Generative artificial intelligence has expanded the range and speed with which candidate chemical structures can be proposed, but molecular originality is not equivalent to therapeutic advantage. A molecule may be unfamiliar in chemical space while remaining difficult to synthesize, poorly aligned with the intended target profile, insufficiently selective for its pharmacological purpose, or unsuitable for further development. This theoretical article examines how medicinal chemistry should interpret these competing dimensions without collapsing them into a single measure of molecular quality. The analysis distinguishes structural novelty from target relevance, synthetic accessibility, intended selectivity, physicochemical and ADME-related developability, and the reliability of the computational predictions on which these judgments depend. It argues that generative molecular design should be treated as a multiobjective decision problem in which some objectives may trade off, some operate as contextual constraints, and some should remain non-compensatory when their failure invalidates otherwise attractive candidates.  The resulting account favors explicit objective-specific evaluation, route-aware synthetic assessment, pharmacology-dependent interpretation of selectivity, prospective developability validation, and reliability-qualified comparison of candidate trade-offs. Therapeutic usefulness therefore depends less on whether generative AI can produce chemically unusual  molecules than on whether those molecules remain defensible after medicinal-chemistry constraints are applied.