7:10 AM
@AncientSwordRage Ah. Gotcha. That's the implicit cognition. However, if you've got a model and you can't figure out ways to get it to "pump the wrong direction" or emit a bad outcome, then you're looking at something that's non-falsifiable, and that's not science.
I think a more tractable example would be something like, "in physics, can you have a perpetual motion machine?" And with the current model, the answer is "yes, but it would require negative mass."
That kind of assumption, "assume negative mass exists", is something that would get noticed (and did for a number of "warp drive" papers a few years back) and flagged for the proposed model.
Similarly, in social science or community health models, one could choose predicates, tunings of parameters, or training data that emits a result affirming some outcome of systemic racism. I would expect few would bother to go through the exercise, and if it were done, the assumptions needed to get such a result get flagged.
A far more prevalent problem currently is observation sets that have baked in racist or systemic racism effects that are then used as training data for models. E.g. risk assessment for mortgages and loans or one that is a terrible problem here is auto and home insurance rates. The actuarial models based on historic data has the problem of being terrible skewed in this way.
7 hours later…
1:57 PM
@GcL I think the fundamental issue with our conversation (leaving anything said by us or anybody else in the main chat aside) is that you seem set on viewing any argument from within science (by which I mean 100% predicated on empirical evidence and logic), and my whole argument is that some decisions need to be made (even if only partly) outside of science (i.e. using something that does not rely on empirical evidence). As a different example, say a person knows they will give birth to a child with a congenital condition. You can look at data collected, look at surveys, build decision tree…
« first day (466 days earlier) ← previous day next day → last day (695 days later) »
Transcript for
Jan24
Jan '2225
Jan26
Here There Be Dragons
For topics that should be opt-in, not opt-out. Please respect ...