hgial

Everyone here seems to be certain the pandemic will have huge and lasting effects on educational outcomes for kids. I would instead bet that while there will be effects in the short term, these effects will mostly fade within a few years (preexisting gaps will remain, but they won't get much bigger).

This post reviews some of the existing evidence on the effect of missing school:

https://astralcodexten.substack.com/p/kids-can-recover-from-...

And generally finds that kids can catch up quite quickly.

I think my one caveat is that there may be a group of at-risk kids for whom school was keeping them from getting into serious trouble with the law. Those effects could conceivably be both large and lasting (e.g., if you carjack someone and kill them in the process and spend the next 15 years in prison it's going to have a big effect on your life!).

Forgive me if I don't consider "you can tell someone's race from physical features" to be quite as extraordinary a result as "particles can travel faster than light".

It's not "conspicuously absent from the paper". They have a whole group of experiments on this: "Experiments on anatomic and phenotype confounders" and conclude "Race detection is not due to obvious anatomic and phenotype confounder variables."

It might be helpful for folks to look at the blog post written by one of the authors:

https://lukeoakdenrayner.wordpress.com/2021/08/02/ai-has-the...

or the paper itself

https://arxiv.org/pdf/2107.10356.pdf

I see a lot of "oh it's probably just picking up on x y z" when x, y, and z are things they explicitly checked for:

1) "It's probably just the names or other metadata" – they only gave it pixel data to train on. To control for things like metadata overlaid on the image (e.g., a name written on the image) they divided the images into 3x3 sections and trained classifiers on each section separately.

2) "It's probably some artifact of how the hospital marked up the images" – they used something like 7 different datasets from different hospitals and different modalities (X-Ray and CT).

If it is cheating somehow, it's not doing it in an obvious way that you can think of in a minute or two. Also note that they had more than just medical folks working on the paper; the author list includes plenty of computer scientists. It's unlikely they're making an elementary ML mistake here.

At the moment crowd-sourcing ML training doesn't work very well because updates to the model's weights have to be shared between all of the nodes in the computation periodically. Giant models tend to be trained on clusters with very fast interconnects for this reason.