AI and Sustainability

Digital tools and services powered by artificial intelligence (AI) have rapidly evolved from the realm of experimental technologies to readily available consumer products. Chatbots, digital assistants, and data analysis tools have all become interlaced with AI and its unique suite of features. This swift and coordinated effort to integrate AI into the digital ecosystem has raised concerns regarding the extent of its energy and human costs. As our global society attempts to address the present burdens associated with climate change, the emergence of AI presents a significant curveball in the sustainability equation. With the presence of AI generally expected to expand in future years, it becomes increasingly important for those concerned with sustainability to understand the nature of its costs and benefits, along with where it stands amongst other pressing sustainability concerns.

Environmental Concerns with AI

One of the more prominent concerns with the proliferation of AI is the costs it will place unto energy grids, the environment, and communities where the industry is most present. The United Nations Environment Programme notes increasing electronic waste, use of exceedingly scarce water resources, mineral demand for computer components, and electricity to power growing AI infrastructure as some of their foremost concerns on the matter. The cost of increased electricity consumption is corroborated by professionals at the Massachusetts Institute of Technology, who write that the collective electricity demand of AI data centers outpaces entire nations and will soon approach the usage of some of the globe’s economic powerhouses.

The energy usage of generative AI, while staggering when viewed in its totality, consumes at a variable rate and is worth recognizing in all its complexity. Large fluctuations in energy use occur during the training period of generative AI models, which can often stress power grids to the point of activating diesel-powered generators to meet these electricity demands. The frequent iterations of updated AI models, which can occur in mere months or even weeks, compound these costs by rendering previous models, and the energy required to train them, obsolete. So while the present energy demand for AI queries bears an energy cost five times greater than that of a standard web search, current forecasts indicate that this is set to increase with the introduction of new models. As the use of AI for data processing and inference tasks becomes increasingly commonplace, concerns amongst research scientists and professionals continue to mount as these models grow more sophisticated and require additional energy supplies to satiate them.

The environmental concerns with generative AI include:

• Energy use from data centers and training AI models
• Water use from cooling data centers
• Mineral mining for critical minerals to build hardware
• Electronic waste from hardware, some containing hazardous substances
• Environmental inequities as vulnerable populations are unfairly burdened with consequences from AI expansion
• Security risks to energy systems as AI capabilities increase the risk of cyberattacks on energy infrastructure
• Rebound effects when increased AI efficiency leads to more use of AI
• Spreading misinformation about climate change and environmental issues due to the proliferation of convincing AI-generated content that is anti-climate change

Using AI to Promote Sustainability

Proponents of artificial intelligence note that there exist several benefits presented by AI which may assist in addressing present and future environmental concerns. The Columbia Climate School states that the current task-based AI we possess can sometimes perform better than humans in certain pattern recognition tasks, such as weather forecasting, so long as it is trained with a sufficient level of information. This, in the eyes of the United Nations Environment Programme, indicates that AI can play a crucial role in assisting the development of more advanced emissions models. This, in theory, could also extend to AI being deployed to measure environmental footprints of various industries and climate metrics such as changes in glacier mass and sea level. By furthering the extent to which climate data can be collected and accelerating the rate at data analysis, AI tools have the potential to present clearer data that informs future research efforts and global decision making.

The use of these tools for enhancing climate data, while beneficial in many regards, does not come without its own associated costs and growing pains. The relative novelty of AI use in data analytics and research means there exists something of a disconnect between the expertise of environmental scientists and actual AI professionals. To better utilize AI in mitigating environmental challenges, further collaboration between the two must be explored first. The environmental costs of sprawling AI data centers will also continue to be of concern, leading to inevitable discussions on the environmental cost to benefit ratio of AI usage in its entirety. Discussions on how AI may be used in a limited and ethical capacity may be on the horizon, but the market for these technologies are presently in a relatively unregulated situation which makes future predictions challenging.

Generative AI can potentially be used to support sustainability goals in the following ways:

• Making energy grid management more efficient. Power companies are now using generative AI to optimize energy distribution, minimize wastage, and integrate renewable energy sources more effectively.
• Improving energy efficiency in homes. One estimate found that AI-run smart homes could reduce a household’s carbon dioxide generation by up to 40%.
• Improving the reliability of renewable energy systems. AI can be used to optimize facility layouts, monitor equipment, stabilize power grids and predict climate patterns, making renewable energy systems more tenable.
• Optimizing resource exploration and production. Generative AI has the potential to lower the carbon footprint of companies who produce natural resources by determining ways to improve efficiency and reduce waste during resource exploration and extraction.
• Decarbonizing transportation. Autonomous electric vehicles and optimized logistics systems have the potential to significantly reduce emissions.
• Discovering new solutions to climate change. Generative AI can simulate weather patterns, improve precision agriculture, track icebergs, identify pollution, chart greenhouse gas emissions, and create better predictive models for natural disasters.
• Facilitating scientific research and innovation. Generative AI can help scientists explore new possibilities and craft models. For example, researchers are currently using generative AI to design more sustainable materials, such as solar photovoltaic materials.

