Perhaps the main goal of AI safety is to improve the final safety/usefulness pareto frontier we end up with when there are very powerful (and otherwise risky) AIs.
Alignment is one mechanism that can improve the pareto frontier.
Not using powerful AIs allows for establishing a low-usefulness, but high-safety point.
(Usefulness and safety can blend into each other in many cases (e.g. not getting useful work out is itself dangerous), but I still think this is a useful approximate frame in many cases.)
Interesting, when you frame it like that though the hard part is enforcing it. And if I was being pithy I’d say something like: that involves human alignment, not AI
“AI Safety”, especially enforcing anything, does pretty much boil down to human alignment, i.e. politics, but there are practically zero political geniuses among its proponent, so it needs to be dressed up a bit to sound even vaguely plausible.
AI safety is easy. There’s a simple AI safety technique that guarantees that your AI won’t end the world, it’s called “delete it”.
AI alignment is hard.
It’s called “don’t build it”. Once you have what to delete, things can get complicated
Sure, this is just me adapting the idea to the framing people often have, of “what technique can you apply to an existing AI to make it safe”.
Perhaps the main goal of AI safety is to improve the final safety/usefulness pareto frontier we end up with when there are very powerful (and otherwise risky) AIs.
Alignment is one mechanism that can improve the pareto frontier.
Not using powerful AIs allows for establishing a low-usefulness, but high-safety point.
(Usefulness and safety can blend into each other in many cases (e.g. not getting useful work out is itself dangerous), but I still think this is a useful approximate frame in many cases.)
Interesting, when you frame it like that though the hard part is enforcing it. And if I was being pithy I’d say something like: that involves human alignment, not AI
“AI Safety”, especially enforcing anything, does pretty much boil down to human alignment, i.e. politics, but there are practically zero political geniuses among its proponent, so it needs to be dressed up a bit to sound even vaguely plausible.
It’s a bit of a cottage industry nowadays.