Open-source artificial intelligence can accelerate sustainable development goals
Open-source artificial intelligence can accelerate sustainable development goals. Open-source artificial intelligence represents a major opportunity to advance global sustainable development goals. Between 2018 and 2024, its use to support these goals increased by 300%, with notable contributions in areas such as agriculture, ecosystem management, and the fight against climate change. This open model, which allows everyone to study, modify, and share tools, was designed to accelerate progress in an effective and inclusive manner.
However, its expansion has revealed complex challenges. The lack of a structured governance framework can lead to excessive resource consumption and ethical vulnerabilities. For example, the decentralized development of redundant models on inefficient local infrastructures worsens the industry’s carbon footprint. Additionally, the ease of access to open-source tools has reduced the production costs of deepfakes and targeted disinformation, posing risks to security and trust in information.
Another issue is digital colonialism, where powerful countries and institutions in the Global North control data and infrastructures, limiting the autonomy of countries in the Global South. Datasets designed for local contexts may not apply well elsewhere, leading to errors and reduced relevance for governance solutions.
To address these challenges, four key actions are proposed. The first involves integrating sustainability throughout the lifecycle of models by optimizing hardware infrastructures and avoiding waste from redundant developments. Techniques such as model distillation can reduce their size by up to 70%, thereby limiting their environmental impact. Another approach is the adoption of liquid-cooled data centers powered by renewable energy, reducing energy demand by 20% and water consumption by 52%.
The second action aims to establish a quantitative evaluation framework to measure the impact of artificial intelligence on sustainable development goals. This involves creating composite indicators that integrate local and global data, enabling a precise assessment of positive and negative effects. For example, sustainability criteria could cover more than 40 social, environmental, and economic indicators, providing a clear view of the real impact.
The third action relies on regulation and transparency. To prevent abuse, it is essential to implement mandatory audits, adversarial testing, and clear licenses defining permitted uses. Developers must use audited datasets to limit biases, while governments can impose strict policies to prevent the spread of malicious content. Users, for their part, have the responsibility to apply these tools ethically and report misuse.
Finally, the fourth action encourages global cooperation and knowledge sharing. Despite the democratization of access, inequalities persist in terms of infrastructure, talent, and data sovereignty. The creation of open-access platforms, adhering to the principles of findability, accessibility, and reusability, is crucial to ensuring equitable access to resources. Institutions such as a Global Dialogue on AI Governance could strengthen cooperation by publishing annual reports and organizing intergovernmental negotiations.
At the regional level, collaboration between global platforms and local research centers allows combining shared foundational models with context-specific data. This facilitates the deployment of tailored solutions while respecting data protection regulations. Institutions such as the Australian Institute for Artificial Intelligence or France’s National Institute for AI Evaluation and Security play a key role in this localization.
Open-source artificial intelligence, combined with appropriate governance, could thus transform decision-making, making it more inclusive and evidence-based. It would enable faster identification of priorities, obstacles, and pathways for sustainable transformation, even in the face of complex challenges. By integrating lessons learned from citizens worldwide, it could also make the post-2030 agenda more representative and effective.
Open-source models also offer concrete advantages. Their cost is 5 to 29 times lower than that of proprietary models for comparable performance, which broadens their accessibility. Tools such as natural language processing models specialized in climate help combat disinformation, while others, like those used in healthcare, could mitigate therapist shortages in certain regions. In the economic sector, small and medium-sized enterprises can adopt and customize these technologies to optimize their production.
However, without appropriate governance, risks persist. The production of electronic materials and the infrastructure required for artificial intelligence contribute to global toxic waste. Additionally, nearly 60% of models labeled as open-source have no license, complicating their secure use. The proliferation of unregulated models also makes it easier for malicious actors to identify vulnerabilities, increasing the risks of phishing or the spread of hate speech.
To maximize the benefits of open-source artificial intelligence, a coordinated approach is essential. It must combine sustainability, rigorous evaluation, security, and cooperation to reduce uncertainties and ensure that this technology effectively serves global progress.
Open-source artificial intelligence can accelerate sustainable development goals. Open-source artificial intelligence represents a major opportunity to advance global sustainable development goals. Between 2018 and 2024, its use to support these goals increased by 300%, with notable contributions in areas such as agriculture, ecosystem management, and the fight against climate change. This open model, which allows everyone to study, modify, and share tools, was designed to accelerate progress in an effective and inclusive manner.
