Debates on artificial intelligence (AI) in art tend to emphasise aesthetic dimensions, while the epistemic capacities of AI in this domain remain underexamined. This paper proposes a conceptual framework for analysing the intersection of artistic research and generative AI, grounded in Karl Popper's World 3 thesis. Popper's triadic ontology distinguishes between empirical outputs (World 1), subjective experiences (World 2), and objective knowledge structures (World 3). Arguably, artistic development can thereby be situated at the level of artistic problems in World 3. Extending this framework through Lakatos's shift from isolated objects to evolving lineages, the paper outlines an evolutionary epistemology of art in which artistic impact is understood in terms of the formation of problems over time. Within this perspective, artworks are evaluated not only by their aesthetic, material, or sensory qualities, but also by their capacity to generate, extend, and propagate artistic problems within World 3. This approach provides a basis for assessing the epistemic role of AI in artistic research. While generative AI can produce compelling aesthetic outputs, there is presently little evidence that these outputs also contribute to the sustained proliferation or development of artistic problems. From this perspective, artistic impact must be evaluated across large temporal scales, where the decisive criterion is not momentary achievement in cultural or commercial contexts, but the capacity to generate and contribute to enduring lineages of problem formation. Measured against this standard, current forms of AI art remain epistemically limited and can, for the time being, be regarded as relatively infertile within the long-term evolution of artistic reasoning. However, by proposing tentative criteria for detecting and assessing epistemic impact in art generally, the paper offers a preliminary framework for evaluating both AI-generated and human-made art in future developments of artistic knowledge production, while also tracing their interactions.