4 ms·

I’ve had the opposite trajectory, and I think there’s a lot of danger in a wider adoption of your stance. The deeper I go on STEM things, the more I realize eve

I’ve had the opposite trajectory, and I think there’s a lot of danger in a wider adoption of your stance. The deeper I go on STEM things, the more I realize everything is, in your words, “bullshit.” I want to be clear that I would phrase that as “based on consensus,” and not as bullshit. I have seen so many doomed/failed projects/models that have come up because people say “well, it’s math, so it has to be right.”

LatteLazy 3y ago
I mean, if the maths was right it was right. If it was wrong and people decided to proceed anyway then that's "consensus based" (or bullshit as I ineloquently put it :) ) like non stem seems to be most of the time. Sorry if I'm missing your point?
- treprinum 3y ago
  Stuff like computing precise results from imprecise inputs etc.
  - LatteLazy 3y ago
    An engineer or scientist will rapidly tell you you cannot do that. A ethicist will never be caught using real inputs at all, precise or not. A Psychologist will tell you 86% of people taking treatment X recovered. But miss out that 85.8% of people on placebo also recovered. Those numbers will be from a study of 10 people.
- mike_hearn 3y ago
  https://en.wikipedia.org/wiki/Mathiness https://en.wikipedia.org/wiki/Mathiness Mathiness is a term coined by Nobel prize winner economist Paul Romer to label a specific misuse of mathematics in economic analyses. An author committed to the norms of science should use mathematical reasoning to clarify their analyses. By contrast, "mathiness" is not intended to clarify, but instead to mislead. According to Romer, some researchers use unrealistic assumptions and strained interpretations of their results in order to push an ideological agenda, and use a smokescreen of fancy mathematics to disguise their intentions.
- JHonaker 3y ago
  The main problem is the extrapolation of the “small world” we construct with mathematics to the “large world” of an application. Your model can be completely internally consistent and allow you to study things in the context of your assumptions perfectly, but that doesn’t mean it doesn’t have critical flaws that disqualify any learning about the system you’ve modeled. The lack of understanding about what implications a representational choice have for the kinds of relationships a mathematical model can describe, and the lack of understanding concerning what are and aren’t critical domain components to model that’s difficult. People often forget that there are two translation steps that have to take place. Domain -> Math -> Domain. That second one is too often implicit or assumed to follow, especially in ML.

LatteLazy 3y ago

I mean, if the maths was right it was right. If it was wrong and people decided to proceed anyway then that's "consensus based" (or bullshit as I ineloquently put it :) ) like non stem seems to be most of the time. Sorry if I'm missing your point?

treprinum 3y ago
Stuff like computing precise results from imprecise inputs etc.
- LatteLazy 3y ago
  An engineer or scientist will rapidly tell you you cannot do that. A ethicist will never be caught using real inputs at all, precise or not. A Psychologist will tell you 86% of people taking treatment X recovered. But miss out that 85.8% of people on placebo also recovered. Those numbers will be from a study of 10 people.
mike_hearn 3y ago
https://en.wikipedia.org/wiki/Mathiness https://en.wikipedia.org/wiki/Mathiness Mathiness is a term coined by Nobel prize winner economist Paul Romer to label a specific misuse of mathematics in economic analyses. An author committed to the norms of science should use mathematical reasoning to clarify their analyses. By contrast, "mathiness" is not intended to clarify, but instead to mislead. According to Romer, some researchers use unrealistic assumptions and strained interpretations of their results in order to push an ideological agenda, and use a smokescreen of fancy mathematics to disguise their intentions.
JHonaker 3y ago
The main problem is the extrapolation of the “small world” we construct with mathematics to the “large world” of an application. Your model can be completely internally consistent and allow you to study things in the context of your assumptions perfectly, but that doesn’t mean it doesn’t have critical flaws that disqualify any learning about the system you’ve modeled. The lack of understanding about what implications a representational choice have for the kinds of relationships a mathematical model can describe, and the lack of understanding concerning what are and aren’t critical domain components to model that’s difficult. People often forget that there are two translation steps that have to take place. Domain -> Math -> Domain. That second one is too often implicit or assumed to follow, especially in ML.