AI and Sustainability: the opportunities and the risks we face

  ^{INESC TEC}

  ^{INESC & Faculty of Engineering of University of Porto (FEUP)}

The tools and concept associated with "Artificial Intelligence" have gained significant prominence among consumers over recent years, as well as great business interest among investors and companies. However, and like any "El Dorado", the implications of these new solutions must be analysed and safeguarded - whether in terms of economic, environmental and social sustainability.

The dawn of significant advances in the transposition and improvement of mathematical algorithms framed within the broad concept of "Artificial Intelligence" (AI), namely with the current high computing capacities and large amounts of data for model training, allowed the emergence of multiple AI-based applications to solve complex problems (e.g., health, energy, industry, transportation, marketing, etc.). The success of those applications and developments led to two global effects: the interest of markets and business investors, and the wide dissemination of this new paradigm to consumers and citizens (namely through Large Language Models, popularised by the ChatGPT solution, by the company OpenAI).

The large investments of the global giants of the software industry are clear and we can say the same about hardware manufacturers (chips, large computing and storage systems). Still relevant, but significantly unbeknownst to the consumers, is the need to intensive consumption of electricity to cover computing power and related cooling systems’ needs, and to address the water used in the cooling infrastructures. Indeed, in this new "El Dorado" (as well as previous ones) there is a natural enthusiasm regarding the discovery and the search for "new mines" of opportunities and applications, towards hopefully the improving of regions and population. Hence, and still concerning the context of AI applications, the reader may ask: "aren't these AI or AI-based apps another type of software capable of automating processes through a dematerialisation supported by digitalisation and the Industry 4.0 paradigm?" The answer is "no"; since there are many pros and cons, and we should analyse this question through “sustainability mindset” approach of triple bottom line: economic, environmental, and social domains. This text does not seek and can’t provide "all the answers" to this analysis. But, it aims to emphasise, through a science-impartial lens, benefits-opportunities and disadvantages-risks, while presenting questions that favour a deeper scientific reflection and improve the citizens' critical awareness.

During each innovation and technological cycle, or even in our history regarding the connection between scientific-technological advances and innovation, there is the usual motivation to seek the capitalisation of results, potential impact in economics and market exploitation terms. Hence, the significant investments made in less than a decade around the new AI paradigm - the "current hype" -, the speculative risks, stock markets and business corrections that will follow (already clear while writing this article) come as no surprise. In this sense, considering economic and financial sustainability, and the broad application of AI solutions - whether in mobility-transportation (e.g., highly anticipated full-autonomous vehicles), industry, or other services -, companies must be careful about their investments and evaluate the potential returns e.g., productivity gains or cost reductions, versus technical (i)maturity, might be for internal processes, products, services or product-services.

In terms of environmental sustainability, natural resources management and pollution levels, it is becoming clear that there are several risks associated with the material, energy and water needs by the applications' operation - despite the benefits of using AI tools and techniques to improve and optimise processes, materials, products, and services (e.g., improved energy or material resources efficiency). We live in a world that currently faces a Triple Planetary Crisis, where the effects of climate change, pollution and biodiversity loss are already clear and tend to threaten the future development of humanity, other species with risk of extinction, and massive destruction of habitats and natural ecosystems. Despite several national or global programmes (notably the Kyoto Protocol 1997 or the UN Agenda Paris 2015) to mitigate the effects of global warming, we regularly witness phenomena like sequential record temperatures, greater frequency, duration and intensity of heat waves, storms and natural disasters. The goal of keeping global warming below +1.5ºC may be eminently compromised in view of the limited results of global decarbonisation of the economy [1]. In this unfavourable context, it is increasingly urgent to reduce the emission of greenhouse gases, namely CO2, to accelerate the energy transition to renewable sources, but also to act in terms of water resources management. Concerning energy, the proportionality between the computing capacities and the required power, the training of increasingly larger models (with a level of exponential computing requirements), and the mass use of AI tools (namely Generative AI) in professional or personal contexts, translates into higher electricity consumption. As an example, the set of data-centres, computational providers and data-transmission networks represent about 3% of the world's energy consumption, which means an annual CO2 emission equivalent to Brazil, and almost doubling electricity consumption over four years forecast - from 460TWh, in 2022, to 1000 TWh, in 2026 [2]. The energy level required is so great, that giant software companies are planning or already signing direct energy contracts with nuclear plants, hydro plants or other power plants, while reviewing their decarbonisation plans for potentially slower roadmaps towards carbon neutrality [3]. Considering freshwater consumption, the most recent data are worrying, due to the growing needs related to the cooling of data and computation-centres, but also due to the exponential demand for chips and, consequently, for greater amounts of purified water to feed the manufacturing processes - which compete with the supply of drinking water to the populations. As an example, safeguarding the fact that there are still few scientific studies in the area, recent estimates point to a consumption of 0.5 L of water (associated with the cooling systems of computation centres) for an interaction of 20-50 questions-answer with a Generative AI application [4].

Despite these risks - or disadvantages, if we consider the context and urgent need to mitigate climate change -, the advantages of AI tools cannot be ignored or overshadowed, also in favour of reducing resource consumption (energy, water and materials) in different domains: industry application, city management systems, public transportation, energy distribution networks, fostering and enhancing circular business models, industrial symbioses, waste management, etc. A question that we should consider is whether the application of AI tools and capabilities should focus on important society domains and needs, while informing other agents - namely, citizens - about a responsible and regulated use of AI based tools. In this sense, should the extensive use of AI application be controlled/regulated, or not? Especially since we haven't ensured sustainable energy consumption (with the elimination of non-renewable energy sources) and access to freshwater for all population.

Finally, and regarding social sustainability, there are different types of opportunities, like improving human skills (regular or reduced due to disease/disability), but also risks or threats. This article does not intend to thoroughly explore said threats, since Ethics is a quite sensitive area - addressed in other articles included in the Science & Society magazine; but we must mention the risks associated with a future where people, as intelligent individuals, have access to machines with similar (or higher) cognitive intelligence and decision-making levels, with the power to kill or control communities. When it comes to "less futuristic" aspects - even those related to environmental and economic sustainability - it's worth mentioning the fierce “competition” for resources among the infrastructures that support AI tools, leading to social pressure (namely in the poorest and drought-stricken territories), as well as the consequences in terms of employment due to a broad automation of tasks (many of them still quite specialised today). Thus, similarly to the past, this requires transformation/adaptation processes within organisations, focusing on their professionals and human dimension.

References