The next generation of machine learning for tracking adaptation texts

Machine learning presents opportunities for tracking evidence on climate change adaptation, including text-based methods from natural language processing. In theory, such tools can analyse more data in less time, using fewer resources and with less risk of bias. However, the first generation of adaptation studies have delivered only proof of concepts. Reviewing these first studies, we argue that future efforts should focus on creating more diverse datasets, investigating concrete hypotheses, fostering collaboration and promoting ‘machine learning literacy’, including understanding bias. More fundamentally, machine learning enables a paradigmatic shift towards automating repetitive tasks and makes interactive ‘living evidence’ platforms possible. Broadly, the adaptation community is failing to prepare for this shift. Flagship projects of organizations such as the IPCC could help to lead the way.