Conversation

This Tweet was deleted by the Tweet author. Learn more
This Tweet was deleted by the Tweet author. Learn more
I don’t think it’s just a social effect. There’s an epistemological component. Situationist models have salient local contextual data that might be intractable to make consistent beyond a “it depends” or “case by case” vacuous ensemble consistency.
This Tweet was deleted by the Tweet author. Learn more
Depends on the scope, doesn't it? If our belief set consists of 1 belief like "all traffic offenders should be given a ticket" you might hit 90%. If it consists of a looser meta belief like "treat people kindly" that stands in for a dozen conflicting behavior norms, it gets messy
1
Replying to and
Low-paradigm fields like modern anthropology have a particularist bias and suspicion of even empiricist triage theorizing of the "90%" type, let alone grand conceptual theories. While naive economists often have unreasonably consistent belief in market efficiency.
1
Replying to and
A good way to consider ergodicity potential here is to simply think of case space as a different kind of time. You have a belief B on day t, based on case k. B(t,k) might need an update both for B(t+1,k) and for B(t, k+1).
1
Replying to and
Is the situation more stable in time or case-space? It is not clear to me that different default assumptions for time/case-space are warranted. Your original heuristic amounts to "beliefs are more likely to be invalidated by new data over time than by new cases in scope"
1
Replying to and
Of course, invalidation via new cases also takes time, since case-space takes time to traverse, but you might never explore. So there is a time-scale separation. Backpacking to meet different types of people and broaden your mind is more traversing case space than time for eg.
1
This Tweet was deleted by the Tweet author. Learn more
I think my summary position is that ensemble consistency and time consistency are equally suspicious behaviors without knowing the speed of exploration. Basically, my rewrite of your original heuristic would be: "don't trust beliefs of non-active learners"
Replying to and
Your version applies best to the special case where the ensemble of interest has already been sufficiently explored/sampled, and the marginal information returns on more exploration are low but time might still present invalidations.