AI and Sustainability Policy and Research Priorities

Generative AI experts agree on a number of policy recommendations and research opportunities to mitigate the environmental impacts of generative AI.

• Develop environmental regulations for generative AI. This includes developing standardized procedures for assessing environmental impacts, creating reporting frameworks for tech companies, and setting benchmarks for energy and water use.
• Mandate transparency in the accounting of the environmental impacts of AI. Many companies are not forthcoming with the energy use, water use, and other environmental impacts of their generative AI models, making it difficult to understand how tech companies are using our shared resources and impacting our planet. Reporting frameworks that do exist are voluntary, so we need mandatory reporting systems to ensure we are getting critical information.
• Improve the efficiency of AI models and data centers. This includes designing more efficient models and algorithms to accelerate AI training and reduce model sizes; improving hardware design to accelerate computing processes; and improving data center power and cooling infrastructure.
• Explore solutions beyond efficiency. Many tech companies are focused on improving the energy and water efficiency of their generative AI training models and data centers. However, this has the potential to generate rebound effects, which would increase our reliance on computation and thus the environmental impact of AI. Therefore, research should explore how to decrease generative AI’s environmental impact beyond just improving computing and technological efficiency. 
• Curb the use of generative AI to accelerate fossil fuel extraction. Technology companies’ partnerships with the fossil fuel industry and reliance on fossil fuel infrastructure will prolong our reliance on fossil fuels. Steps must be taken to mitigate this and ensure that generative AI is being utilized to promote sustainable resource use instead. 
• Address the environmental inequalities of generative AI. Identify which groups and locations are likely to be most affected by the environmental impacts of generative AI. Protect areas that are prone to adverse environmental impacts, such as drought, electricity grid overload, pollution, and depletion of freshwater resources. Include relevant stakeholders in decision-making processes, such as local and Indigenous communities.

Sustainable Personal Use of AI

The meteoric speed at which AI has integrated into digital ecosystems and internet services speaks to its staying power in a world still adjusting to its presence. Inevitably, this has raised concerns amongst the public regarding how to use these tools in a sustainable fashion. There are many instances in which one can forego AI altogether, though many digital tasks have become so thoroughly entangled with this technology that avoiding it can prove tedious in even the best of circumstances. Many of the most prolific consumer AI tools such as Google Gemini, Microsoft Copilot, and Apple Intelligence are encouraged for casual use and thus do not have clear instructions for how to moderate or turn them off. While some workarounds do exist that allow users to circumvent the use of these otherwise omnipresent AI features, the processes can often be convoluted, and such methods are often quickly patched in updates to devices and internet browsers.

For an in-depth guide and how to disable certain AI tools from your digital space, visit this Consumer Reports article.

Less frequent use of these AI features often comes with the added benefit of increasing device storage space while decreasing power usage and computational strain. Though as previously mentioned, mainstream digital platforms have continued to solidify the presence of AI in their infrastructure, thus making it more difficult to curate your experience on their platforms. In the case of Google’s search engine, it is possible to switch to “Web Mode” in the quick settings beneath the search bar, though this must be done manually and seemingly does not mitigate the energy cost of AI search queries. This remains true of other services, which offer more classic experiences, but often only cloaks the presence of AI which continues to run in the background. A more sustainable option would be to utilize alternative internet browsers that have more customizability in terms of AI experience and stronger sustainability commitments.

There are also many instances in which AI tools cannot be ignored, especially when it comes to completing certain work tasks. The professional applications of AI have existed for longer than the consumer models that recently debuted to the public, and can be efficiently used for pattern recognition tasks, writing code, and the interpretation of large datasets that can otherwise prove difficult for researchers to process on their own. Given this history of use and the growing scope of AI assistants and tools, determining whether the utility of AI exceeds traditional task management can be challenging to do. These decisions should account for the efficiency of using AI in accordance with the scope of the task at hand. General writing tasks, whether they be work emails, essays or articles, can each be accomplished independently in lieu of defaulting AI. Image generation can also be easily avoided and is often a poor use of energy when considering the already impressive library of art and stock images on the internet. Determining whether AI is providing mere convenience or genuine assistance in a task can often guide the decision to utilize it or to forego it entirely.

Sustainable Internet Browser Alternatives

• Brave – Privacy focused browser with optional AI features
• Ecosia – Uses smaller AI models and engages in many sustainable practices
• DuckDuckGo – Focused on privacy and ad-blocking with optional AI features
• Firefox – Compatible with Google products and has optional AI features