This is typical of politicized reporting that really really wants to push diversity. First, the article says Many people think they know what the founder of a

This is typical of politicized reporting that really really wants to push diversity. First, the article says Many people think they know what the founder of a tech start-up looks like: a 20-something man who spent his childhood playing on computers in his basement and who later dropped out of college to become a billionaire entrepreneur. And then refute this with the more "diverse" reality that The average founder is 38, with a master’s degree and 16 years of work experience. Note no mention of the gender of the average founder. Next, they consider the study by Bloomberg Beta that the article implies refutes the "myth in our heads of what the prototypical start-up founder is, and that myth is an early- to mid-20s white male who studied computer science at an elite school and dropped out" But the methodology of the study is to create a machine learning model that uses the "backgrounds" of people (which seem to include which companies a person worked for, and I would guess also which school) to predict which people will become a founder. They then define the people who the machine learning model scores highly, as "potential founders" and examine the demographics of potential founders. The results quoted are: While only 12 percent of current founders are women, when they searched for potential founders based on the other characteristics of successful founders, 20 percent of the people they found were women. “If you look at just the professional histories of the people who got funded, then it suggests people who share those histories are much more diverse than the people who get funded,” Mr. Bahat said. Only 53 percent of founders have technology backgrounds, indicating that a computer science degree is not a requirement. Of potential founders, 8 percent of those with technology backgrounds are women. So in the strange logic of the author, the machine learning model's validity is not called into question even when the difference in gender bias between "potential founders" and actual founders is mediated by highly relevant information, such as a technology background. To be specific about the problem with this logic, if the machine learning model selects a completely different looking population (ignoring gender) from existing founders, that is evidence that the VCs are biased, but it is also evidence that the machine learning model is wrong. The high level problem with this approach is that the same variable they are trying to show is biased (being an actual founder) is the one they are using to train the model which selects "potential founders" in order to prove the bias.