AI is here (and has been for a while). It provides hope and potential solutions to many scientific problems, but also raises many ethical problems that can’t be ignored.
For insect biodiversity and conservation research, AI tools can support ID and analysis of species, traits, interactions and behaviours and have huge potential for large scale monitoring. If you use iNaturalist, AI has helped you narrow down your observation ID to relevant ‘suggested species’ (and is usually pretty accurate).
But on social media, does it help or hinder?
Two common unregulated AI uses I’ve seen on social media that I think could be potentially problematic for insect/invertebrate scicomm in the long term:
AI-generated fake insects
With the rise of accessible AI/CGI software, fabricating fiction has also become more accessible. In the last few months, I’ve seen a few posts ‘seeking ID’ of obviously fake invertebrates. They are designed to prank and shock, not educate, playing on the negative stereotypes of terrifying ‘creepy crawlies’. Yet, even though most commenters call them out, some comments show that some people still fall for the horror of the fiction.
What effect do these pranks have on audience awareness of and engagement with insects? It’s disinformation, so should Facebook groups and other online communities moderate and remove such posts?
There are also many ‘legitimate’ uses of AI to fabricate fictional insect ecology images – e.g. images generated for blogs like this at Entomology Today (the image doesn’t show an ‘ecosystem’, as is labelled in the caption, and the ‘insect’ depicted is taxonomically dubious). They are engaging and seemingly harmless, but do we know how this widespread use will impact public awareness and engagement?
Non-specialists using AI apps to publicly ID insects
A lot of people use Facebook groups to seek IDs for critters they encounter, and the comment sections of these posts are dominated by people who have clearly used free digital apps and search functions to dump the original photo for AI recognition. They quickly provide a very confident ID, but sometimes that ID is wrong. Without insect expertise, it’s challenging to know if the returned ‘result’ is accurate.
I’ve tested a few of the ID phone apps and Google image search on Australian insects, and they often return a northern hemisphere species that doesn’t occur in Australia, or a morphologically similar species. I’ve also been involved in a few different research projects to develop and improve AI capability for insect ID, and none have (so far) been successful.
I think AI capabilities have definitely improved in the last few years, but insects are notoriously difficult to identify. There is potential and we’ll get there, but we have a long way to go – about 70% of Australian insect species are still undescribed, so how can we rely on apps (that are trained on existing known species, most from the northern hemisphere) to know what we’re looking at here?
These tools, regardless of accuracy, are now very accessible and popular, so we need to consider how this affects scicomm now. Social media community ID takes pressure off experts, but if that ID is provided at the expense of accuracy, how does it influence scicomm in the long term?
A fly (Pyrgotidae) photographed in my house. Online AI-based ID tools identified it as a cicada or a house fly (Musca domestica) – it is neither.
© Manu Saunders 2024