I'd say this is a huge reason to actually push for more AI, especially open models. I agree that there are serious environmental, etc concerns BUT there will be an even larger problem in the future when this tech becomes incorporated into pretty much everything if there are only a couple providers with a -poly on the really good models. This thing is here to stay and will only get qualitatively better, and the way to prevent/offset the "omnipresent big brother" is to ensure that everyone has reasonable access to models with decent capability and know how to properly evaluate and use them.
I don't know if more models will offer a meaningful counter to those in power, who will have not only more resources, but also the military, law enforcement and effectively the law itself.
Probably the most we can hope to do at this point is to dismantle the surveillance architecture. "Starve the beast" so to speak.
I don't think there's a way to "starve the beast", unless completely isolating from society is an option (which it isn't for any but a sparse few). But the data it lives on can be diluted and made more piecemeal by keeping things local first, and spreading usage across multiple jurisdictions when more than local is needed.
This, but we need to prevent as much of this data from being collected in the first place. All our traffic cameras for instance are being coopted into a massive surveillance network with no oversight, but this network doesn't exist if you don't have traffic cameras in the first place.
I propose the J Edgar Hoover heuristic. Would X be something you'd want Hoover to have access to? If not, then don't build it.
