As I was going home from the champaign airport after a very eventful EAG conference, I struck up a conversation about AI safety with my lyft driver. I thought to myself “It’d be good to see what the public thinks. Maybe inform him of the risks. It’s positive counterfactual.”
But as I spoke to him he started dropping words like “UBI” and drawing analogies to power concentration and gradual disempowerment. I was surprised to see how much he understood already. The conversation went one way or another and he went on to talk about his experience in the military. He spoke about the efficiency of the military, how they function with clear hierarchies, straightforward agendas, and utter coordination. Amidst his thorough and proud explanation, I began to envy how the military is able to function so smoothly as a “well-oiled machine”.
He kept talking about how life in the military is completely structured. How instructions are simple, people know what to do, and that they make sure to do it.
I said to him “So none of your work is bottlenecked by other people.” And then he said something that stuck with me:
“If we have bottlenecks, people die”.
...
Isn’t this the same situation we are in?
If we truly believe in short timelines, and that there is a non-negligible probability of catastrophic outcomes, shouldn’t we carry forward such a mindset when approaching AI Safety?
I'm not suggesting that AI Safety should adopt military hierarchies but I think we can take a page out of the military's book in terms of mindset and the seriousness with which we approach this problem.
It seems like we are struggling to internalize the stakes at hand, which is expected, the feedback loop in the military is much faster. Still, I think more people should think of the issue in this way and rethink where their attention goes.
As I was going home from the champaign airport after a very eventful EAG conference, I struck up a conversation about AI safety with my lyft driver. I thought to myself “It’d be good to see what the public thinks. Maybe inform him of the risks. It’s positive counterfactual.”
But as I spoke to him he started dropping words like “UBI” and drawing analogies to power concentration and gradual disempowerment. I was surprised to see how much he understood already. The conversation went one way or another and he went on to talk about his experience in the military. He spoke about the efficiency of the military, how they function with clear hierarchies, straightforward agendas, and utter coordination. Amidst his thorough and proud explanation, I began to envy how the military is able to function so smoothly as a “well-oiled machine”.
He kept talking about how life in the military is completely structured. How instructions are simple, people know what to do, and that they make sure to do it.
I said to him “So none of your work is bottlenecked by other people.” And then he said something that stuck with me:
“If we have bottlenecks, people die”.
...
Isn’t this the same situation we are in?
If we truly believe in short timelines, and that there is a non-negligible probability of catastrophic outcomes, shouldn’t we carry forward such a mindset when approaching AI Safety?
I'm not suggesting that AI Safety should adopt military hierarchies but I think we can take a page out of the military's book in terms of mindset and the seriousness with which we approach this problem.
It seems like we are struggling to internalize the stakes at hand, which is expected, the feedback loop in the military is much faster. Still, I think more people should think of the issue in this way and rethink where their attention goes.