Thread · 4 tweets · 17 May 2021

But that's not the heart of the issue. The problem isn't ML, the problem is how it's used. The assumption is that it should replace user action rather than enhance it. The Gmail promotion tab is actually a great example of how to get this completely wrong.

If you have Clippy sort your mail, what's going to happen is that there'll be false positives but you're never looking in that folder because it's mostly crap. So either you have to slog through a folder of crap, or put up with losing some things you want. It's dumb.

A better way to do it would be to have Gmail's UI actually designed so it's possible to be fast when using it, and to make your own calls sorting email quickly. Clippy could add a little flag next to an email indicating likely promo — so you can kill is faster.

This, of course, would require there to be an actual UI designer involved in creating Gmail instead of a pile of engineering gimmicks in a product-like trenchcoat.