I studied computer science and took a few machine learning classes.

[davidculley@hachyderm.io]

RE: https://dair-community.social/@timnitGebru/116270042818651080

I studied computer science and took a few machine learning classes. Naturally, every introductory class teaches you about logistic regression, a relatively simple machine learning modeling technique.

Today I learned that its "inventor", Karl Pearson, who was active about 100 years ago, was a eugenicist.

A brief explanation:

The simpler method, linear regression, let's you predict a continuous variable. For example: Knowing that a man weighs 100 kg, how tall do you think he is? You predict a man's height given his weight.

Logistic regression does not let you predict a continuous variable such as height. It predicts the probabilities of categories. For example: Given a man weighs 120 kg, would you say he is obese or not obese? A man of height 170 cm would surely be classified as obese. But a man of height 210 cm would not. But given that there aren't that many men taller than 2 m, logistic regression would say something like "Given the weight 120 kg, I'd say it's 90% likely that he is obese and 10% likely that he isn't."

Pearson's mentor, Francis Galton, had used similar methods to measure the attractiveness of African and European women. Both Galton and Pearson were deeply eugenicist and racist.

Galton was a cousin of Charles Darwin, btw.

Pearson contributed many more methods that are now fundamental to statistics:

- the histogram
- the p-value
- the chi-squared test
- Principal Component Analysis

https://en.wikipedia.org/wiki/Karl_Pearson#Contributions_to_statistics

Check out the Wikipedia pages for Pearson and his mentor Galton.

> Pearson saw war against "inferior races" as a logical implication of the theory of evolution. […] He reasoned that […] the nation is wasting money when it tries to improve people who come from "poor stock".

I am once again saddened that university teaches you all the math about these methods but nothing about the people or the social context behind them, and that today is the first time I hear about any of this.

Even with the benefit of hindsight this connection between fundamental statistical methods and vile ideologies isn't clear to most people who further Pearson's statistical legacy.

The idea that human intelligence can be measured, which is so common in AI enthusiasts and normalized, is deeply rooted in eugenics.

#TIL