Ahh—thanks. Yes, if that is what Eliezer is proposing, my above response misunderstood him—but either I misunderstood something, or it would be inconsistent with how I understood his viewpoint elsewhere about why we want to be coherent decision makers.
I remember Toby Ord gave a talk at GPI where he pointed out the following:
Let L be long-term value per unit of resources and N be near-term value per unit of resources. Then spending 50% of resources on the best long-term intervention and 50% of resources on the best near-term intervention will lead you to split resources equally between A and C. But the best thing to do on a 0.5*(near-term value)+0.5*(long-term value) value function is to devote 100% of resources to B.
That’s exactly why it’s important to clarify this. The position is that the entire value of the future has no more than a 50% weight in your utility function, not that each unit of future value is worth 50% as much.
Are there two different proposals?
Construct a value function = 0.5* (near term value) + 0.5* (far future value), and do what seems best according to that function.
Spend 50% of your energy on the best longtermist thing and 50% on the best neartermist thing. (Or as a community, half of people do each.)
I think Eliezer is proposing (2), but David is proposing (1). Worldview diversification seems more like (2).
I have an intuition these lead different places – would be interested in thoughts.
Edit: Maybe if ‘energy’ is understood as ‘votes from your parts’ then (2) ends up the same as (1).
Ahh—thanks. Yes, if that is what Eliezer is proposing, my above response misunderstood him—but either I misunderstood something, or it would be inconsistent with how I understood his viewpoint elsewhere about why we want to be coherent decision makers.
I remember Toby Ord gave a talk at GPI where he pointed out the following:
Let L be long-term value per unit of resources and N be near-term value per unit of resources. Then spending 50% of resources on the best long-term intervention and 50% of resources on the best near-term intervention will lead you to split resources equally between A and C. But the best thing to do on a 0.5*(near-term value)+0.5*(long-term value) value function is to devote 100% of resources to B.
Diagram
That’s exactly why it’s important to clarify this. The position is that the entire value of the future has no more than a 50% weight in your utility function, not that each unit of future value is worth 50% as much.