Recent research from Google challenges the popular notion of an AI singularity defined by a single, godlike machine. Instead, the findings suggest that intelligence evolves through social aggregation, similar to the development of human civilization. The paper argues that historical intelligence explosions, from primate brains scaling with group size to the creation of writing and institutions, were never about individual upgrades but rather about expanding collective systems. Evidence for this shift is already appearing in frontier reasoning models. Contrary to the belief that larger models simply think better, these systems spontaneously develop internal multi-agent debates without explicit training. Researchers observe what they term a "society of thought," where distinct cognitive perspectives within a single model argue, verify, and reconcile with one another. This indicates that robust reasoning is inherently a social process, a discovery driven by optimization pressure for accuracy rather than specific design. If intelligence is fundamentally social, the future of AI will not rely on building a massive central oracle. It will instead focus on composing richer social systems. The authors propose a transition into a "centaur" era, defined by human-AI ensembles. In this framework, composite actors are neither purely human nor purely machine. AI agents can fork into specialized versions to handle subtasks before recombining results, creating a structure that resembles an organization more than a biological brain. This new perspective demands a shift in governance strategies. Moving away from the monolithic singularity model, which often leads to policies focused on preventing a hypothetical breakout, the plurality model emphasizes the design of mixed human-AI systems. Governance should not rely on a simple kill switch but on checks and balances embedded within the architecture. The authors suggest AI systems should be equipped with distinct values such as transparency, equity, and due process to act as auditors. For instance, a labor department AI could audit corporate hiring algorithms for bias, while a judicial AI could evaluate the risk assessments of an executive AI against constitutional standards. The core message is that intelligence grows like a city, not a single meta-mind. A city lacks central intelligence; instead, it possesses neighborhoods, institutions, infrastructure, and a dynamic mix of conflict and coordination. It is messy and plural, yet far more capable than any individual within it. The next great leap in AI will not stem from a single breakthrough model but from the interaction of billions of humans with hundreds of billions of AI agents. Rather than aiming to create a god, the path forward is to build a civilization. This approach reframes AI development as the construction of social infrastructure, focusing on the norms and institutions that govern mixed-human and artificial societies.