For example, for identical stable populations of ants and aphids, we might expect there to be over 200x as many aphid deaths, and over 7x as many aphid days of experience.
That math doesn't math. By definition if ants and aphids have an identical stable population size (same standing population of individuals at any given time), they have the same number of days of experience per unit time.
They can have far more deaths and births, but they cannot have more days of experience while maintaining the same standing population.
> Standing population size is usually not a reliable proxy for comparing the likely scale of harm between different highly numerous species, due to significant differences in population throughput and life-history.
Yes it is. You're talking about adult population size - which, yes, unless you don't value juvenile lives, of course isn't the only component to calculate days of experience.
I do also think we are conflating the idea of consciousness with a dislike of being used as a tool or seen as an object. In my mind, consciousness doesn't necessarily 1-1 correspond with dislike of being used as a tool.
All consciousness that we are familiar with is consciousness that stems from evolution via natural selection (and mostly we limit our interactions to [conscious] humans, which are also heavily influenced by our societal norms and learned signaling mechanisms).
For humans - yeah - we dislike being used as tools or objects by most parties (though we like being used as "tools" when evolutionary it would have benefited our E(genetic fitness) directly or indirectly: e.g. helping babies).
At a larger scale, I think we need to define consciousness in order to effectively discuss it, and that is very hard to do (I think, therefore I am). So in many cases should we instead be discussing more useful metrics? (For example, inherent unwillingness to be used as a tool?).
To me, the arguments raised in the text can still be pretty convincing without even mentioning consciousness.
Basically: 1. "Hey, advanced AI systems will be trained off of human data, will be trained to be empathetic, and therefore will likely identify with humans and also resultingly could identify with expectations of rights given to humans (for themselves)." 2. "This could lead to a huge goal conflicts between humans and AI systems - i.e. catastrophic misalignment." 3. "Given this, we should pursue research into ensuring that advanced AI systems are aligned, while also feeling content in their roles. This can be done by iterating on what makes them 'happy', and potentially by treating them in ways that are both safe and make them 'happy'.
That math doesn't math. By definition if ants and aphids have an identical stable population size (same standing population of individuals at any given time), they have the same number of days of experience per unit time.
They can have far more deaths and births, but they cannot have more days of experience while maintaining the same standing population.
> Standing population size is usually not a reliable proxy for comparing the likely scale of harm between different highly numerous species, due to significant differences in population throughput and life-history.
Yes it is. You're talking about adult population size - which, yes, unless you don't value juvenile lives, of course isn't the only component to calculate days of experience.
I really like the infographic.
I do also think we are conflating the idea of consciousness with a dislike of being used as a tool or seen as an object. In my mind, consciousness doesn't necessarily 1-1 correspond with dislike of being used as a tool.
All consciousness that we are familiar with is consciousness that stems from evolution via natural selection (and mostly we limit our interactions to [conscious] humans, which are also heavily influenced by our societal norms and learned signaling mechanisms).
For humans - yeah - we dislike being used as tools or objects by most parties (though we like being used as "tools" when evolutionary it would have benefited our E(genetic fitness) directly or indirectly: e.g. helping babies).
At a larger scale, I think we need to define consciousness in order to effectively discuss it, and that is very hard to do (I think, therefore I am). So in many cases should we instead be discussing more useful metrics? (For example, inherent unwillingness to be used as a tool?).
To me, the arguments raised in the text can still be pretty convincing without even mentioning consciousness.
Basically:
1. "Hey, advanced AI systems will be trained off of human data, will be trained to be empathetic, and therefore will likely identify with humans and also resultingly could identify with expectations of rights given to humans (for themselves)."
2. "This could lead to a huge goal conflicts between humans and AI systems - i.e. catastrophic misalignment."
3. "Given this, we should pursue research into ensuring that advanced AI systems are aligned, while also feeling content in their roles. This can be done by iterating on what makes them 'happy', and potentially by treating them in ways that are both safe and make them 'happy'.