The Asymmetry of the Wager
In the seventeenth century, Pascal argued that rational self-interest demanded belief in God, not because the evidence supported it but because the asymmetry of outcomes — infinite salvation versus infinite damnation — made belief the rational hedge. Jonas saw in Pascal's wager a structure that could be repurposed for the technological age. Not the specific content — he was not arguing about God — but the logical architecture: when stakes are asymmetric, when one outcome is catastrophic and irreversible while the other is merely costly and recoverable, the rational strategy is to prioritize avoidance of the catastrophic outcome regardless of its estimated probability. The wager is not about what is likely. It is about what cannot be undone.
In the AI Story
Applied to AI, the asymmetry takes a specific form. If the optimistic projection proves correct — AI tools genuinely democratize capability, raise the floor of who can build, enhance human creativity — the cost of having been cautious is real but recoverable: months or years of slower adoption, deferred benefits, the developer in Lagos gaining access later than she might have. These costs are genuine and fall disproportionately on those who can least afford to wait. But they are recoverable: the capability arrives, the ideas are still there, the building happens on the same foundation.
If the pessimistic projection proves correct — the quieter and more plausible version, the systematic erosion of conditions under which genuine human cognitive development occurs — the cost of having been bold is not a deferred benefit. It is a developmental deficit in an entire generation, a deficit that cannot be corrected retrospectively because critical developmental windows in which cognitive capacities are built have passed. The child whose capacity for sustained attention was not cultivated between eight and sixteen cannot be given those years back at twenty-five. Neural architecture that forms during critical periods is not infinitely plastic.
The asymmetry is this: the cautious strategy, if wrong, produces recoverable loss. The bold strategy, if wrong, produces irrecoverable loss. The magnitude of the irrecoverable loss — potential deformation of the cognitive conditions under which an entire generation develops — dwarfs the magnitude of the recoverable one. Pascal's wager holds, not because the pessimistic outcome is more likely, but because its consequences, if realized, cannot be corrected by subsequent action.
The counterargument is familiar: history is littered with false alarms, Socrates worrying about writing destroying memory, the Luddites predicting machines would destroy skilled labor. Jonas's reply applies the heuristics of fear's two-condition test: the feared outcome must be plausible (grounded in identifiable mechanisms, not speculative fantasy) and must be irreversible (not merely harmful, but harmful in a way that forecloses future correction). The Berkeley study's documentation of intensification, task seepage, and the blurring of voluntary engagement and compulsion satisfies the plausibility condition. The irreversibility condition — whether the cognitive effects will prove reversible — cannot yet be assessed, and that uncertainty is itself the relevant moral fact.
Origin
The wager formulation appears most explicitly in Jonas's 1973 essay 'Technology and Responsibility' and is developed in The Imperative of Responsibility. It draws on Pascal's structure while redirecting it from theological to technological application.
The argument has influenced discussions of long-term risk in climate ethics, existential risk studies, and contemporary AI safety discourse, though Jonas's specific philosophical grounding is often underappreciated in these extensions.
Key Ideas
Priority by irreversibility, not probability. The wager operates on the structure of outcomes, not their likelihood. An improbable catastrophic outcome deserves priority over a probable recoverable one.
Epistemic dimension of irreversibility. The most dangerous cases are those in which damage alters the perceptual apparatus of those who suffer it, making the damage undetectable from inside the damaged condition. A generation whose attentional capacities have eroded may lack the very capacities needed to recognize what has been lost.
The collective wager. Unlike Pascal's individual bet, the AI wager is collective — requiring coordinated restraint among actors whose short-term incentives point uniformly toward maximum deployment. The structural logic of competition makes the wager harder to win than Pascal's.
Merit of unrealized alarm. The alarmist whose warnings produce the precautions that prevent the catastrophe has not been proven wrong. The alarmist has been proven effective. The comfortable retrospective judgment that 'it wasn't so bad after all' misreads the function of the alarm.
Debates & Critiques
Economists and policy analysts argue that Jonas's wager, if applied consistently, would prevent most innovation, since sufficiently imaginative analysts can construct irreversible-harm scenarios for nearly any new technology. Jonas's defenders counter that the wager applies only when the irreversibility condition is actually satisfied — not speculatively but structurally — and that the AI transition's unique feature is that it operates on the cognitive apparatus itself, making the detection of harm uniquely difficult.
Appears in the Orange Pill Cycle
Further reading
- Hans Jonas, 'Technology and Responsibility,' Social Research 40, no. 1 (1973)
- Hans Jonas, The Imperative of Responsibility, Chapter 2 (University of Chicago Press, 1984)
- Blaise Pascal, Pensées, especially fragment 418 (Wager)
- Toby Ord, The Precipice: Existential Risk and the Future of Humanity (Hachette, 2020)
- Nick Bostrom, Superintelligence (Oxford University Press, 2014)