Proleptic Epistemology for Societal Impacts of AGI

The societal impacts of artificial general intelligence are studied almost exclusively through foresight: forecasting, scenario analysis, risk assessment, and anticipatory governance. We argue that foresight carries structural epistemic limitations no methodological refinement can overcome, because its temporal orientation renders present-day preconditions of structural transformation invisible as ordinary. Drawing on the philosophical tradition of prolepsis, we propose proleptic epistemology: a method of generating knowledge about the present by assuming a future state as already realized and examining what becomes visible in retrospect. Applying this lens to four domains of AGI’s societal impact: power structures, labor dispensability, popular sovereignty, and political agency, we show that foresight frames each as an independent risk to be managed. While proleptic epistemology reveals them as stages of a single integrated trajectory toward the structural dispensability of the governed. We conclude that the ML community's epistemic toolkit requires not only better foresight but a fundamentally different temporal orientation, one that generates diagnostic knowledge about the present rather than predictive claims about the future.