Sustainable AI implementation with Ethical oversight

1. Understanding sustainable AI

Sustainable AI refers to “the use of artificial intelligence systems that operate in ways contingent with sustainable business practices.” True sustainability extends beyond environmental considerations to include social justice and economic viability, with experts noting that “making AI sustainable is crucial to ensuring its accessibility in all countries and reducing the digital divide.”

1.1 Three fundamental dimensions

Environmental Sustainability: Focuses on reducing AI’s ecological footprint throughout its lifecycle, including energy consumption during training and inference, carbon emissions, and resource usage. Training a single large AI model can emit more than 626,000 pounds of carbon dioxide equivalent to nearly five times the lifetime emissions of an average car.

Social Sustainability: Ensures AI systems promote fairness, inclusivity, and human well-being by addressing algorithmic bias, protecting privacy, and ensuring equitable access to AI benefits. Research shows AI’s impacts are unevenly distributed, with the IMF warning that “AI will affect almost 40% of all jobs around the world and deepen inequality.”

Economic Sustainability: Examines how AI affects markets, labour, and economic structures. Studies indicate that “automation-driven AI strongly boosts output but intensifies wealth inequality,” requiring deliberate policy interventions to ensure benefits are broadly shared.

2. Key principles for sustainable AI

Several organisations have developed comprehensive frameworks for sustainable AI implementation. Access Partnership has developed 12 key principles for sustainable AI implementation:

2.1 Core principles:

1. Responsibility for all: Sustainability responsibility extends across the entire ecosystem, from the physical foundation to end users.

2. Human-centred and Purpose-driven AI: AI should be deployed thoughtfully, “ensuring that its use cases justify the environmental cost.”

3. Efficiency by design: “Sustainability should be embedded in AI systems from the design phase rather than being an afterthought.”

4. Data efficiency & Minimisation: The AI ecosystem must “emphasise data efficiency by reducing reliance on massive datasets.”

2.2 Additional principles

Additional principles include prioritising renewable energy, adopting a lifecycle approach, ensuring accountability and transparency, implementing sustainable governance, utilising edge computing when possible, building for resilience, and fostering collaboration.

3. Environmental sustainability solutions

Addressing AI’s environmental footprint requires multifaceted approaches:

3.1 Technical solutions

1. Adapt technology architecture: Chipmakers are “developing ways to cut energy use from the outset,” with innovations achieving “in some cases a 96% improvement.”

2. Optimise training geography: Computationally intensive tasks can be strategically relocated to “regions with abundant, low-cost, low-carbon electricity.”

3. Select appropriately-sized models: Many specialised AI applications “consume far less electricity than generative AI systems.”

3.2 Infrastructure solutions

Power Infrastructure and Operations: Data centres should prioritise renewable energy sources and implement advanced cooling technologies, as cooling systems consume “38% to 40% power” in data centres.

4. Social and Economic considerations

Without intervention, AI risks widening existing divides between:
– Those who can and cannot effectively use AI
– Advanced and emerging economies
– Workers with different skill levels
– Capital owners and labour

The IMF emphasises the importance of establishing comprehensive social safety nets and offering retraining programs for vulnerable workers to make the AI transition more inclusive.

5. Ethical oversight and Governance

5.1 Established ethical guidelines

The World Economic Forum identifies nine core ethical AI principles derived “from globally recognised fundamental human rights, international declarations and conventions or treaties.” These include principles like fairness, safety, lawfulness, and compliance.

The principles can be categorised as:
– Epistemic principles: Prerequisites for investigating AI ethicality, including interpretability and reliability
– General ethical principles: Behavioural principles valid across cultures, addressing accountability, privacy, human agency, and more.

