Then frankly I find this all overblown and somewhat insane. The AI safetyism moral panic is something I find very demoralizing. It seems like the main legacy of AI will be its use as leverage to annihilate computing freedom and privacy for all time.
I am sympathetic to arguments that quantity has a quality all of its own - for example, I think this has deeply impacted privacy. It may have been legal to discreetly follow someone and note their movements for most of history, but because it wasn't scalable, there was no major privacy threat for doing so. Now it can be done at scale and very cheaply. "Quantity" has qualitatively transformed privacy in practical, real terms.
But nearly all the AI panic seems to be oriented around very different core principle - rather than "technological changes are making us less free in practice" it is "technological changes threaten public harmony." This is the core concern in all authoritarian empires. The idea is that people simply shouldn't be allowed to generate images or texts The Experts have decided they shouldn't be allowed to, because (in this case) it might make journalists' jobs harder, or it might trick too many people, or its offensive, or it offends some really vague notion of "safety" (or can be construed as such by people writing breathless articles for engagement dollars that will be bad PR for the AI company involved.) Yes, doubtless many people will be deceived by AI imagery. Many people already are. This number will certainly be higher than the number of people deceived by altered photographs, which are less in number and require more work to create. But this does not represent a earth-shattering disaster that must be dealt with by banning (somewhat dumb and silly) product ideas like "generate an AI photo based on this satellite imagery" buttons on Google Maps. It can be dealt with by cultural understandings of any imagery shifting, and a general return to older (not unprecedented) information environments. I think in the West, the rise of broadcast and state media really broke people's mental models of information - it really inculcated the idea that people cannot be trusted even in a democratic society to evaluate things on their own: someone must be doing it for them, top-down. The way things worked in the pre-broadcast environment is maybe inconceivable to them now. (And to some extent, there was always a "parallel" information environment that was in the background, but not highly visible to the people in charge, full of conspiratorial nonsense - but they could ignore that because they didn't see it.)
The ultimate endpoint of AI Safetyism is local models are illegal or highly regulated, computing devices are locked down, and only a small list of government-approved AI vendors with tight restrictions are allowed to Do AI. This is totally unnecessary, but it certainly has massive financial benefits for those who thing they can get on the list.
alt Hacker News
Then frankly I find this all overblown and somewhat insane. The AI safetyism moral panic is something I find very demoralizing. It seems like the main legacy of AI will be its use as leverage to annihilate computing freedom and privacy for all time.
I am sympathetic to arguments that quantity has a quality all of its own - for example, I think this has deeply impacted privacy. It may have been legal to discreetly follow someone and note their movements for most of history, but because it wasn't scalable, there was no major privacy threat for doing so. Now it can be done at scale and very cheaply. "Quantity" has qualitatively transformed privacy in practical, real terms.
But nearly all the AI panic seems to be oriented around very different core principle - rather than "technological changes are making us less free in practice" it is "technological changes threaten public harmony." This is the core concern in all authoritarian empires. The idea is that people simply shouldn't be allowed to generate images or texts The Experts have decided they shouldn't be allowed to, because (in this case) it might make journalists' jobs harder, or it might trick too many people, or its offensive, or it offends some really vague notion of "safety" (or can be construed as such by people writing breathless articles for engagement dollars that will be bad PR for the AI company involved.) Yes, doubtless many people will be deceived by AI imagery. Many people already are. This number will certainly be higher than the number of people deceived by altered photographs, which are less in number and require more work to create. But this does not represent a earth-shattering disaster that must be dealt with by banning (somewhat dumb and silly) product ideas like "generate an AI photo based on this satellite imagery" buttons on Google Maps. It can be dealt with by cultural understandings of any imagery shifting, and a general return to older (not unprecedented) information environments. I think in the West, the rise of broadcast and state media really broke people's mental models of information - it really inculcated the idea that people cannot be trusted even in a democratic society to evaluate things on their own: someone must be doing it for them, top-down. The way things worked in the pre-broadcast environment is maybe inconceivable to them now. (And to some extent, there was always a "parallel" information environment that was in the background, but not highly visible to the people in charge, full of conspiratorial nonsense - but they could ignore that because they didn't see it.)
The ultimate endpoint of AI Safetyism is local models are illegal or highly regulated, computing devices are locked down, and only a small list of government-approved AI vendors with tight restrictions are allowed to Do AI. This is totally unnecessary, but it certainly has massive financial benefits for those who thing they can get on the list.