Scaling Shadows

In the AI Story

The semiconductor industry did not treat shadows as problems to be solved and forgotten. It treated them as permanent companions of the scaling trajectory. Each generation required new accounting of shadow costs, new engineering solutions to keep them manageable, and new assessments of whether gains still outweighed costs. The accounting was continuous. The solutions were never final. This discipline of measuring shadows with the same rigor applied to measuring gains is the single most transferable lesson from semiconductor history to the AI transition — and the lesson least absorbed by the current discourse.

The shadows of AI scaling are documented in the empirical literature Edo Segal engages in The Orange Pill. The Berkeley study by Ye and Ranganathan, embedded for eight months inside a technology company, measured them with engineering specificity. The first shadow is intensification: AI tools did not reduce work, they increased it. Workers took on more tasks, expanded into adjacent domains, and filled every freed minute with additional activity. This is the thermodynamic consequence of reducing execution friction — when execution becomes cheaper, the system does not rest. It executes more. The per-unit improvement is real; the system-level consequence is the opposite of what the per-unit improvement would naively suggest.

The second shadow is boundary erosion — what the Berkeley researchers called task seepage. AI-assisted work colonized previously protected time: lunch breaks, elevator rides, gaps between meetings. These interstices had served as informal cognitive rest periods. When AI made productive work possible in any thirty-second gap, the gaps disappeared. The third shadow is dependency: as AI tools become more capable, users become more dependent on continued access. The fourth is judgment atrophy — the neurological consequence of outsourcing cognitive capacities that, unexercised, weaken the way unused muscles weaken.

Moore's framework does not suggest shadows should prevent scaling. The semiconductor industry did not stop shrinking transistors because heat dissipation scaled with density; it measured the heat, developed cooling solutions, redesigned architectures, and continued. The shadows were managed, not eliminated. The same discipline applies to AI: intensification requires organizational structures that manage it (what the Berkeley researchers called AI Practice); boundary erosion requires new explicit boundaries; dependency requires redundancy; judgment atrophy requires workflows that preserve the practice that maintains judgment. These are engineering responses to engineering problems.

Origin

The shadow concept is implicit throughout semiconductor engineering culture, articulated most clearly in the thermal and power-management literature of the 1990s and 2000s. The application to AI — treating intensification, task seepage, dependency, and judgment atrophy as scaling shadows analogous to semiconductor waste heat — is articulated in this volume as a synthesis of Moore's management discipline with the empirical findings of Ye and Ranganathan's 2026 Berkeley study and related research on AI-induced skill degradation.

Key Ideas

Shadows are thermodynamic, not accidental. Every scaling law produces a quantity that grows alongside the gain in the opposite direction of value; these are features of the physics, not aberrations.

Management, not elimination. The semiconductor industry's greatest achievement was not scaling itself but the continuous shadow-management systems that allowed scaling to continue.

Intensification, not reduction. AI tools do not reduce work; they increase it, because reducing execution friction enables more execution per unit of human effort.

Engineering responses to engineering problems. Shadows require measurement and structural response, not philosophy, not celebration, not withdrawal.

The margin determines when walls arrive. Scaling laws describe the average relationship; shadows operate at the margin, and it is the margin that determines when the trajectory breaks.

Debates & Critiques

The most contested shadow is judgment atrophy. Some researchers argue that AI tools free cognitive capacity for higher-order judgment rather than eroding it — that the capacity moves rather than disappearing. Others, including the Berkeley researchers and studies on skill degradation in automated environments, document measurable decline in professional capability among heavy AI users. Moore's framework suggests both may be true simultaneously: the capacity relocates for some users and atrophies for others, with the distinction determined by the organizational structures that surround the tool use.