5.2 Regulatory approaches

Different regions have developed distinct approaches to AI governance:

– UK: Recently announced “plans to introduce legislation in 2025 to address AI risks, making voluntary agreements with AI developers legally binding.”
– EU: The pioneering AI Act represents “a step towards a comprehensive, risk-based regulatory regime.”
– US: A “decentralised, sector-specific approach, which relies on a mosaic of federal and state regulations, agency guidelines, and industry standards.”
– China: “Integrates AI regulation within its broader data security framework.”
– Singapore: Has “crafted a non-mandatory, guidelines-based framework that prioritises industry growth and agility.”

6. Stakeholder involvement in oversight

Effective AI governance requires diverse stakeholder participation throughout the AI lifecycle:

6.1 Models of collaboration

– Consultative model: Seeking input without granting decision-making power
– Participatory model: Involving stakeholders directly in decision processes
– Co-creation model: Collaboratively developing AI policies

6.2 Governance structures

– AI ethics committees: Multidisciplinary teams addressing ethical considerations
– AI governance boards: Providing strategic oversight and policy development
– Cross-functional teams: Integrating diverse expertise for comprehensive solutions

7. Measuring AI sustainability

Approaches generally assess three key dimensions:

7.1 Sustainability impact categories

– Social impacts: Effects on employment, privacy, and equity
– Ecological impacts: Environmental footprint, including energy consumption and resource usage
– Economic impacts: Market concentration, economic inequality, and overall economic benefits

7.2 Specific measurement metrics

– Energy efficiency metrics: Assessing energy consumption during training and deployment
– Social equity indicators: Measuring impact on different demographic groups
– Economic performance metrics: Evaluating economic benefits like job creation and market growth

8. Real-world implementation examples

8.1 Corporate leadership

– IBM: Developed the AI Fairness 360 Toolkit, “an open-source library with over 70 fairness metrics and algorithms.”
– Google: Refined its diabetic retinopathy detection algorithm after early deployment “revealed issues with accessibility and real-world usability.”
– Microsoft: Addressed bias in speech recognition technology after discovering “higher error rates for African American users” by “expanding data collection and involving sociolinguists.”
– Meta: Uses AI to “detect and reduce the spread of misinformation on its platforms.”

8.2 Sector-specific applications

– Healthcare: AI systems aid in disease diagnosis with potential to “save the U.S. healthcare system $150 billion annually by 2026.”
– Finance: AI enhances fraud detection by “analysing transaction data to identify unusual patterns.”
– Agriculture: AI-driven precision farming helps “optimise crop yields by analysing soil conditions, weather patterns, and crop health data.”
– Energy Management: AI optimises smart grids by “predicting energy demand and managing distribution efficiently.”
– Public Services: AI systems assist in disaster response by “analysing data to predict and manage natural disasters.”

9. Conclusion: Recommendations for implementation

9.1 Technical recommendations

Organisations should adopt frameworks while implementing technical solutions such as efficient architectures, optimised training locations, and appropriately-sized models. Renewable energy adoption and advanced cooling technologies can significantly reduce environmental impacts.

9.2 Policy recommendations

To address social and economic dimensions, policy interventions like safety nets and retraining programs are essential to ensure AI benefits are broadly shared. Diverse stakeholder involvement strengthens governance and builds public trust.

9.3 Future outlook

By balancing innovation with responsibility, we can harness AI’s transformative potential while mitigating negative impacts and ensuring equitable benefit distribution. The implementation of sustainable AI is not merely a technical challenge but a societal imperative requiring collaboration across disciplines, sectors, and borders.

10. Atlas Forge: Your partner in ethical AI implementation

Atlas Forge, as a specialised AI consultancy, offers comprehensive support for organisations navigating the complex landscape of ethical AI oversight and governance. Our multidisciplinary team combines technical expertise in AI systems with deep knowledge of regulatory frameworks across global jurisdictions, helping clients develop customised governance structures that align with both regional requirements and organisational values.

