THE UNITED Nations Environment Programme (UNEP) has noted the potential of artificial intelligence (AI) to help to solve global environmental challenges, while also cautioning of the need to carefully manage the risks, for people and the planet.
This is reflected in an Issues Note on the environmental impacts of AI, published by the UNEP last month and which champions the need for the comprehensive assessment of the full AI life cycle.
According to the issues note, AI offers “transformative opportunities for the environment”, with the ability to play an essential role in the triple planetary crisis of climate change, nature loss and pollution.
“However, AI’s rapid pace of innovation, exponential growth and demand for resources have raised significant concerns about its potential negative impact on the environment. As AI has evolved, its demand for larger datasets and increased computational power has grown, with associated pressures on natural resources,” the document said.
Also important, the issues note explained, is the need to take stock of the two life cycles of AI – the software and the hardware.
“When assessing the environmental footprint of AI, it is important to examine both life cycles. Until recently, the environmental impact of the AI software life cycle was thought to be mainly in the training phase due to the high data and computational load. However, with generative AI and large-language models, the inference stage now requires equal or more resources,” it said.
“The AI hardware life cycle is much more complex and difficult to assess, as it requires a thorough analysis of every stage of the process. From mining and extraction practices to transportation methods, to water and energy consumption, all the way to waste and e-waste generation, with each stage having a different environmental footprint,” it added.
“While the software and hardware life cycles are inextricably linked, it is important to measure and seek to mitigate the effects of both life cycles on all aspects of the environment,” it said further.
It is against this background, together with other identified issues, that the UNEP has made a series of recommendations to help to tackle the AI risks to the environment. They include the pursuit of further research to determine “the indirect and higher-order impacts, including possible negative impacts from overconsumption”.
Other recommendations include that countries establish standardised methods and metrics for measuring AI’s environmental impacts, with the immediate priority being “the focus on the most concerning direct effects, and those for which data is more accessible, namely, consumption of energy, water and mineral resources, and the production of emissions and e-waste”.
Also included in the recommendations is that countries, with UNEP’s support, develop mechanisms and frameworks for mandatory reporting and disclosure of AI’s direct environmental impacts by companies offering AI products and services.
“UNEP, together with member states and the private sector, [should also] agree on ways to make the above metrics transparent and accessible, to empower end users to understand their environmental impact and make informed decisions, promoting behavioural changes towards more sustainable AI use – similar to individual carbon footprint calculations and offsets in aviation,” it also said.
This is in addition to, among other things, a recommendation for the private sector and research community to prioritise research on optimising AI algorithms for energy efficiency by reducing computational complexity and data usage, thereby minimising the energy required for AI models.