# Data Harvesting的所有文章

You've Been Training Google's AI for Free for 15 Years, Completely Unaware

For 15 years, Google has leveraged reCAPTCHA to harness free human labor to train its AI, unbeknownst to users. Initially created to digitize books by having users transcribe distorted text, the system evolved under Google's ownership. With reCAPTCHA v2, users were tasked with identifying objects like traffic lights and crosswalks in images from Google Street View. This provided massive, free training data for Google's computer vision models, directly benefiting products like Google Maps and the autonomous vehicle company Waymo, valued at $45 billion. At its peak, 200 million reCAPTCHAs were solved daily, amounting to 500,000 hours of free human labor—worth an estimated $5 million per day at minimum wage. This data-labeling operation, embedded as a mandatory gateway to essential websites, was unparalleled in scale and cost-efficiency. The latest version, reCAPTCHA v3, invisibly analyzes user behavior to verify humanity, further feeding AI systems. The profound irony is that users spent years proving they were human by performing tasks AI couldn't do, thereby training the very systems that now make their contributions obsolete. Google never asked for consent, paid for this labor, or disclosed its purpose, turning the entire internet-using population into unwitting, unpaid trainers for its commercial AI empire.

marsbit03/18 09:32