Atlas Forge provides end-to-end services including ethical risk assessment, stakeholder engagement facilitation, development of transparent AI documentation processes, and implementation of continuous monitoring frameworks that evolve with changing regulations. Our collaborative methodology ensures that ethical considerations are embedded throughout the AI lifecycle rather than treated as compliance checkboxes, enabling organisations to build AI systems that are not only powerful and efficient but also trustworthy, fair, and aligned with broader sustainability goals. Through strategic partnerships with Atlas Forge, organisations can transform ethical governance from a potential constraint into a competitive advantage that builds stakeholder trust and ensures long-term AI sustainability.

Sources

[1] 12 Key Principles for Sustainable AI – Access Partnership https://accesspartnership.com/12-key-principles-for-sustainable-ai/
[2] What is sustainable AI? – TechTarget https://www.techtarget.com/searchenterpriseai/definition/sustainable-AI
[3] The Carbon Footprint of AI: How AI Impacts Climate Change https://www.carma.earth/blog-posts/the-carbon-footprint-of-ai
[4] As generative AI asks for more power, data centers seek … – Deloitte https://www2.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2025/genai-power-consumption-creates-need-for-more-sustainable-data-centers.html
[5] How will artificial intelligence affect wealth equality? – Phys.org https://phys.org/news/2025-03-artificial-intelligence-affect-wealth-equality.html
[6] Artificial Intelligence will affect jobs and worsen inequality, says IMF https://www.thomasandyoung.co.uk/news/latest-news-for-business/archive/news-article/2024/February/artificial-intelligence-will-affect-jobs-and-worsen-inequality-says-imf
[7] Understanding the carbon footprint of AI and how to reduce it https://www.carbon-direct.com/insights/understanding-the-carbon-footprint-of-ai-and-how-to-reduce-it
[8] 9 ethical AI principles for organizations to follow https://www.weforum.org/stories/2021/06/ethical-principles-for-ai/
[9] [PDF] Towards AI ethics-led sustainability frameworks and toolkits – Pure https://pure-oai.bham.ac.uk/ws/portalfiles/portal/227076712/1-s2.0-S2950370124000038-main.pdf
[10] Ethical AI: what does it mean for environmental tech? – Nymark https://www.nymark.agency/journal/ethical-ai-what-does-it-mean-for-sustainability
[11] A Comparative Analysis of AI Governance Frameworks https://wjlta.com/2024/07/09/a-comparative-analysis-of-ai-governance-frameworks/
[12] AI Watch: Global regulatory tracker – United Kingdom https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker-united-kingdom
[13] Real-World Case Studies of Ethical AI in Action – LinkedIn https://www.linkedin.com/pulse/real-world-case-studies-ethical-ai-action-bright-next-academy-xvyff
[14] 30 Inspiring Examples of Responsible AI in Action – Convin https://convin.ai/blog/responsible-ai
[15] Stakeholder Engagement and Governance Structures in Artificial … https://aign.global/ai-governance/stakeholder-engagement-and-governance-structures-in-artificial-intelligence-ai/
[16] Civil society perspectives on AI in the EU – WISERD https://wiserd.ac.uk/blog/civil-society-perspectives-on-ai-in-the-eu/
[17] Sustainable AI Metrics for Evaluation | Restackio https://www.restack.io/p/sustainable-ai-answer-evaluating-ai-metrics-cat-ai
[18] The Sustainability Index for Artificial Intelligence – AlgorithmWatch https://algorithmwatch.org/en/sustain/
[19] Comprehensive sustainability criteria and indicators for AI systems https://arxiv.org/abs/2306.13686
[20] These 7 principles ensure AI remains human-centric https://www.weforum.org/stories/2024/01/7-principles-integrate-artificial-intelligence-impact/
[21] The role of artificial intelligence in achieving the Sustainable … https://www.nature.com/articles/s41467-019-14108-y
[22] [PDF] Foundations for Environmentally Sustainable AI https://nepc.raeng.org.uk/media/2aggau2j/foundations-for-sustainable-ai-nepc-report.pdf
[23] AI Principles – Google AI https://ai.google/responsibility/principles/