Like the Rationalist's "Bayesian priors," the election models were a remnant of the "big data" hype from a decade and a half ago. This article is a decent overview for anyone who forgot about it[1]. Like with many hype cycles, there was something actually important underneath the surface (useful statistical modeling), but then people with a poor understanding of the limitations ran wild thinking it could do things far beyond its capabilities (in this case, the degree to which one could use statistics to predict the future).
Industry gave up on the more extreme claims fairly quickly because it wasn't able to produce. But it lingered on in other places where there was less direct feedback or it was telling people what they wanted it to hear.
To add to this, it became obvious that many of the leaders in this "field" were people who believed they had an expertise that was far beyond their actual capabilities. Nate Silver ended up accusing much of the polling industry of fraud recently, because he wasn't able to do basic statistical math[2].
Industry gave up on the more extreme claims fairly quickly because it wasn't able to produce. But it lingered on in other places where there was less direct feedback or it was telling people what they wanted it to hear.
To add to this, it became obvious that many of the leaders in this "field" were people who believed they had an expertise that was far beyond their actual capabilities. Nate Silver ended up accusing much of the polling industry of fraud recently, because he wasn't able to do basic statistical math[2].
[1] https://slate.com/technology/2017/10/what-happened-to-big-da... [2] https://x.com/JustinWolfers/status/1853302476406993315