Hacker News Reader

Comment by bitL | Hacker News Reader

It should be trivial to measure performance these days and figure out if there indeed is/isn't any biological difference or if certain parts of DNA/cultural background etc. are the decisive factors. It's Google anyway, "The ML company". They should internally test this hypothesis and then publish their findings (hopefully they can do objective studies with high integrity). If diversity programs cause them become less competitive, they will wither at some point; if diversity helps them they will earn economical success (unless all companies become alike). They have a perfect playground internally for that.

View on HN

And what if people don't like the result? If the result will be that there is a biological difference between the sexes will people accept the result or just start calling google a sexist and bigoted company and then google will suffer? There is a Norwegian documentary about biological differences but that didn't change anything.

Let's wait until some research is done. It's premature to say what the real result would be even if we can argue whether pure primitive biology-based behaviorism/evolutionary psychology will be more important than more gentle approaches the civilization is built upon. Maybe we will even gain an insight where the breaking points are.

What makes you say that it should be trivial to measure performance?

You have so many metrics these days as well as a complete view of the bottom-line. Google's ML is amongst the best, they have possibilities most people don't even get a hint of.

It's very much not trivial, though. Given two solutions to a coding problem, how do you objectively determine which one is better? More importantly, how do you objectively tell if either of them is good enough? Or if they're poor?

Probably wrong granularity; we would likely have to wait a bit for this detailed ML. But you can make various groups that follow certain "ideology" and see which ones thrive under what circumstances/in what kind of environment or not. Kinda like A/B testing but internally on their own coders. Then you can view e.g. maintenance costs (that might be your code quality metrics aggregated over time) etc. If there is one company that can pull this off, it's Google. I would be very interested in results to be honest. Then this could be generalized for governments and provide fact-based guidelines on how to govern well.

Unfortunately, how do you control that both groups are the same except for the metric you're trying to measure? And, once again, how do you accurately measure maintenance costs, especially over different projects that would likely have variations in requirements? And, most importantly, how do you come up with objective measurements for these things?

First, it's going to be an approximation and would take decades to carry out, as we humans are slow and no GPU/cloud would accelerate us :-( Second, the idea I had on mind was to use ML to figure out which signals/features ire the most important ones; just dump all raw data onto it and expect it would figure it out. Kinda like what you see from PCA which gives you base vectors guaranteeing some correlation properties, or CNNs where you expects neural network to come up with convolutions it needs on its own.

With maintenance costs I'd probably record as much data as I can and then try to find in a given case what kinds of signals were present. If I see some are repeating and are beneficiary or not, I can have some classifier from them. Also, it could unearth parasitistic relations when certain group of individuals flourishes only because they take advantage of certain individuals with different ethics, but never thrive in other kinds of environments.

Reply on news.ycombinator.com