Why do people disagree about when powerful AI will arrive?

Abstract

Projections for the arrival of Artificial General Intelligence (AGI) vary significantly, ranging from a few years to several decades. Advocates for short timelines emphasize the rapid saturation of academic benchmarks, exponential increases in the time horizons of autonomous task completion, and the potential for automated AI research to trigger a recursive intelligence explosion supported by massive compute scale-ups. Conversely, skeptics argue that existing evaluations fail to reflect complex, contextual real-world workflows, pointing to Moravec’s paradox, hardware and data bottlenecks that limit self-improvement, and the dependence of scientific discovery on factors beyond abstract reasoning. Because current compute scaling rates may reach financial and physical limits by 2030, the coming years represent a critical test period, underscoring the urgent need for robust safety and alignment research despite timeline uncertainties. – AI-generated abstract.