Every AI system can be placed on a chart with two rough axes: task breadth (how many different kinds of problems it can competently handle) and autonomy (how much it can act, plan and decide without a human in the loop). Ten real and hypothetical systems are plotted on the floor grid below — from a spam filter to a hypothetical superintelligence — and the coloured terrain shows the three informal regions those labels describe.
Nothing built so far occupies the AGI region on this chart — today's most capable models are wide but shallow (broad task coverage, still short of matching human competence and judgement everywhere). Superintelligence remains entirely hypothetical: a useful boundary concept for reasoning about risk, not a system anyone has built.
A 3D capability-space chart plots real and hypothetical AI systems by task breadth and autonomy, revealing narrow-AI, AGI and superintelligence as regions on a continuous, fuzzily-bounded map rather than a strict ladder of labels.
Ten systems — from a spam filter to a hypothetical superintelligence — cluster clearly by task breadth and autonomy. Nothing built today sits inside the AGI region, and the colour blending shows how contested and gradual these category boundaries really are.
Jump to a preset system, or drag the breadth and autonomy sliders to place your own probe point anywhere on the chart and read its live classification. Widen the boundary-fuzziness slider to see the regions blur into one another.
There is no agreed technical test for "AGI" — researchers argue over whether it means matching human performance on a fixed benchmark suite, on any economically valuable task, or something broader still, which is exactly why the boundary here is deliberately soft.