Artificial superintelligence (ASI) is a system that vastly exceeds the best humans in nearly every domain. The canonical definition is Bostrom's: "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest" (Bostrom, 2014). Yudkowsky and Soares, in the passage translated by Santelli (2026), ask for more: smarter than any human and than humanity collectively, able to do science, plan, scheme and improve itself (Yudkowsky & Soares, 2025).
In the jargon, AGI is the station and ASI the destination: an AI able to do AI research is expected to accelerate its own improvement (recursive self-improvement, FOOM) until it leaves us behind. Yudkowsky (2008) added an argument about scale: in the space of possible minds, the "village idiot" and Einstein are nearly indistinguishable points, so an AI could cross that stretch very quickly. The song sums it up with a nerd joke, since on these forums John von Neumann serves as the unit of human genius: "Now von Neumann's obsolete" (osmarks, n.d.).
For doomers, building ASI with current techniques is suicidal, because nobody knows how to set its goals (alignment, orthogonality thesis). For accelerationists it is the gateway to abundance. Critics doubt the premise. Mitchell argues that supposed human "limitations" (bodies, emotions, biases) may be part of what makes us general, and that the urgent problem is the opposite of superintelligence: the brittleness of current systems (M. Mitchell, 2019, ch. 16). The underlying philosophical question is whether "intelligence" names a scalar magnitude that can grow without ceiling, like temperature, or a heterogeneous set of situated capacities (measuring intelligence).