This paper critically examines the evolving relationship between architects and artificial intelligence, arguing that current debates—largely focused on generative AI and image production—overlook the deeper potential of human–machine collaboration in design processes. The authors revisit earlier research in AI for design, particularly from the 1980s and 1990s, to reconstruct a conceptual continuity with contemporary developments and to reframe AI not as a tool for visual output, but as an active design partner.
Through a historical and theoretical review, the study traces the evolution from rule-based systems and early CAD environments to contemporary machine learning and generative models, highlighting persistent limitations in terms of interaction, interpretability, and integration within real design workflows. While current generative tools can produce visually compelling outputs, they often fail to encode architectural logic, resulting in formally plausible but technically inadequate solutions (Fig. 1, p. 5).
The paper further presents two experimental applications that explore alternative collaboration scenarios: one based on plan-driven generation of thin-shell structures, and another using sketch-based interaction for conceptual bridge design. These experiments demonstrate how AI can support designers not only in generating solutions, but also in learning design strategies and exploring performative forms through iterative interaction (Figs. 6–10, pp. 19–23).
The authors conclude by identifying three main directions for future human–AI collaboration in architecture: accelerating decision-making processes, organizing and retrieving complex information, and supporting the acquisition of design knowledge. However, they emphasize that achieving truly collaborative systems requires advances in both technical infrastructures—such as intuitive interfaces and training methods—and theoretical frameworks defining the role of AI within design practice.