Body
In Quilicura, an urban edge area of Santiago, a group of about 50 residents spent Saturday trying to run a chatbot in a way they said avoided artificial intelligence entirely—by answering questions and handling image prompts through people working on laptops in a community center. Organizers behind Quili.AI said the 12-hour effort drew more than 25,000 requests from around the world, and they framed it as an experiment meant to shift attention from how quickly chatbots respond to the environmental costs linked to AI infrastructure.
One prompt aimed at producing an image of a “sloth playing in the snow” did not generate an instant output. Instead, the Quili.AI website told users in Spanish to wait a few moments and reminded them that a human was responding, before eventually returning a hand-drawn sketch—about 10 minutes later—showing a cartoon-like sloth in a pile of snowballs. Organizers said the demonstration was designed to make visible a part of the technology stack that most users do not see.
Organizers said the responses came from a rotating crew of volunteers, with people taking turns answering in real time. When asked by an Associated Press reporter who made the sloth drawing, the Quili.AI website said it was a local youth helping with illustrations. The website’s answers also reflected local knowledge, according to AP’s account: when users asked how to make Chilean sopaipillas, a fried pastry, volunteers who did not know the answer said so and then looked around the room to find someone who did.
Lorena Antiman, representing the environmental group Corporación NGEN, said the goal was to highlight the “hidden water footprint behind AI prompting” and to encourage “more responsible use.” She also said Quili.AI was built around the idea that not every question needs an immediate answer, adding that the project aims to encourage a different kind of interaction—one in which residents can admit uncertainty, share perspective, or respond with curiosity rather than certainty.
Antiman said the campaign was not designed to reject what she called the “incredibly valuable” uses of AI. In her remarks, the project centered on considering what she described as the impacts of so much “casual prompting,” specifically in places like Quilicura that experience water stress. The volunteers’ slower, human-paced responses were part of that messaging, organizers said, because they forced users to notice the difference between instant chatbot outputs and the labor and resources behind them.
The demonstration comes amid a broader debate in Chile and elsewhere about the heavy costs attributed to AI data centers. Computer chips that run AI systems require large amounts of electricity, and some data centers also use significant volumes of water for cooling, with usage varying depending on location and the type of equipment, according to AP’s report. The Santiago region has attracted data center development from major cloud providers, including companies such as Amazon, Google and Microsoft.
AP reported that Google has said a Quilicura data center it switched on in 2015 is the “most energy efficient in Latin America,” pointing to investment in wetlands restoration and irrigation projects in the surrounding Maipo River basin. The company has also faced legal challenges tied to water usage concerns for another project near Santiago, adding to the pressure around how environmental costs are assessed and managed.
Chile has faced a decade of severe drought, a factor experts say contributed to the spread of recent deadly wildfires. In that context, the Quili.AI project used a local, human-run chatbot event to bring the water-footprint discussion directly to residents—using their own questions, delays, and local knowledge as the demonstration content.