Data Ecology

Recently, the environmental impact of information and communications technology infrastructure has emerged as a critical issue. Data centers and their attendant network infrastructure are energy guzzlers, consuming anywhere from 1% to 5% of the world’s electricity, depending on how one calculates the estimate. In non-digital contexts, woodblock carving, mass printing, and climate-controlled archives also consumed energy, but humanists do not traditionally associate our research activities with energy use.

This research area encourages humanities scholars to acknowledge the ecological costs of digital systems and adopt more sustainable practices. Digital archives require continuous electricity for storage, processing, and retrieval. While small-scale use may seem negligible, energy demands escalate rapidly as data scales. A notebook computer may be powered with a solar panel the size of a backpack; hundreds of thousands of servers at a data center, however, demand megawatts of electricity and produce an insurmountable amount of heat and carbon emissions.

Energy efficiency and access to environmental resources have become keys to understanding the relationship between corporate interests and public institutions. Every major cloud provider has become deeply invested in energy production. Amazon, for instance, claimed in 2020 to be the “world’s largest corporate purchaser of renewable energy,” boasting 6.5 GW of capacity. By April 2021, Facebook reported it had achieved its 100% renewable energy target. China leads the world in renewable energy capacity, but most of that generation occurs in sparsely populated regions far from high-tech urban centers, requiring the installation of high-voltage direct current (HVDC) lines. We also investigate the environmental and geopolitical implications of mining rare earths and other critical minerals used in data centers and AI systems.

BDSL conducts global comparative studies of big data energetics that critically review scientific studies, corporate PR claims, and alarmist reports. We also examine cultural and regional differences, from the widespread district heating systems in Nordic countries to the persistent concerns in South Korea about the alleged carcinogenic emissions from data centers.

The advent of transformer-based artificial intelligence (AI) systems has further exacerbated the environmental impact of centralized, hyperscale infrastructure that our modern-day digital activities have depended on since the transition to Web 2.0. In response, BDSL follows the principles of lean and modular AI, marked by the separation of preprocessed knowledge bases and a compound system consisting of task-specific classifier, interpretive, and generative models. Rather than aspiring to build an artificial general intelligence for the humanities, we focus on targeted, efficient applications. BDSL’s flagship projects, DeepPast and Nabi X, are both developed according to our modular AI principles.