Civilizational Intelligence

In the AI Story

The question this concept poses is hard: who reasons on civilizational timescales in a world where most institutions don't? Firms are quarterly. Elections are cyclical. Careers are a few decades. Legal doctrine is annual. Yet the consequences of capable AI — value lock-in, power concentration, existential risk, the shape of work and meaning — will unfold across centuries. The mismatch between the timescale of problem and the timescale of institution is the operational core of contemporary AI-governance concern.

Asimov's Foundation novels are the most detailed fictional model of civilizational-scale planning ever written. The Seldon Plan operates on a thousand-year timescale; the novels treat individual generations as waves on a longer current. The key Asimovian insight, now increasingly taken up by contemporary governance thinkers, is that civilizational intelligence cannot be vested in a single individual or a single generation — it has to be carried by institutions that outlast their founders.

Real-world examples of (modest) civilizational intelligence: the Catholic Church (two thousand years of institutional continuity); the British constitutional tradition (unwritten, cumulative, case-precedent-based); the scientific method (cross-generational inheritance of questions and answers); common-law legal systems; nuclear-safety cultures at institutions like Sandia and Los Alamos that persist decades past their founders; the International Atomic Energy Agency; some long-lived corporations. None are perfect instances; all are partial realizations of the idea. Contemporary longtermist organizations (FHI, CSER, the Long Now Foundation) are explicit attempts to build new institutions of civilizational intelligence for AI-shaped problems.

The question of whether AI systems themselves could be civilizational intelligences — carriers of long-horizon reasoning across human generations — is live and contested. Systems with persistent memory, explicit long-term objectives, and institutional authority could, in principle, carry civilizational reasoning. Whether this is desirable, how it would be accountable, and how it would interact with existing human institutions are open questions that the next two decades will answer empirically.

Origin

The concept is implicit throughout political philosophy — Burke's "partnership between the dead, the living, and the unborn" is one articulation — and becomes explicit in the longtermist philosophy tradition of the 2010s, particularly in the work of Nick Beckstead, Toby Ord, and William MacAskill. Asimov's fiction made the idea vivid half a century before the philosophical treatment.

Key Ideas

Timescale mismatch. Most institutions operate on timescales short relative to the problems AI poses. Civilizational intelligence is the resource that could close the gap.

Knowledge persistence. Civilizational intelligence requires institutions that carry knowledge and judgment across human generations. The Catholic Church is the extreme longevity case; scientific disciplines a more modest one.

Generational delegation. A civilization that reasons well across generations requires ways to bind later generations to earlier reasoning — a hard political-philosophical problem.

Plan robustness. Asimov's Plan worked for a thousand years until the Mule appeared. Real civilizational plans need robustness to outliers they cannot foresee.

AI as potential substrate. Unlike any prior carrier of civilizational intelligence, AI systems could in principle preserve reasoning across generations without relying on institutional succession. This is both a promise and a risk.

Longtermism. The explicit philosophical tradition behind effective-altruist long-termist thought. Controversial in its strong forms; foundational in its weak forms.