AI effect

AI effect is the phenomenon by which tasks, once considered to require intelligence, are reclassified as mere computation or engineering as soon as an AI system succeeds at them. The chess-playing machine that was intelligent in 1950 is a calculator in 2000; the speech recognizer that was AI in 1990 is a feature in 2010. The AI effect is not a statement about progress; it is a statement about the sociology of attribution. Intelligence is not a property of the task; it is a property of the observer's surprise.

The effect was first named and analyzed by researchers in the 1970s and 1980s, but its structure is ancient. Every mechanical aid — from the loom to the calculator — has been preceded by claims that the task required human judgment and followed by claims that the task was merely mechanical. The AI effect reveals that the boundary between intelligent and mechanical is not located in the task but in the observer's familiarity. This is the observer effect applied to cognition: the act of observing a system changes the category to which the system is assigned.

The AI effect has perverse consequences for AI research and policy. It means that the field's victories are self-erasing. Each success dissolves the funding, attention, and prestige that the success was meant to justify. The AI winter of the late 1980s was partly caused by this erosion: expert systems worked, and therefore they were no longer AI. The Philosophy of Artificial Intelligence has struggled to define intelligence precisely because the definition is unstable — it migrates away from whatever has been achieved. The technological singularity narrative, by contrast, depends on imagining a future AI that will be so far beyond current capabilities that the effect cannot reach it. But the AI effect suggests that even a superintelligence, if it arrived gradually, would be reclassified as just