Can AI read (biodiversity) papers ... and understand its contents?

Millions of academic papers are published every year and nobody can even read the papers in their own field anymore. Can AI do it for us? We ran experiments using various chatbots and their knowledge about taxonomy (specifically reptile taxonomy). With web-based searches chatbots can access most of taxonomic knowledge now but there is still a substantial body of knowledge behind paywalls and in the old literature (even though most chatbots can read text in the Biodiversity Heritage Library). Next, we investigated their ability to read taxonomic papers, especially species descriptions, to find data in them and to extract them. The results so far are mixed. Information extraction works well when tested if an AI can extract data into a table, but there are limitations. For instance, in our first assessment of more than 100 species descriptions, chatbots extracted all correct diagnostic features in about 50% of cases. In fact, the rates are higher if certain traits, such as color patterns are ignored. We are currently (summer 2026) investigating other features of AI, such as (1) can AI extract drawings and table data from papers? Can AI draw taxonomic conclusions based on multiple lines of (sometimes contradictory) evidence? How good are computer vision models to recognize species from images (e.g. in iNaturalist) etc. Results from these studies will be presented.