No proof, but I have a feeling its more of a way around regulations for most tech companies. For example, in hiring, IQ tests are known to have predictive value but they're basically banned because they discriminate against women and minorities.
So you build a grey box neural network that analyzes resumes to find "good" candidates. Underneath it might draw correlations between female or minority names. It might decide that people who lived in certain places or came from country X are "bad employees". Since nobody can explain how it works once you've trained it, you're free to build it using whatever biases you want.
Same with financial regulations. The network could rely 100% on insider trading. Analyzing the HTML of websites for leaked information that isn't visible to humans yet. Again nobody knows this is how it's so good at taking in bytes at one end and buying stocks on the other.
So you build a grey box neural network that analyzes resumes to find "good" candidates. Underneath it might draw correlations between female or minority names. It might decide that people who lived in certain places or came from country X are "bad employees". Since nobody can explain how it works once you've trained it, you're free to build it using whatever biases you want.
Same with financial regulations. The network could rely 100% on insider trading. Analyzing the HTML of websites for leaked information that isn't visible to humans yet. Again nobody knows this is how it's so good at taking in bytes at one end and buying stocks on the other.