The relational database has been a fixture of the modern research laboratory -- used to catalog and organize specimens and petri dishes, as well as to organize and store research data and analyses. Yet, though there are numerous textbooks on database design and short-term maintenance, there is still a need for deeper exploration of how these artifacts change, grow and are maintained in the long term, and how their very structure can affect their users' work. Findings from a deeper, more extended exploration of database use over long periods of time would have implications for not just data curation, preservation and management, but also for our understanding of actual, situated information organization practices and needs in science: designing for actual practice rather than for unrealistic idealization of these practices and needs. We draw inspiration, and our title, from Brand's highly influential book; "How Buildings Learn" (1995). We believe many of the topics Brand discusses regarding buildings' change and growth over time might usefully be applied to certain aspects of databases. This work is a first step towards understanding how databases, like buildings, learn.