Two Types of AI Existential Risk: Decisive and Accumulative

Existential AI risks may emerge gradually rather than suddenly: The paper introduces an "accumulative AI x-risk hypothesis" suggesting that existential catastrophe could result from a series of smaller, interconnected disruptions that gradually cross critical thresholds over time, rather than from a single dramatic event like an uncontrollable superintelligence takeover.

Current AI risk discourse may be incomplete: Conventional discussions focus heavily on abrupt, severe events from advanced AI systems, but this focus "often neglects the serious possibility of AI x-risks manifesting incrementally," potentially leaving society unprepared for slower-moving threats.

Multiple AI-induced threats could converge to undermine society: The paper describes a "boiling frog scenario" where "incremental AI risks slowly converge, undermining societal resilience" through accumulated vulnerabilities and systemic erosion of economic and political structures, eventually triggering irreversible collapse.

Different risk pathways require different prevention strategies: By contrasting the "decisive" hypothesis (sudden AI takeover) with the "accumulative" hypothesis (gradual degradation), the paper suggests that policymakers and researchers may need distinct approaches depending on which pathway poses the greater threat.

The two perspectives on AI risk may be reconcilable: The paper argues that distinguishing between these two hypotheses "can reconcile seemingly incompatible perspectives on AI risks," suggesting that both gradual and sudden risks deserve serious consideration rather than being viewed as competing theories.