AI development causes environmental collapse and labor exploitation – while AI itself is already environmentally harmful through infrastructure networks that require unethical mining operations and endless pollution regimes – it also causes a rebound effect in which its proliferation multiplies its own impacts since its increased capabilities are utilized by businesses that are environmentally harmful
Hagendorf 21 (Thilo, AI researcher at the University of Tübingen. 11-25-21, “Blind Spots in AI Ethics,” Orcid, https://doi.org/10.1007/s43681-021-00122-8)-qcl
Ecosystem services do not only have an unprecedented monetary value for humans [164], but they are the very reason for the possibility of human life on earth. Over-exploitation of ecosystems harms future generations [165], poor and already underprivileged people [166], animals [167], and many more. Whereas the building industry, agroindustry, or transport industry seem to be main drivers of ecosystem destruction and climate change, the information and communication industry also play a tangential role. This holds AI and Ethics 1 3 especially true for the AI field. The term “AI” bears a linguistic similarity to the term “cloud”. The notion of cloud computing suggests that it lacks materiality, that it is invisible and placeless, that data are “stored in the troposphere” [168], where in fact, big data is anything but transcendent or amorphous. It is grounded in fragile physical infrastructures, cables, hardware, routers, server buildings, power grids, cooling systems, satellites, etc., all of whom require natural resources. A similar situation unfolds with AI. The term “artificial intelligence” again suggests something immaterial, a mental quality that has seemingly no physical implications. This could not be farther from the truth. To appreciate this, one must switch from a data level, where the real material complexities of AI systems are far out of sight, to an infrastructural level and to the complete supply chain. Here, and frst of all, it becomes visible that the cloud and AI systems are intrinsically intertwined. The cloud builds the necessary condition of AI, and the material implications of the cloud cling onto AI, too. These implications are far-reaching and manifest themselves in global networks of cable infrastructures, labor division, logistics, distribution, and manifold externalities. These networks comprise lithium, tin, cobalt and other mines that deliver essential minerals for electronic components or batteries that are part of every digital mobile device, smelters and refiners that produce acidic, radioactive, and otherwise harmful waste products, storage systems and warehouses for logistics and transportation operations, energy and water hungry data centers, afliated cooling systems, diesel powered generators for backup purposes in cases of blackouts, data annotation factories, collection operations for toxic electronic waste consisting of technical devices with a lifespan of a few years, and many more [2, 169]. All these mining, shipping, manufacturing, and garbage incineration operations are heavily destructive, have a high burden on ecosystems, and come at the cost of human lives, child labor, wildlife populations, natural habitats, toxins in the ground, water, and air, public health, political instability and tensions and low wage labor markets. The material conditions that allow AI usage, especially rare earth elements or “confict minerals”, are also triggers for military operations, violence, murder, and migration that surround the already brutal and slavery-like industry of mining [170]. AI systems are not just demanding in terms of material resources, they also require a lot of energy. Electronic machines, in contrast to combustion engines, can in principle be used sustainably by consuming electricity from renewable energy sources. In practice, however, in many countries, only small proportions of electricity are renewable [171]. Accordingly, powering the computational resources that are required to collect large amounts of training data and to train, test, and apply large AI models comes with a signifcant carbon footprint [172]. Strubell et al. [173] conducted a life cycle assessment of several large AI models and found out that they can emit around three hundred thousand kilograms of carbon dioxide equivalent. The reason for this lies in the many ways machine learning methods go along with a “bigger is better” approach which prioritizes accuracy over efciency [174] with costly trial and error processes which span from practitioners intuitively setting up model parameters all to neural architecture search and other tuning and automated optimization processes. Ultimately, the information and communication industry, which incorporates the AI feld, has a carbon footprint that is bigger than that of the aviation industry [175]. But while there is fight shame, there is no such remorse for AI use, although some AI ethics researchers have tentatively started to develop a critical perspective on the role that AI has in contributing to climate change [174, 176, 177]. Much of this is dependent on where training servers are located, which energy grid is used, how long models are trained, and what hardware accelerators are in use [178]. However, even under perfect conditions where only renewable energy sources are used, it seems likely that AI remains a polluting technology in many industry sectors due to the business purposes it is utilized for. On one hand, AI technologies are heralded as technical solutions to the climate crisis by helping to develop lowemission infrastructures, operate smart grids, help foster sustainable consumption and production, etc. [21, 179]. On the other hand, AI technologies are used to buttress industries and business models that are environmentally harmful—let alone rebound efects in industries that are deemed to be sustainable [180]. In this regard, the tip of the iceberg is the collaboration between the largest AI companies and the fossil fuel industry [181, 182] which does not only comprise the optimization of oil and gas extraction, but goes so far as to actively support climate change deniers [183].