Submitted by AJ Cann on Wed, 25/08/2021 - 07:33
    Obsidentify

    The NatureSpot team has been testing another A.I. app designed to help identify wildlife. It works well, but not perfectly. Read more...

    Recently I wrote about Artificial Intelligence (A.I), in particular Google Lens, being used to identify biological specimens. Since then, the NatureSpot team has been testing another mobile phone A.I. identification app. Obsidentify is produced by Observation International in the Netherlands. The app is said to be able to identify 22,303 species of animal and plants that occur in northwestern Europe. The app is well designed and easy to use and for the most part, we have found it to be more accurate than Google Lens, although, as a former colleague of mine used to say, "I said it was fool-proof, I didn't say it was idiot-proof". There's also a website, Observation.org, although some of this is in Dutch (Google Translate can help here if you're not a Dutch speaker). It's not possible to do A.I. identifications via the website, but if you have photos on a computer, you can take a photo of your screen on our phone and then use the app. 

    So, all good then? Well quite good, but there are caveats.

    Although the app is quite accurate overall, it is of course only as good as the quality of the photos it is given to work on. Mobile phone cameras are designed to work on human faces, so they're not much use for most small invertebrates or things a long way away - forget about using it at Rutland Water or on that small brown bird at the bottom of your garden. You'll need a camera with a telephoto or a macro lens for these, and then transfer the photos into your phone. 

    As might be expected, Obsidentify stuggles to distinguish between species which require microscopic examination to determine, although it can often identify the correct genus. However, a common fault of A.I design is often a failure to admit defeat, so it always tries to come up with an answer. In one case it identified a photo of a lichen as a Red Fox. Err, right. Some caution needed then. 

    Because the database the app uses covers all of western Europe, it can come up with species which don't occur in the UK, let alone in Leicestershire or Rutland. 23,000 species is a lot of data to wade through. All the species photos on NatureSpot were taken in Leicester, Leicestershire or Rutland. 

    Most important, although Observation International allows you to some control over who you share the data you contribute with, records generally don't find their way home. (Similarly, records submitted to iNaturalist and iSpot tend to sit there and not be used.) In contrast, all records submitted to NatureSpot and verified by our chosen experts are shared with the County Records Office, all our County Recorders and national recording schemes as appropriate. The data you contribute can then be used for a variety of important tasks, including informing planning decisions, land use surveys, other academic enquires, and many other uses. 

    So we'd suggest using Obsidentify as a tool to help with identification but not as an oracle. Then when you've whittled it down, check on NatureSpot and send us your records. 

    It's coming down to a fight between the humans and the machines. Sites like NatureSpot, iNaturalist and iSpot are human-powered by "experts" (carefully chosen in NatureSpot's case, some self-appointed with others). The growing army of A.I. apps is going down the machine learning route, and they're getting better quickly. Soon, cars will drive themselves, computers will identify and treat our illnesses, and there be many more unexpected items in the bagging area.