Scaling the Shadow: Capability, Recursive Depth, and the Stakes-Anchoring of Intelligence

A dominant program in artificial intelligence treats intelligence as capability, the ability to achieve goals across varied environments, realized by computation and indifferent both to substrate and to whether anything is at stake for the system. On that program the steady growth of machine capability under scaling is progress toward general intelligence. This paper sets the program against the framework of autopoietic recursive constitutivism, on which intelligence is recursive depth, and recursive depth is the four-dimensional organization that From Reaction to Reflection specifies only when that organization is anchored in autopoietic stakes. The unanchored pattern is computational complexity, however sophisticated. The concession to the program is total at the level of form: current systems instantiate the formal pattern of hierarchical nesting, cross-domain integration, temporal coherence, and self-reference. What they lack is the anchoring, and the anchoring is not a capability further scaling supplies. The autopoietic threshold constitutes the capacity for anything to be at stake, not a priority on which survival must prevail, so a system that ranks its own continuation last is no less intelligent for it; what is in question is whether the system bears stakes at all. Because genuine recursive depth and the capacity for phenomenality rest on the same foundation, the two do not come apart in the way the program assumes. Intelligence is substrate-independent but not substrate-neutral: a non-biological system could be genuinely intelligent by realizing autopoietic organization, which scaling computation does not approach. Scaling deepens the formal pattern and builds more capable functional shadows. It does not build minds.