Sampling Nature: LOOKING TO THE FUTURE FOR AI-READY SAMPLES

Physical samples are always one of a kind, often difficult and expensive to collect, and crucial for advancing research across many scientific disciplines, including geology, environmental science, archaeology, and biology. However, many samples are lost or hidden without adequate metadata or curation, making sharing or reusing challenging. This presentation reports on the Sampling Nature Research Coordination Network’s examination of the challenges and opportunities that emerge when samples are better documented. Most recently, Sampling Nature, with Omic BON and iSamples, explored how to reuse (meta)data and samples to improve biodiversity observations, combine them with data from other domains (e.g., geoscience, archeology), and contribute to global digital interoperability frameworks such as UNESCO-coordinated Ocean Data and Information System (ODIS). This led to the evocative question: What is an “AI-ready sample?” Through an in-person workshop at the Smithsonian National Museum of Natural History, the community drafted a collective definition of “AI-ready samples.” The crux of this definition is that the (meta)data representing samples must be 1) serialized using modern, globally adopted, portable, and cross-domain standards and 2) composed with machine-actionable and widely used semantic resources. Further, AI-ready sample metadata must adequately describe the provenance of the sample. Provenance includes not only scientific metadata to enable (re)use across domains but also legal and social metadata, which are vital for ethical reuse across a range of potential purposes. While machines will play an evermore important role, humans must remain in the loop to set goals, define the rules, and decide how to steward samples and their derived data in an AI world for the common good.