Where in 2015 student rooms were still the place for experiments, there are now huge "super data centers" full of specialized chips all over the world. More than three hundred interviews with over two hundred and sixty people from the world of artificial intelligence show how quickly everything is changing in the book Empire of AI.
These data centers use enormous amounts of power. They now account for some 3 to 3.5 % of all carbon emissions. According to a recent report, the growth of these centers over the next five years could create an additional energy demand of 50 % to 120 % of the annual consumption of the entire United Kingdom.
Water resources are also coming under pressure. Two-thirds of new data centers are springing up in areas with little water. Data centers use pure drinking water to keep servers cool, putting drinking water supplies for residents at risk. In some regions, this even led to construction bans on new homes.
Yet the industry believes that scaling is the way to enable "general artificial intelligence. Big tech companies are investing billions in computing power. However, academic researchers show that compact AI models with hundreds of chips often perform just as well. But the standard remains, and it evidently remains because they want more and more and bigger clusters.
Behind the scenes, a large group of people work under very harsh and unsafe conditions. In East Africa, content moderators were hired for just a few dollars an hour to read and classify unfiltered, abusive texts ranging from hate speech to sexual abuse. Through daily exposure to violent and abusive messages, many suffered severe psychological symptoms similar to post-traumatic stress disorder (PTSD). In Latin America, refugees were attached to digital platforms where they had to claim new tasks at lightning speed via a loud alarm plugin as soon as a task became available, you then had to respond within seconds, otherwise it was with someone else. As a result, they worked day and night, running to their computers each time to avoid missing out on the next assignment and keep their income.
In South America, villagers managed for years to block data centers and enforce public participation; now schools, hospitals and municipalities are also setting up AI committees to decide for themselves what systems they use and under what conditions. In Suriname, too, we can tighten oversight by submitting objections to NIMOS against plans for data centers in water-scarce areas, refuse unnecessary cookies on websites to protect your data and set up a small AI advisory group within your organization or neighborhood, and involve technical experts, policy makers and end users to clearly define which AI applications are allowed and how they will be used responsibly.
