I especially liked the numerical computation of the Piranha theorem of correlation with 8 independent causes. A great intuition pump − 0.35 correlation is almost nothing.
In many cases we should probably expect causes to be exponentially distributed: 1 or 2 causes account for the majority of variance, with a steep dropoff.
On the other side of the coin: if something is influenced by many approximately equal independent causes we again have something easy: the CLT gives us Gaussians. This describes many systems.
More generally, thermodynamics/statistical physics techniques become relevant in these domains.
A well-known but still major problem with high multiple cause explanation is how easy it becomes to blabla away any discrepancy—too much wiggle room. ofc this is primarily epistemic.
A much less-well known problem is that systems with many causal pathways quickly become chaotic.
there is work from Stuart Kauffmann that proves this. I don’t remember the details not but IIRC it was something like if you make a model of gene-protein pathways as a bunch of nodes connected by causal wires than if the number of causal wires of a given node is on average higher than 2 the whole system will be chaotic.
Chaotic systems certainly exist—i.e. the weather, gene-diagrams—but they’re inherently hard/impossible to predict. Any simple stories are therefor necessarily bs.
EDIT: Another relevant link is gwern’s post detailing how in general causal graphs the number of correlations overwhelms genuine causal relations.
Hear, hear! A most excellent post, sir!
I especially liked the numerical computation of the Piranha theorem of correlation with 8 independent causes. A great intuition pump − 0.35 correlation is almost nothing.
In many cases we should probably expect causes to be exponentially distributed: 1 or 2 causes account for the majority of variance, with a steep dropoff.
On the other side of the coin: if something is influenced by many approximately equal independent causes we again have something easy: the CLT gives us Gaussians. This describes many systems.
More generally, thermodynamics/statistical physics techniques become relevant in these domains.
A well-known but still major problem with high multiple cause explanation is how easy it becomes to blabla away any discrepancy—too much wiggle room. ofc this is primarily epistemic.
A much less-well known problem is that systems with many causal pathways quickly become chaotic.
there is work from Stuart Kauffmann that proves this. I don’t remember the details not but IIRC it was something like if you make a model of gene-protein pathways as a bunch of nodes connected by causal wires than if the number of causal wires of a given node is on average higher than 2 the whole system will be chaotic.
Chaotic systems certainly exist—i.e. the weather, gene-diagrams—but they’re inherently hard/impossible to predict. Any simple stories are therefor necessarily bs.
EDIT: Another relevant link is gwern’s post detailing how in general causal graphs the number of correlations overwhelms genuine causal relations.