My takeaway: without short feedback loops it is hard to build expertise. If it takes 2 years to learn the individual you paroled reoffended, you can't learn (so there are virtually no expertise in parole officers or university admissions). In these scenarios, simple linear models outperform "experts" by a wide margin.
See: "In Praise of Epistemic Irresponsibility: How Lazy and Ignorant Can You Be?" by Michael Bishop
https://psycnet.apa.org/record/2009-13007-001
My takeaway: without short feedback loops it is hard to build expertise. If it takes 2 years to learn the individual you paroled reoffended, you can't learn (so there are virtually no expertise in parole officers or university admissions). In these scenarios, simple linear models outperform "experts" by a wide margin.
See: "In Praise of Epistemic Irresponsibility: How Lazy and Ignorant Can You Be?" by Michael Bishop
https://philpapers.org/rec/BISIPO