The essay responds directly to a policy discourse increasingly shaped by two assumptions that Narayanan and Kapoor regard as empirically unfounded: that AI is approaching or will soon approach general intelligence, and that the principal governance challenge is preventing a loss of control to a superintelligent system. Both assumptions treat AI as a nascent species rather than a technology, and both lead to governance recommendations that are either ineffective against the real harms of deployed AI or potentially harmful in their own right—concentrating power in incumbent companies, stifling beneficial applications, or distracting regulatory energy from the mundane reforms that would actually help the people currently harmed by predictive AI tools.
The normal-technology claim rests on a distinction between invention and diffusion that the discourse consistently conflates. The pace of AI capability improvement is genuinely remarkable. But the pace at which new capabilities translate into societal change is governed not by the capability trajectory but by the absorption rate of institutions, professions, and the workforce. Hospitals must reorganize, regulators must understand what they are regulating, workers must acquire new skills, and none of this moves at the pace of a model release. The history of electricity—which took decades to reorganize industrial production after its invention, because the factory floor, the supply chain, and the labor market had to restructure simultaneously—is Narayanan and Kapoor's template for AI's actual trajectory.
The most consequential implication of the normal-technology frame concerns control. If AI is a separate species racing toward superintelligence, the central problem is alignment, and the response may need to be drastic. If AI is normal technology, the loss-of-control scenario rests on a misdiagnosis of what the technology actually is, and keeping AI under human control does not require drastic interventions. It requires the ordinary, evidence-based, institutional work of governance that societies have applied to every previous powerful technology.
The invention-diffusion distinction. New AI capabilities are invented at the pace of research and compute. New capabilities diffuse through society at the pace of institutional absorption, which is governed by organizational redesign, regulatory adaptation, professional retraining, and the slow accumulation of trust and experience. The capability trajectory and the societal trajectory are not the same, and conflating them produces predictions about AI's near-term impact that are systematically too fast. Even electricity and the internet are normal in this sense: transformative but slow to diffuse, and their transformative effects visible only in retrospect rather than in real time.
Governance by evidence rather than speculation. The normal-technology frame redirects governance attention from speculative future risks to documented present harms. The false predictions deployed in consequential decisions, the opaque systems that judge people without recourse, the data practices that violate privacy, the automated injustice embedded in hiring and criminal justice—these do not require novel emergency powers. They require the application of ordinary instruments: transparency requirements, evidentiary standards for consequential systems, accountability for harm, the right to understand and contest automated decisions.
The net-zero aggregate. Consistent with Damodaran's analysis of previous technology cycles, the normal-technology frame predicts that AI's aggregate economic impact will be approximately neutral: value created by winners roughly offsets value destroyed among losers, as it did with PCs, the internet, and smartphones. This is not pessimism about AI but historical pattern recognition, and it implies that the relevant question is not whether AI will create value in aggregate but which specific actors will capture the value it creates—a question that requires the granular analysis that the normal-technology frame makes possible.
The humanlike-intelligence error. Narayanan and Kapoor argue explicitly that viewing AI as humanlike intelligence does not capture what the systems are and does not illuminate how they will matter. The systems are powerful statistical tools, not emerging persons, and the persistent tendency to imagine them as autonomous agents generates a category of concern—alignment with a superintelligent agent—that is irrelevant to the actual harms that deployed AI produces. The error unites the utopians and the dystopians beneath their apparent opposition, directing energy toward the wrong problems and away from the real ones.