-scale claims — in either direction. Precision here is not pedantry; it’s the difference between a proportionate response and a wasted one.
The blurry middle
In practice, 2025–2026 systems sit in a genuinely ambiguous middle: they show broad competence across language, code, and some scientific reasoning tasks, well beyond classic narrow AI, but still fail in ways no similarly “generally intelligent” human would. That ambiguity is precisely why capability thresholds (see the overview page) are more useful than a single yes/no label.