Search

Objective functions in materials artificial intelligence are routinely presented as neutral computational devices that merely minimize formation energy or maximize ionic conductivity. Yet, they quietly embed normative assumptions about what constitutes a “good” material, thereby encoding ethical, social, and scientific value judgments that shape downstream discovery pathways. These functions operate as value vehicles by translating ostensibly descriptive metrics into prescriptive targets that privilege certain outcomes—stability over metastability, efficiency over sustainability—while rendering competing priorities invisible. This critique identifies four interlocking problems: the hidden normativity concealed within technical loss functions, the resulting value monoculture that narrows the space of desirable materials, the measurability trap that biases discovery toward easily quantified properties, and the democratic deficit that excludes affected stakeholders from objective formulation. The consequences of these unexamined assumptions include narrow and path-dependent discovery trajectories, unresolved value conflicts, and a systematic exclusion of societal considerations from materials innovation. Alternative approaches are therefore proposed that treat objective design as an explicit exercise in value articulation, multi-objective negotiation, and participatory governance, thereby transforming materials AI from a value-blind optimizer into a reflexive, value-aware sociotechnical practice.