Reader TTS

Imagine we have a software development process, P. Let’s take the simplest case possible, and assume that P is a function of a _single_ human decision, H. (In actual fact, P will be a

function of _thousands_ or _millions_ of human decisions.) The output of P will be a model M. Let’s suppose there are two possible outputs, so that M will be either M0, which has bias B0, or

M1, which has bias B1. Suppose P will produce model M0 if and only if the human decision is H0. P will produce model M1 if and only if the human decision is H1. Suppose, by definition, that

the human decision will be H0 if and only if the human decision-maker has bias B0'. And the decision will be H1 if and only if the human decision-maker has bias B1'. Then, the

output model M will have bias B0 if and only if the human decision-maker has bias B0'. And, M will have bias B1 if and only if the human decision-maker has bias B1'.