HN user
The barber dishes out haircuts with average utility of X, where the average price for a X utils of haircutting is Y. I personally receive more than X utils from this guy’s haircuts. As a courtesy to compensate the additional value that I personally receive from him, which would not be fungible to a randomly chosen barber of equal skill based on anecdata, I provide a tip.
I like tipping my barber. His skill and care makes a substantial difference on the outcome. This feels like an appropriate place for tipping.
That’s backwards. If the problem is that the algorithm says healthy white gets treatment more than sick black; that means the algorithm has something else that it’s considering besides healthy and sick. If that’s the problem you’re solving for, you need to remove the input that creates the difference. Once it’s gone, the algorithm will be unable to produce a different result for two people with the same sickness level.
It is not possible to avoid racial bias. If you make a model predict equally for two classes, you’ve likely only made your predictions worse for both.
Prediction outputs come with a degree of confidence. It is much better to change how you evaluate your predicted probabilities for each protected class than to try and force the probabilities themselves to come out even.
The algorithm is rarely going to be the problem. It is the input data that is allowed into the model. If you have a problem with how healthy whites are getting recommended for treatment more than sick blacks, you need to remove the inputs that otherwise differentiate them.
I would say every business and the vast majority of individual people occasionally act as fluffers (as I believe I am to interpret that term) so sure.
My point is not to reveal a gotcha that McKinsey is not the word of God. You will find nobody at the firm thinks this way.
Many people have strongly negative perceptions of consulting firms, and I’m sure some people have strong anecdotes to justify their beliefs. Yet I’m brought in to address certain problems and it’s very clear that help is needed. Sometimes the solution is something that could have come from someone inside, or is perhaps something that someone on the inside is already saying beforehand, but the consultant gets communicated better to leadership because they have more political capital and credibility on the issue. If this happens, then you’re already experiencing some of the problems highlighted in this article. Political and organizational friction is real. If that is the “only” thing preventing an organization from solving a problem, then they are still incapable of solving the problem on their own. I think a lot of people would benefit from seeing how things are from the consultants point of view.
Current McKinsey, not associated with this report but kind of close. Yes, this one is very fluffy. There’s a few high level things that are important though around Org. Lots of companies just don’t have a good strategy for org, and there’s a lot of low hanging fruit for things like talent, culture, etc.