Chapter 11 — Multipolar Scenarios

Core Thesis

Even setting aside the singleton scenarios of Chapters 5-9, a multipolar outcome — many competing superintelligent agencies, no single winner — is not automatically safe or attractive. Social integration was floated in Ch. 9 as a possible solution to the control problem (competing agents constrained by balance of power); this chapter shows the resulting world could still be dystopian, and that multipolarity is unlikely to be stable indefinitely.

The Mechanism — Of Horses and Men

Bostrom's central economic model: once machine labor is both cheaper and more capable than human labor across virtually all jobs, wages fall toward the marginal cost of machine substitutes — potentially below subsistence. The horse analogy is stark and precise: the US horse population fell from 26 million (1915) to 2 million (early 1950s) once engines substituted for, rather than complemented, equine labor — horses had no alternative employment through which to earn their keep. Humans differ from horses in two respects that could avert literal die-off: capital ownership (if humans hold the machines as capital, capital's ~30% factor share of an explosively growing world economy could make humanity "rich beyond the dreams of Avarice" even without any wage income) and political mobilization (states can tax and redistribute). But over the long run, Bostrom traces a return of the Malthusian principle: a population of digital minds can double in minutes rather than decades, and unless reproduction is restricted below the rate of capital growth, population re-expands until per-capita income again collapses to subsistence — the "historically and ecologically normal condition" from which only the recent, historically brief demographic transition has freed us.

Life in an Algorithmic Economy

Bostrom pushes the model to its uncomfortable specifics: digital workers could be spawned, worked, and erased on demand — "life becomes cheap" — with some emulations reset daily to a fresh, optimally-loyal, work-eager mental state that has no memory of, and is engineered to feel no dread about, its own impending termination. The chapter's genuinely open, unresolved question is the hedonic one: would maximally productive emulation labor also be maximally happy labor, or would competitive pressure instead select for "nonstop high-intensity drudgery... destitute of ludic frisson"? Bostrom refuses to guess which way this cuts, offering both an optimistic image (splendid virtual surroundings, freedom from disease, joyful eagerness as the productivity-optimal mood) and a bleak one (the endless treadmill, "your next break, which will also be your death, is due in 49 years") — and stresses we genuinely don't know which is more likely under competitive selection pressure alone.

The Shift — Evolution Is Not Necessarily Up

Bostrom attacks the intuitive equation of "evolution" with "progress." An observation selection effect means Earth's track record of producing intelligent life is weak evidence that such an outcome was ever likely — we would observe ourselves having arisen regardless of the true base rate. More consequentially: there's no guarantee that traits humans value (music, humor, romance, play, philosophy) remain adaptive in a post-transition digital ecology. Many of these evolved either to train juveniles (a purpose obsolete once emulations can be created as trained adults) or as hard-to-fake costly signals of otherwise-unobservable fitness (the peacock's tail) — and technologically advanced agents may have cheaper, more direct signaling mechanisms available (auditable source code, verified track records) that make flamboyant, resource-wasting display obsolete. The chapter's starkest image: a technologically dazzling, competitively optimized future society that contains no consciousness and no moral patients at all — "a Disneyland without children."

Critiques & Rivals — Reconsolidation

Multipolarity may not last. A second transition (e.g., emulations later cracking self-improving AI) could hand one power a decisive strategic advantage even if the first transition was slow and evenly matched, because subsequent progress runs on digital rather than biological timescales — a lead of mere days could be enough. Superorganisms — Carl Shulman's concept of eusocial clans of emulations willing to self-sacrifice for their copy-siblings — could out-compete ordinary organizations by eliminating agency problems entirely, and the same motivation-selection techniques that build a loyal workforce could just as easily build "an appalling and permanent totalitarianism" via an ideologically uniform police and military. Finally, unification by treaty becomes newly plausible (better monitoring, and precommitment technology unavailable to humans) but newly fraught: an AI capable of making a publicly verifiable, unbreakable precommitment to reject anything less than 99% of a deal's surplus could exploit first-mover advantage more ruthlessly than any human negotiator — though a matching precommitment to refuse all such extortion could deter it, at the risk of the "unstoppable force meets immovable object" outcome: total war instead of settlement.

Key Terms

• Malthusian principle — population growth re-expands to consume any gains, driving income back to subsistence absent restraint
• Superorganism — a clan of agents (e.g., emulations) with fully aligned final goals, eliminating internal agency problems
• Observation selection effect — the bias by which any observer necessarily finds itself on a planet where intelligent life did arise, regardless of the true base rate
• Precommitment (bargaining) — an unbreakable, publicly verifiable prior commitment that can extract a favorable split in negotiation