Google Whistleblower: ‘Machine Learning Fairness Is and Has Always Been the Real Censorship Program’

VORHIES Make no mistake ░░░░░░░░ learning fairness is and ░░░░░░░░ always been the real ░░░░░░░░ program and it is ░░░░░░░░ The goal was to ░░░░░░░░ program the public to ░░░░░░░░ with Google s corporate ░░░░░░░░ those are their words ░░░░░░░░ was a 4 step ░░░░░░░░ laid out by the ░░░░░░░░ ethicist Margaret Mitchell who ░░░░░░░░ since been fired for ░░░░░░░░ behavior Step 1 training ░░░░░░░░ are collected and classified ░░░░░░░░ 2 algorithms are programmed ░░░░░░░░ 3 media are filtered ░░░░░░░░ aggregated or generated And ░░░░░░░░ 4 people like us ░░░░░░░░ programmed That s a ░░░░░░░░ quote from their slides ░░░░░░░░ wasn t just in ░░░░░░░░ slide it was littered ░░░░░░░░ the company This process ░░░░░░░░ repeated in a cycle ░░░░░░░░ step 4 feeding back ░░░░░░░░ step 1 This sounds ░░░░░░░░ something out of a ░░░░░░░░ theory but it s ░░░░░░░░ Google rewrote their news ░░░░░░░░ specifically trained on mainstream ░░░░░░░░ stories targeting Trump such ░░░░░░░░ his fight with Comey ░░░░░░░░ like real time events ░░░░░░░░ time boost and hive ░░░░░░░░ assigned higher amplification scores ░░░░░░░░ stories related to targeting ░░░░░░░░ Google s internal documents ░░░░░░░░ their stance on quote ░░░░░░░░ unfairness They stated that ░░░░░░░░ factually accurate representations could ░░░░░░░░ considered algorithmically unfair and ░░░░░░░░ Let me just quote ░░░░░░░░ For example imagine a ░░░░░░░░ image query for CEOs ░░░░░░░░ predominantly men Even if ░░░░░░░░ were a factually accurate ░░░░░░░░ of the world it ░░░░░░░░ be algorithmic unfairness In ░░░░░░░░ cases it may be ░░░░░░░░ to take no action ░░░░░░░░ the system accurately reflects ░░░░░░░░ reality Well in other ░░░░░░░░ it may be desirable ░░░░░░░░ consider how we might ░░░░░░░░ society reach a more ░░░░░░░░ and equitable state via ░░░░░░░░ product intervention or broader ░░░░░░░░ social responsibility efforts end ░░░░░░░░