The AI Effect: How AI Achievements Stop Being Seen as Intelligence

Summary

The AI effect describes a recurring change in how people define intelligence: once an AI system acquires a capability, that capability may be reclassified as ordinary computation rather than “real” intelligence. The idea covers both a cognitive shift in expectations and a sociotechnical process in which successful techniques become embedded in other fields. Pamela McCorduck described this pattern as a feature of AI research, while Rodney Brooks observed that understood systems are often regarded as “just computation.” Chess and checkers programs, optical character recognition, and speech recognition illustrate the pattern. IBM’s Deep Blue victory over Garry Kasparov in 1997 was treated by critics as brute-force computation rather than understanding, while once-unreliable recognition technologies became standard engineering components. The related saying “AI is whatever hasn’t been done yet,” often associated with Tesler’s theorem, captures the moving-target character of the field. AI techniques have also been absorbed into marketing, automation, and software applications, sometimes losing their AI label as they become widespread. The effect is still debated in relation to large language models and generative AI, whose capabilities are sometimes described as statistical or mechanical after they become better understood. Researchers interpret the phenomenon variously as cognitive bias, technological normalization, or a legitimate philosophical distinction about intelligence. The article also notes that AI terminology was sometimes avoided during AI winters but has become highly visible in contemporary public discourse and marketing. This visibility complicates the older pattern: AI methods may disappear into ordinary products, while the label itself is now widely used. The debate ultimately connects the AI effect to questions about human uniqueness, the boundaries of intelligence, and how humans assign meaning to machine capabilities.