Continuous Novel Discovery: The Minimum Configuration for Artificial Superintelligence

Every existing definition of artificial superintelligence (ASI) inherits an unmeasurable baseline —the upper limit of human cognitive ability. We propose a definition that avoids this dependency:ASI is continuous novel discovery, a self-sustaining loop in which a system expands its knowledgespace by discovering what was not known, updating its own representations, and moving to thenext unknown, without termination. We formalize this as an infinite sequence of verified strictexpansions of a knowledge state, and distinguish three tiers of novelty by whether the expansionadds content within a fixed representational space (Tiers 1 and 2) or enlarges the space itself(Tier 3). We identify a five-component minimum configuration for the loop — a T3-capablecognitive layer, a physical verification layer, a knowledge-space update mechanism, a directionmechanism, and a cross-domain propagation system — formalize the components as operatorsin a state-transition system, and argue that each is necessary: removing any one yields a systemthat already exists — the AI Scientist, A-Lab, Hakken, the Robot Scientist, or standard LLMinference — whose documented limitations are traceable to the missing component. We evaluatesix current AI discovery systems against this specification and find that none satisfies all fivecomponents, though several achieve genuine results within Tier 1. We propose Experiment as aService (EaaS) as the architecture that closes the physical-verification loop, building on existingclosed-loop infrastructures in pharmaceutical screening and self-driving laboratories, and providea technology inventory identifying what exists and what is missing. No fundamental scientificbreakthrough appears to be required; the distance from the current state to a minimally functionalsystem is engineering, not science.