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

VORHIES: “Make no mistake, machine learning fairness is and has always been the real censorship program and it is massive. The goal was to, quote, ‘program the public to align with Google’s corporate values,’ those are their words. This was a 4-step process laid out by the AI ethicist Margaret Mitchell who has since been fired for unethical behavior. Step 1: training data are collected and classified. Step 2: algorithms are programmed. Step 3: media are filtered, ranked, aggregated or generated. And step 4: people like us are programmed. That’s a direct quote from their slides. It wasn’t just in one slide, it was littered throughout the company. This process was repeated in a cycle with step 4 feeding back into step 1. This sounds like something out of a conspiracy theory, but it’s real. Google rewrote their news algorithms specifically trained on mainstream media stories targeting Trump, such as his fight with Comey. Systems like real-time events, real-time boost, and hive mind assigned higher amplification scores to stories related to targeting Trump. Google’s internal documents revealed their stance on, quote, ‘algorithmic unfairness.’ They stated that even factually accurate representations could be considered algorithmically unfair and removed. Let me just quote them: ‘For example, imagine a Google image query for CEOs shows predominantly men. Even if it were a factually accurate representation of the world, it would be algorithmic unfairness. In some cases, it may be appropriate to take no action if the system accurately reflects current reality. Well, in other cases, it may be desirable to consider how we might help society reach a more fair and equitable state via either product intervention or broader corporate social responsibility efforts,’ end quote.”