rogershaw.vercel.app

I need to remind myself to not let perfect be the enemy of good. I don't need to be so serious about future-proofing straight away. Getting off the ground with the basics is the most important part. Posting publicly to hold myself accountable.

People care deeply about iNaturalist, and have a knee-jerk reaction when they see the words 'AI' and 'Google'. I am not a fan of AI or Google whatsoever, and I've been an iNat evangelist for over 12 years. But I see nothing to be concerned about here. Especially since it will almost certainly fail.

Re: Google getting data, again this is using all the same data that is already being used for ID suggestions (image, location, date(?), ID), and nothing more. All iNat photos are available for this under the Creative Commons. And surely Google have already scraped these anyway? ->

The project will be attempting to run those sets of images through an image-to-text transcriber. Like, "This photo shows a person running through a meadow", except with highly-similar taxa. I am 99% sure this will fail, because AI is a stochastic parrot. And that's fine, iNat will ditch it then. ->

There is not enough published material about taxa identification, on iNat or anywhere else, to train an AI model. I am 99.9% certain that this model will solely be analyzing images, which is already being done for the ID suggestions. As everyone has already seen, those suggestions are imperfect. ->

Can we please all just get behind this man? It would heal our world.