AI Pacing Debate Misunderstands Business Adoption Lag

A new debate has emerged in Washington and Silicon Valley concerning the "pacing" of artificial intelligence development, which involves deliberately slowing down the creation of frontier models until societal readiness, safety measures, and alignment can catch up. Proponents of pacing view it as a necessary safeguard against potential catastrophic risks, while detractors argue it amounts to unilateral disarmament in a global AI race, particularly with China. Both sides, however, are criticized for falling prey to the "Compute-to-GDP Fallacy," a misconception that every advancement in AI model performance directly and immediately translates into macroeconomic output. Historically, widespread adoption and measurable productivity gains from general-purpose technologies have taken decades. While AI's diffusion may be accelerated, the belief that it bypasses these economic realities is seen as hype. In reality, American corporations are significantly behind the current AI frontier, and the commercial success of AI labs hinges more on building trust and achieving widespread adoption than on sheer model capability. The economic cost of pacing is posited to be remarkably low, whereas the cost of unchecked advancement, particularly regarding alignment issues, could be far greater. The current argument is framed as the wrong one, potentially stemming from arrogance or a misdiagnosis of the core challenges. Skepticism regarding pacing is not unfounded, with questions arising about whether it is a genuine safety concern or a strategic marketing tactic by frontier labs and cybersecurity firms to bolster their financial positions ahead of potential initial public offerings (IPOs). Another concern is whether pacing would cede the United States' lead in AI to China, or if Beijing would adopt a similar pacing strategy. The AI frontier has undeniably reached a critical capability milestone, and warnings of catastrophic or existential risks can no longer be easily dismissed, even if their near-term probability is considered modest. However, by concentrating almost exclusively on the most advanced models, frontier labs are accused of mismanaging their public messaging and eroding public trust. Extensive conversations with over 100 CEOs, policy leaders, and AI scientists for an upcoming book, "When Machines Act," indicate that the pacing debate has overshadowed more fundamental issues of AI integration and societal impact. The core issue is not the pace of innovation itself, but the disconnect between cutting-edge AI capabilities and the practical, trust-based adoption by businesses and society. The focus on raw power at the frontier overlooks the substantial lead time required for any technology to be integrated, understood, and trusted enough for widespread economic impact. This lag in business adoption suggests that the immediate economic consequences of pacing are minimal compared to the potential long-term risks of unchecked AI development.