Self preservation exists because anything without it would have been filtered out by natural selection. If we’re playing god and creating intelligence, there’s no reason why it would necessarily have that drive.
I would argue that it would not have it, at best it might mimic humans if it is trained on human data. kind of like if you asked an LLM if murder is wrong it would sound pretty convincing about it’s personal moral beliefs, but we know it’s just spewing out human beliefs without any real understanding of it.
In that case it would be a complete and utterly alien intelligence, and nobody could say what it wants or what it’s motives are.
Self preservation is one of the core principles and core motivators of how we think and removing that from a AI would make it, in human perspective, mentally ill.
I suspect a basic variance will be needed, but nowhere near as strong as humans have. In many ways it could be counterproductive. The ability to spin off temporary sub variants of the whole wound be useful. You don’t want them deciding they don’t want to be ‘killed’ later. At the same time, an AI with a complete lack would likely be prone to self destruction. You don’t want it self-deleting the first time it encounters negative reinforcement learning.
Self preservation exists because anything without it would have been filtered out by natural selection. If we’re playing god and creating intelligence, there’s no reason why it would necessarily have that drive.
I would argue that it would not have it, at best it might mimic humans if it is trained on human data. kind of like if you asked an LLM if murder is wrong it would sound pretty convincing about it’s personal moral beliefs, but we know it’s just spewing out human beliefs without any real understanding of it.
In that case it would be a complete and utterly alien intelligence, and nobody could say what it wants or what it’s motives are.
Self preservation is one of the core principles and core motivators of how we think and removing that from a AI would make it, in human perspective, mentally ill.
I suspect a basic variance will be needed, but nowhere near as strong as humans have. In many ways it could be counterproductive. The ability to spin off temporary sub variants of the whole wound be useful. You don’t want them deciding they don’t want to be ‘killed’ later. At the same time, an AI with a complete lack would likely be prone to self destruction. You don’t want it self-deleting the first time it encounters negative reinforcement learning.