Study Finds Weak Foundations in AI Doomsday Arguments

Summary

A new study, published in AI and Ethics, reviews 81 papers on existential risks from artificial intelligence and examines the assumptions supporting their arguments. It finds that a substantial part of the literature treats consciousness, sentience, autonomy, AGI, and superintelligence as plausible future conditions without clearly defining them or showing how they would arise. The review also criticizes the use of benchmarks and present-day technical performance to generalize about future autonomy and danger. A second concern is the use of probability estimates, timelines, and intelligence-explosion scenarios: the study says these numerical forms can create an impression of precision and urgency even when their premises are highly uncertain. The review further argues that many x-risk papers treat AI as an autonomous system whose capabilities steadily increase, while giving too little attention to infrastructure, politics, material conditions, history, and counterarguments. It says interdisciplinary perspectives and critical voices are underrepresented in the corpus. The article presents the review as an explanation for why parts of the existential-risk debate have been criticized for weak scientific and normative validity, while its conclusions remain a critique of the literature’s assumptions rather than evidence that existential risks are impossible.