Last update 02/08/2023
AI & Environment: 5 Key Points to Remember
- AI can foster advancements in tackling current challenges about environment and energy.
- Nevertheless, numerous studies highlight the significant negative effects of AI on the environment, especially due to the energy and resources it consumes.
- The sustainable impact of an AI system involves three pillars: environmental balance, socio-economic contribution, and health impact.
- To minimize the environmental cost of AI systems, using short supply chains and reducing the required energy are recommended solutions.
- In its latest version dated June 14, 2023, the AI ACT has incorporated new environmental obligations for AI systems.
Introduction
Artificial intelligence has a complex relationship with the environment. Although it offers solutions to energy challenges, reports increasingly warn about its environmental impact. AI depends on infrastructures that store vast amounts of data, consuming significant resources and energy. How can we reconcile AI with environmental protection? This article explores these challenges and presents solutions and obligations from the new European regulation on Artificial Intelligence, which AI systems must comply with.
The Concept of Sustainable Impact and Measuring the Energy Consumption of an AI System
Sustainable impact must be taken into account when exploiting artificial intelligence. The white paper from Hub France IA Ethics identifies three sub-principles for including sustainable impact in the use of AI.
Environmental Balance
This refers to the environmental footprint of an AI system. It is calculated from the sum of direct, indirect, and induced environmental impacts due to its production, development, and use. Using an AI system involves various types of energy consumption. These include GPU (graphics processing unit) consumption, server consumption, and the consumption of equipment used to operate the servers. There is a trend towards initiatives to calculate the energy consumption and carbon footprint of an AI system. Green Algorithms and ML CO2 Impact are examples. Criteria such as resource consumption for the process or the overall power consumption of the data center are used to perform this calculation.
Socio-Economic Contribution
Socio-economic contribution is part of the sustainable impact of an AI system. The criteria evaluating this contribution include the system’s impact on collective well-being, the goal of complementarity with humans, and equal access conditions.
Health Impact
The last principle relates to the health impact that an AI system can have, for example, in research or the implementation of public health policies. In such cases, the AI system must adhere to ethical values as well as personal data protection.
This article will focus on the first principle concerning environmental impacts. The other two principles will be covered in other NAAIA articles.
The Impact of AI on the Environment
AI indeed has a significant ecological impact. Cédric Villani, a former French deputy and specialist in mathematical analysis, highlights the environmental impact of AI. He explains that the use of data storage infrastructures necessary for AI operation contributes significantly. This includes « pollution from mining operations, resource consumption, energy use, and even space. » Villani authored a report on implementing a French and European strategy in artificial intelligence, emphasizing these concerns.
Various researchers have looked into the subject and provide studies that help understand the extent of negative effects on the environment. A team from the University of Massachusetts Amherst, including Emma Strubell, stated in a 2019 study that some neural networks used in natural language applications, trained for 4 to 7 days, consumed as much energy as a human does in 57 years.
AI to Address Energy Challenges?
AI can sometimes serve as a solution to current climate and energy challenges. According to the UNESCO working paper on AI for sustainable development, AI can foster advances in ecological and biodiversity research, as well as ecosystem management. For example, AI can be useful in disaster prevention, as experimented in a Japanese city for detecting and implementing an early warning prototype in case of tsunamis. Freshwater management has also been efficiently monitored thanks to AI through statistical modeling.
Moreover, the ecological transition can benefit from AI to develop innovative and environmentally friendly solutions. For instance, AI can optimize the construction of renewable energy production sites by identifying the most favorable locations in terms of sun exposure, climate, or winds.
What Solutions for Responsible AI?
To limit the pollution generated by artificial intelligence, researchers recommend taking sustainable impact criteria into account. This involves minimizing the environmental cost of producing materials needed for AI systems. One approach is using short supply chains. For the development and operation of AI systems, reducing the necessary energy is a potential solution to minimize the associated environmental cost.
Furthermore, to offset the environmental cost of the production, development, and operation chain, it would be beneficial to promote uses that have a positive environmental impact.
According to Cédric Villani, « For AI to be responsible, responsible manufacturing, implementation, and control processes must be put in place. Responsibility lies with humans. » He thus recommends questioning « the purpose and architecture of algorithmic systems. This includes their transparency, ex-post evaluation, biases, and responses to environmental challenges. » Digital sobriety is also highlighted, which would consist of limiting AI uses through the adoption of good practices.
European AI Regulation: The Principle of « Social and Environmental Well-Being »
The European Parliament, in its amendments dated June 14, summarized in an NAAIA article, emphasizes one of the main principles of the AI ACT: social and environmental well-being. New environmental provisions have been integrated. In the newly created recital 46 bis of the AI ACT, the European Parliament states, « AI systems can have a significant impact on the environment and high energy consumption during their life cycle. »
The new recital 28 bis supports the idea of a « fundamental right to a level of environmental protection. » This consideration must factor in when assessing the severity of harm caused by an AI system, especially its environmental consequences. According to these amendments, an AI system may be deemed « high risk » if it poses a substantial environmental harm.
More than just high-risk systems, all AI systems will be, according to the new Article 4 bis of the AI ACT, « developed and used in a sustainable and environmentally friendly manner. »
The AI ACT thus intends to impose environmental obligations when implementing an AI system.
Although artificial intelligence can contribute to the energy transition, it presents numerous environmental challenges. Therefore, companies should integrate the concept of sustainable impact into their AI systems as well as the idea of digital sobriety. The European Union’s AI ACT aims to include environmental obligations as major principles that different AI systems in Europe must comply with.