AI systems, such as neural-network-based deep learning (DL) and other machine learning (ML) algorithms, can extract valuable insights from data. A major downside of these algorithms is dependence on the availability of sufficient amounts of relevant and structured data. This is clearly problematic for uses in settings where data are scarce and may hamper the development of innovative, creative ML solutions. Hence, there are tensions between ambitions expressed in previous studies to build “universal” solutions based on available (big) data, and the need to contextualize data for specific uses in distinct domains. However, in this paper, we argue that recent advances in ML reduce these tensions and call for more understanding of how these systems can facilitate human creativity. To contribute to such understanding, we present an illustrative application of transfer learning (TL) to facilitate conceptual leaps by broadening algorithmic affordances. This application, involving the use of data on the shapes of lunar craters to identify archeological remains in Swedish forests, highlights how TL can act as a catalyst for cross-domain idea generation through a practical example. By doing so, we theoretically link research on ML development with creativity research, while also demonstrating this connection in practice.