Comment by VBprogrammer
7 years ago
Wouldn't it be easier to create an algorithm for compensation which clearly didn't involve gender (i.e. no deep learning stuff). Something which boils down to base salary + experience factor + google service factor.
Honestly, I can imagine that being less troublesome even outside of attempting to be fair an equitable regardless of race / gender / sexual orientation.
Blind hiring can have the exact opposite effect:
https://www.abc.net.au/news/2017-06-30/bilnd-recruitment-tri...
My point is - there's zero guarantee that a "logical" algorithm would be "fair". Just like if you trained an ML algorithm on the database of people currently in American prisons it would most likely conclude that black=more likely to be a criminal, which is obviously an unfair assumption to make.
doesn't that mean that women have already artificially high wages? Genuinely curios
The problem is that it can turn out men are more aggressive when it comes to getting counter offers which has a much larger impact than anything else. If that's the case then what do you propose, ban people from negotiating and lose everyone good enough to get competing offers?
In short; yes. I think it's fair to say that google aren't going to run out of engineers if they implemented a completely rigid payment structure. Firstly, their pay is very high compared to the industry and secondly they attract the best engineers by reputation more than anything else.
What happens in a few years when the best engineers start leaving because they can negotiate higher pay elsewhere? If they start to get a reputation as merely good, but not the best, not where the exciting work happens?
1 reply →
> Firstly, their pay is very high compared to the industry
As evidenced by the lowball they just tossed me.
How do you quantify experience? It cannot be empirically measured in all the ways that it matters.
Do you think even the most basic measure, say number of years, is any worse than the current process?
Of course! Have you never met a programmer who had ~20 years of experience but actually was extremely weak at programming? Even the most basic of interviews would disqualify them from the job, but an ML algorithm can give them huge points on the "experience" alone.
3 replies →
I do, some people can do the same job for decades and somehow never improve their skills or personally develop.
I'm not just interested in pure technical skills from somebody, I want somebody who has the common sense and risk aversion to keep their projects running steadily and effectively without needing to put out fires weekly.
I also appreciate people who have something that I like to call "applied laziness", that being an aversion to repeating the same task over and over or building things that will require constant attention down the line. People with this skill are happy to put as much effort in as necessary in the short term to ensure that they don't have to expend needless effort in the future.
> Wouldn't it be easier to create an algorithm for compensation which clearly didn't involve gender (i.e. no deep learning stuff). Something which boils down to base salary + experience factor + google service factor.
You mean, the kind of thing basically every public sector employer has, where alleged gender pay gap is never an issue?
Yes, of course that’s easy to do.
OTOH, private sector employers like to have personal productivity be a factor, but don't have good objective metrics for that, so they let subjective assessment play a major role, such that the biases of managers become a substantial factor in setting pay.
I think this is part of the perception that machine learning is magical. This problem is as difficult as getting an accurate IQ.
If you’re going to base pay- something really serious- on such an algorithm, it must be open and reproducible.
Google is really good at this and that it doesn’t exist in a useful way is a signal that there is no algorithm. It would be extremely valuable to companies to know this like a credit score.
I think when there are natural incentives, resources, and open problem it means that we don’t really have resources or doesn’t exist and needs more time and innovation to solve the problem.
That's why I specifically excluded deep learning stuff. I was imagining something much simpler only marginally more complex than title base salary + x * years experience + y * years at google. Where the factors are initially chosen to get close to the current state.
How about standardised bands for competency? E.g. a software engineer level 3 step 2 has mentored three people, been the tech lead/written the OKRs for a 6-month project, and has X years of experience on other projects (in the earlier levels)?
Then it would be a simple HR policy to assign that engineer to whichever compensation band covered those levels when they joined a new company.
Amazon used software to try and hire people without discrimination, instead they unintentionally discriminated against women https://www.businessinsider.com/amazon-built-ai-to-hire-peop...