However, its expansion has revealed complex challenges. The lack of a structured governance framework can lead to excessive resource consumption and ethical vulnerabilities. For example, the decentralized development of redundant models on inefficient local infrastructures worsens the industry’s carbon footprint. Additionally, the ease of access to open-source tools has reduced the production costs of deepfakes and targeted disinformation, posing risks to security and trust in information.
Another issue is digital colonialism, where powerful countries and institutions in the Global North control data and infrastructures, limiting the autonomy of countries in the Global South. Datasets designed for local contexts may not apply well elsewhere, leading to errors and reduced relevance for governance solutions.
To address these challenges, four key actions are proposed. The first involves integrating sustainability throughout the lifecycle of models by optimizing hardware infrastructures and avoiding waste from redundant developments. Techniques such as model distillation can reduce their size by up to 70%, thereby limiting their environmental impact. Another approach is the adoption of liquid-cooled data centers powered by renewable energy, reducing energy demand by 20% and water consumption by 52%.
The second action aims to establish a quantitative evaluation framework to measure the impact of artificial intelligence on sustainable development goals. This involves creating composite indicators that integrate local and global data, enabling a precise assessment of positive and negative effects. For example, sustainability criteria could cover more than 40 social, environmental, and economic indicators, providing a clear view of the real impact.
The third action relies on regulation and transparency. To prevent abuse, it is essential to implement mandatory audits, adversarial testing, and clear licenses defining permitted uses. Developers must use audited datasets to limit biases, while governments can impose strict policies to prevent the spread of malicious content. Users, for their part, have the responsibility to apply these tools ethically and report misuse.
Finally, the fourth action encourages global cooperation and knowledge sharing. Despite the democratization of access, inequalities persist in terms of infrastructure, talent, and data sovereignty. The creation of open-access platforms, adhering to the principles of findability, accessibility, and reusability, is crucial to ensuring equitable access to resources. Institutions such as a Global Dialogue on AI Governance could strengthen cooperation by publishing annual reports and organizing intergovernmental negotiations.
At the regional level, collaboration between global platforms and local research centers allows combining shared foundational models with context-specific data. This facilitates the deployment of tailored solutions while respecting data protection regulations. Institutions such as the Australian Institute for Artificial Intelligence or France’s National Institute for AI Evaluation and Security play a key role in this localization.
Open-source artificial intelligence, combined with appropriate governance, could thus transform decision-making, making it more inclusive and evidence-based. It would enable faster identification of priorities, obstacles, and pathways for sustainable transformation, even in the face of complex challenges. By integrating lessons learned from citizens worldwide, it could also make the post-2030 agenda more representative and effective.
Open-source models also offer concrete advantages. Their cost is 5 to 29 times lower than that of proprietary models for comparable performance, which broadens their accessibility. Tools such as natural language processing models specialized in climate help combat disinformation, while others, like those used in healthcare, could mitigate therapist shortages in certain regions. In the economic sector, small and medium-sized enterprises can adopt and customize these technologies to optimize their production.
However, without appropriate governance, risks persist. The production of electronic materials and the infrastructure required for artificial intelligence contribute to global toxic waste. Additionally, nearly 60% of models labeled as open-source have no license, complicating their secure use. The proliferation of unregulated models also makes it easier for malicious actors to identify vulnerabilities, increasing the risks of phishing or the spread of hate speech.
To maximize the benefits of open-source artificial intelligence, a coordinated approach is essential. It must combine sustainability, rigorous evaluation, security, and cooperation to reduce uncertainties and ensure that this technology effectively serves global progress.
Source Credits
Primary Source
DOI: https://doi.org/10.1038/s41467-026-73866-8
Title: Steering open-source AI to accelerate the sustainable development goals
Journal: Nature Communications
Publisher: Springer Science and Business Media LLC
Authors: Min Chen; Kai Wu; Prajal Pradhan; Cameron Allen; Stefano Nativi; Klaus Hubacek; Alexey Voinov; Felix Creutzig; Tatiana Filatova; Niklas Boers; Michael Meadows; Peilong Ma; Frank Biermann; Hans Joachim Schellnhuber; John Ludden; Maria Paradiso; Michael Batty; Huadong Guo; Min Cao; Peng Hou; Guonian Lü