I’m not a massive contributor to iNat, but it’s extremely important to me as an epicenter of citizen science and a source of truth for identifying and documenting the wildlife around me.
It’s extremely worrying for me to see that iNat is accepting this grant, because there are a ton of problems I can see from reading both the public grant terms, and from my personal experience with genAI in the tech industry.
I’m sure I’m repeating some of the things said up-thread, but it is a terrible, terrible idea to embed genAI into iNat’s recognition algorithms. GenAI algorithms have a persistent and incurable problem with hallucinating statistically likely but baseless information when they’re invoked, and it’s an artifact of the way they’re made. This is not a correctable problem. This is not fixable by some kind of multi-model oversight, by some kind of “reasoning” framework, or by anything else. It’s embedded into how genAI frameworks are created and trained.
GenAI algorithms are also, at this point in their development, experiencing significant issues with diminishing returns on capability based on the saturation of their training data with genAI-created content. This will be a problem for any genAI system that is intended to produce output in natural language, or that is descended from any major model in use today. Like the hallucination issue, this is not a correctable problem.
Additionally, after reviewing the public grant terms, I can immediately see a poison pill embedded into them:
[each grant recipient] will receive a share of $30 million along with six months of technical training, pro bono support from Google’s AI experts and access to Google Cloud credits.
(emphasis mine)
This is a trap, pure and simple. You can’t run a genAI model without access to compute power, and Google will definitely be ensuring that you’re running on Google Cloud. Cloud credits are ruinously expensive - the only way that Anthropic, OpenAI, and all the other getAI providers can exist is because they’re getting free cloud credits from Google and Microsoft.
I know iNat is always concerned with server costs and infrastructure costs – a nonprofit science website doesn’t exactly rake in the cash. Think about what happens if you integrate genAI, and then Google pulls your Cloud Credits. Suddenly, you’re not just paying for your own servers and infrastructure, you’re paying whatever Google wants to charge you for a critical portion of your infrastructure. Your site has become a hostage to Google – and you can’t remove the genAI because Google’s experts helped you integrate it in a lynchpin position. This outcome would OBLITERATE iNat’s funding, and the only assurance you’d have of it not happening would be Google saying “we would never!”
This is not an exhaustive list of my concerns, but it’s the top few. We cannot allow the seductive lure of money from an untrustworthy benefactor to compromise the functionality, integrity, and usability of iNaturalist, under any circumstances. This is one of the few places left where science holds sway – we cannot lose this, not now.