The AI Effect: How Solved Problems Become Invisible

The article describes the AI effect: as soon as a task is solved, it stops being perceived as requiring intelligence. Originated from Larry Tesler and described by Turing in 1950, the effect has several mechanisms: intelligence defined by ignorance, Moravec's paradox (tasks that are hard for humans are considered intelligent), and successful components breaking off and getting their own names. The author goes through typical e-commerce stack showing former research topics that became routine features: spelling correction and ranking in search (now just 'search'), collaborative filtering (now 'recommendation block'), text classification (now 'feed mapping'), entity resolution (now 'offer stitching'), vehicle routing (now 'delivery planner'), fraud detection (now 'anti-fraud'). Reverse process also occurs: 'AI' label is attached to things for marketing or funding. The consequences of renaming are serious: loss of dedicated owner, quality monitoring (only availability), postmortems, and accountability. Regulatory requirements (GDPR, transparency of recommendation systems) appear later, but there is no one to explain the logic to regulators.