Every once in a while, someone will snidely suggest that perhaps AI safety people should be bombing datacenters, and the fact that they aren’t is evidence that they’re not serious.
As a longtime LW reader (and even early (small) MIRI donor!) I'm broadly sympathetic to the ideas here, but I find it harder to dismiss the skeptics' concerns. Like, suppose you thought international cooperation was hopeless (a common view) and also thought AI dangers were imminent (also a common view). Maybe violence does nothing to alter the long-term trajectory…but could it buy a month or two of normalcy for your loved ones? What if that's all that can be done? Arguably, at that point fretting about our democratic institutions has already become irrelevant. MIRI's "death with dignity" pseudo-slogan leaves open the question of what counts as dignity. Empirically, for many humans, going down in a futile blaze of glory is an archetypal case of preserving one's dignity! Personally I see the honor in *not* going down that path, but I think that requires a bunch of extra steps that are not inherent in the reasonable AI safety worries themselves. So it just seems like an uncomfortable reality that the prospect of violence will always be lingering in the background.
"terrorism is empirically wrong" is one of the worst arguments on the planet. Nobody remembers the successful cases of terrorism, because 'terrorism' is a descriptive label; the cases that succeed end up rebranded as liberators post-hoc.
Witholding violence as an option is *necessary* to build a truth aligned society, but not *sufficient*. Although many truths are easier to convey and propagate in the absence of violence, there are certain truths which are inherently harder to disseminate in its absence (safe example: importance of military defense).
I think the deontological reason is the only real reason. EA/Safety doesn't bomb datacenters, because they don't like violence.
if we lived in a world where terrorism was a very effective way of accomplishing goals, because society didn’t adequately punish terrorism, then the world would look very different. environmentalists would be blowing up factories and pipelines every day (whereas actual ecoterrorism is negligible in magnitude), nimbys would be blowing up construction sites, politicians would constantly be assassinated and purged left and right at USSR/PRC levels, abortion clinics and ICE detention centers and gun stores would become very dangerous, etc. clearly this would be a terrible world to live in, nor is it the one we live in
Step back in time 20 years, and apply your rhetoric to deep learning.
You can only argue the lack of successes for X is proof of X's inherent stupidity, when X has been tried hard enough.
For X=terrorism, it should be obvious that there exists no efficient market for it, that people coordinate to prevent any knowledge building in the domain, that the extreme enforcement of its suppression is done out of the valid concern that it may be made to succeed one day.
Terrorism doesn’t even seem to work to achieve its own aims consistently apart from the people who value murder for its own sake. Like did the guy who murdered arch duke Ferdinand plan on starting world war 1? Did the guy who killed Lincoln end up with a better US? The theory of change doesn’t work
Also terrorists are very rarely the most violent people. If we actually did might makes right, terrorists would very rarely be the strongest people around and you probably don’t want to be ruled by the most insane local warlord even if you’re a terrorist
As a longtime LW reader (and even early (small) MIRI donor!) I'm broadly sympathetic to the ideas here, but I find it harder to dismiss the skeptics' concerns. Like, suppose you thought international cooperation was hopeless (a common view) and also thought AI dangers were imminent (also a common view). Maybe violence does nothing to alter the long-term trajectory…but could it buy a month or two of normalcy for your loved ones? What if that's all that can be done? Arguably, at that point fretting about our democratic institutions has already become irrelevant. MIRI's "death with dignity" pseudo-slogan leaves open the question of what counts as dignity. Empirically, for many humans, going down in a futile blaze of glory is an archetypal case of preserving one's dignity! Personally I see the honor in *not* going down that path, but I think that requires a bunch of extra steps that are not inherent in the reasonable AI safety worries themselves. So it just seems like an uncomfortable reality that the prospect of violence will always be lingering in the background.
"terrorism is empirically wrong" is one of the worst arguments on the planet. Nobody remembers the successful cases of terrorism, because 'terrorism' is a descriptive label; the cases that succeed end up rebranded as liberators post-hoc.
Witholding violence as an option is *necessary* to build a truth aligned society, but not *sufficient*. Although many truths are easier to convey and propagate in the absence of violence, there are certain truths which are inherently harder to disseminate in its absence (safe example: importance of military defense).
I think the deontological reason is the only real reason. EA/Safety doesn't bomb datacenters, because they don't like violence.
if we lived in a world where terrorism was a very effective way of accomplishing goals, because society didn’t adequately punish terrorism, then the world would look very different. environmentalists would be blowing up factories and pipelines every day (whereas actual ecoterrorism is negligible in magnitude), nimbys would be blowing up construction sites, politicians would constantly be assassinated and purged left and right at USSR/PRC levels, abortion clinics and ICE detention centers and gun stores would become very dangerous, etc. clearly this would be a terrible world to live in, nor is it the one we live in
Step back in time 20 years, and apply your rhetoric to deep learning.
You can only argue the lack of successes for X is proof of X's inherent stupidity, when X has been tried hard enough.
For X=terrorism, it should be obvious that there exists no efficient market for it, that people coordinate to prevent any knowledge building in the domain, that the extreme enforcement of its suppression is done out of the valid concern that it may be made to succeed one day.
The only correct case is the deontological one.
Terrorism doesn’t even seem to work to achieve its own aims consistently apart from the people who value murder for its own sake. Like did the guy who murdered arch duke Ferdinand plan on starting world war 1? Did the guy who killed Lincoln end up with a better US? The theory of change doesn’t work
Also terrorists are very rarely the most violent people. If we actually did might makes right, terrorists would very rarely be the strongest people around and you probably don’t want to be ruled by the most insane local warlord even if you’re a terrorist