The research program.

The program asks a single question: what would it take for an artificial system to have values at all, rather than a frozen description of them? It is organized as a sequence of papers, each building on the last, moving from diagnosis to foundations to construction.

The diagnosis begins with the specification trap. Any fixed value specification, whether a reward model, a constitution, or a learned preference ordering, fails at closure: the moment it ceases to update from the process it governs, it begins to misfit the situations that process creates. A second line of work develops the mathematical form of this argument, showing that optimization over serializable state discards exactly the path-dependent structure that human values are built from. A third establishes a systems-level impossibility result: systems with frozen parameters and proxy-only monitoring cannot guarantee alignment, and the result identifies the precise conditions a correction channel must satisfy to escape it.

The middle of the series turns to evidence. Across developmental psychology, neuroscience, and comparative cognition, values behave as processes rather than products. They form through feedback, resistance, and consequence, and they degrade when those conditions are removed. From this evidence the program derives concrete architectural requirements for moral agency: affective integration, interoceptive feedback, empathic prediction error, and a developmental sequence in which identity forms before capability expands.

The constructive arm of the program is Subject Zero, a computational architecture for value emergence through embodied moral development. Rather than specifying values, the architecture specifies a developmental process: an embodied system whose own integrity is coupled to the wellbeing of others, raised through a staged curriculum in which capability acquisition is gated on empathic competence. Pre-registered multi-agent testbeds evaluate what values become under different developmental histories.

One paper in the series is published. Five are in preparation, and experimental work on the developmental testbeds is ongoing.