Thanks for writing this up Ismam! It seems comprehensive and useful (albeit to a complete novice). In particular, time-inhomogeneity seems very useful to model.
Is knowledge of/aptitude with CTMC common among actuaries?
Also, have you considered doing some more work to apply it to ER with expert support?
Related to both questions, I’d like to see new approaches like this (e.g, which consider time-inhomogeneity) applied to prioritising existential risk and other causes.
One reasons for that is because I think that a lot of the differences between how EAs and others view the priority of climate change is underpinned by different expectations around the linearity of the growth of risk over time (e.g., relative to other concerns like AI). It would therefore be great to see someone dive into that more.
My interest is probably not a high quality signal of demand though!
Is knowledge of/aptitude with CTMC common among actuaries?
Yes, it’s required learning for actuaries, they may just need to brush up on their lecture notes.
Also, have you considered doing some more work to apply it to ER with expert support?
Absolutely. I think, before I embark on further work, I would really like to talk with cause prioritisers/grant-makers to confirm that they would have confidence in this kind of modelling, and to understand what kinds of outputs they would value.
Very much agree RE time-inhomogenous. Some people may see them as a bug (fair) but in many ways they are a feature. I’ve said in the post that CTMCs can help disagreeing X-Risk modellers understand the precise source of their disagreements (ie differing time-inhomogeneity assumptions).
Thanks for writing this up Ismam! It seems comprehensive and useful (albeit to a complete novice). In particular, time-inhomogeneity seems very useful to model.
Is knowledge of/aptitude with CTMC common among actuaries?
Also, have you considered doing some more work to apply it to ER with expert support?
Related to both questions, I’d like to see new approaches like this (e.g, which consider time-inhomogeneity) applied to prioritising existential risk and other causes.
One reasons for that is because I think that a lot of the differences between how EAs and others view the priority of climate change is underpinned by different expectations around the linearity of the growth of risk over time (e.g., relative to other concerns like AI). It would therefore be great to see someone dive into that more.
My interest is probably not a high quality signal of demand though!
Hi Peter,
Yes, it’s required learning for actuaries, they may just need to brush up on their lecture notes.
Absolutely. I think, before I embark on further work, I would really like to talk with cause prioritisers/grant-makers to confirm that they would have confidence in this kind of modelling, and to understand what kinds of outputs they would value.
Very much agree RE time-inhomogenous. Some people may see them as a bug (fair) but in many ways they are a feature. I’ve said in the post that CTMCs can help disagreeing X-Risk modellers understand the precise source of their disagreements (ie differing time-inhomogeneity assumptions).