Beyond CO2 Emissions: The Overlooked Impact of Water Consumption of Information Retrieval Models

Proceedings of the 2023 ACM SIGIR International Conference on Theory of Information Retrieval

As in other fields of artificial intelligence, the information retrieval community has grown interested in investigating the power consumption associated with neural models, particularly models of search. This interest has become particularly relevant as the energy consumption of information retrieval models has risen with new neural models based on large language models, leading to an associated increase of CO2 emissions, albeit relatively low compared to fields such as natural language processing. Consequently, researchers have started exploring the development of a green agenda for sustainable information retrieval research and operation. Previous work, however, has primarily considered energy consumption and associated CO2 emissions alone. In this paper, we seek to draw the information retrieval community’s attention to the overlooked aspect of water consumption related to these powerful models. We supplement previous energy consumption estimates with corresponding water consumption estimates, considering both off-site water consumption (required for operating and cooling energy production systems such as carbon and nuclear power plants) and on-site consumption (for cooling the data centres where models are trained and operated). By incorporating water consumption alongside energy consumption and CO2 emissions, we offer a more comprehensive understanding of the environmental impact of information retrieval research and operation.