causal

Type I vs Type II error in AI reading filter

An AI reading filter that's roughly 87 percent accurate is almost certainly a help rather than a hindrance, because one can always set the trade-off between type I and type II error—if it leaves in too much, the reader would have faced that content anyway, so there is no realistic scenario where it isn't a net help.

causalpending

Speaker

Tyler Cowen

Evidence Quote

“I think it’d be 87 percent accurate, and I’d get to read much more, and then I would apply my own rules to what’s left in there. If it leaves in too much, I would’ve been faced with that anyway. You can always set type I versus type II error.”

Source

Conversations with Tyler 2023 Retrospective— Conversations with Tyler
Created: 6/13/2026, 3:23:56 AM

Your notes

Loading notes...