At aihorizon R&D, we believe AI should drive innovation without compromising ethical and sustainability principles. However, greater efficiency can paradoxically increase energy consumption—an effect known as Jevons’ Paradox. While improving model efficiency reduces per-task energy use, the widespread adoption of AI ultimately raises total carbon emissions.

To address this challenge, we are launching the CO₂ Tracker—a transparency-driven tool designed to measure, monitor, and mitigate AI-related carbon emissions. Designed for transparency and accountability, it directly addresses Jevons’ Paradox, giving users real-time insights to make informed decisions about their environmental impact.

To maximize its impact and ensure accuracy, we are collaborating with LMU Munich for empirical validation. As part of this effort, 200 students from the Educational Foundation Freiburg will participate in a structured test phase, generating empirical data that will inform both the CO₂ Tracker’s development and a forthcoming research paper on AI sustainability tracking.

 

The CO₂ Tracker: Bridging AI and Sustainability

To fully integrate sustainability into AI, we focus on three foundational pillars—legitimacy, legality, and domain-specific insights—ensuring that AI sustainability efforts are ethically justified, legally grounded, and contextually relevant.

 

Legitimacy: Ethical & Normative Justification

Rooted in our Responsible AI Framework, the CO₂ Tracker upholds ethical principles that extend sustainability efforts beyond legal compliance. This ensures AI development remains within ethical boundaries even before legal structures evolve.

​

The following examples are taken from the RAI framework:

• Intergenerational Justice → AI should safeguard future generations, making sustainability a core ethical responsibility.

• Ecological Integrity → AI must function within environmental limits, ensuring technological progress does not compromise ecological balance.

• Transparency & Accountability → Sustainability tracking must be clear and accessible, allowing users to clearly understand the carbon dioxide impact of AI.

 

By embedding these principles into AI sustainability monitoring, the CO₂ Tracker establishes environmental accountability as a fundamental pillar of Responsible AI.

Beyond ethical justification, regulatory alignment is crucial for reinforcing AI sustainability as both principled and enforceable.

 

Legality: Compliance & Anticipating Future AI Regulations

The CO₂ Tracker aligns with international sustainability regulations and human rights frameworks, reinforcing AI’s role in advancing environmental accountability.

 

The following examples are taken from the RAI framework:

• Universal Declaration of Human Rights (UDHR) – Recognizes a healthy environment as fundamental to human well-being.

• United Nations Framework Convention on Climate Change (UNFCCC) – Seeks to stabilize greenhouse gas concentrations at levels that prevent harmful human-induced impacts on the climate.

• Sustainable Development Goal 13 (SDG 13) – Calls for immediate action to combat climate change by regulating emissions and encouraging renewable energy solutions.

• Rio Declaration on Environment and Development – Highlights the importance of public participation and ensuring citizens have access to environmental information.

 

While the EU AI Act hasn’t mandated AI sustainability reporting yet, the CO₂ Tracker stays ahead of future regulations. It embeds sustainability tracking as a standard before compliance becomes mandatory.

 

Domain-Specific Insights: Adapting Sustainability for AI Sectors

The CO₂ Tracker is designed to address sector-specific sustainability challenges, making AI’s environmental impact actionable for diverse stakeholders.

• Industry & Enterprise AI → Supports corporate sustainability goals by enabling organizations to track and reduce AI emissions.

• Nonprofits & NGOs → Helps mission-driven organizations integrate AI responsibly without undermining sustainability commitments.

• Education & Research → Empowers students and researchers to develop AI with sustainability in mind, fostering early awareness and accountability.

 

By providing tailored sustainability metrics, the CO₂ Tracker makes responsible environmental stewardship measurable and achievable.

However, ensuring true accountability requires more than just sector-specific applications—it necessitates a deeper alignment between ethical principles and technical execution.

​

Bridging Normative and Technological Dimensions through Epistemic Values

AI Horizon R&D’s Responsible AI Framework employs epistemic values as a methodological bridge, demonstrating that normative and technological spheres are deeply interwoven rather than separate. Values such as transparency, coherence, and accuracy illustrate how ethical principles and technological precision reinforce one another, ensuring that AI is both ethically grounded and functionally robust.

The CO₂ Tracker exemplifies this integration by embedding epistemic values into its design. It enables rigorous CO₂ measurement, incorporates user-centric transparency features, and prioritizes accountable design choices. By embedding epistemic values into its very structure, the tracker transforms abstract ethical imperatives into a tangible, actionable AI tool. This reinforces that Responsible AI must not only assess impact but also uphold standards of trustworthiness, accountability, and real-world usability.

 

Comprehensive CO₂ Modeling and Visualizations

The CO₂ Tracker calculates AI-related CO₂ emissions by using the fundamental concept of tokens—the smallest meaningful units of text processed by language models (LLMs)—and by estimating emissions related to AI-generated visuals. Tokens represent words, word fragments, or punctuation marks, and their exact breakdown varies by AI model. For example, the sentence "I am going for a walk today." may be tokenized into "I", "am", "go", "ing", "for", "a", "walk", "today", and ".".

By associating each token and generated image with model-specific approximations of CO₂ emissions, the tracker precisely estimates the carbon footprint of individual prompts. These per-prompt emission values are recorded in a user-specific database, enabling transparent and cumulative tracking of each user's environmental impact.

Our methodology is based on a European energy mix at its median value, using AI models deployed on Azure cloud resources for consistency.

Additionally, we have modeled 25 different CO₂ equivalent values, from small-scale activities like using an electric toothbrush to large-scale emissions from driving a car or launching a rocket. This helps users contextualize AI emissions with real-world activities.

 

To make CO₂ data more accessible, we offer varied visualizations, including:

• Linear and non-linear consumption curves → Showing how emissions scale with AI usage.

• Consumption over time → Highlighting cumulative emissions from extended AI interactions.

• Increase and decrease comparisons → Demonstrating the impact of AI efficiency optimizations.

• Equivalent visualizations → Translating AI carbon footprints into relatable real-world activities.

These visualizations simplify emissions data, making sustainable AI decisions more intuitive.

To ensure that these insights are not just theoretical, we have integrated them directly into our Horizon v2 platform.

 

Seamless Integration into Horizon v2

Embedded within our enhanced Horizon v2 platform, the CO₂ Tracker provides precise CO₂ measurement metrics while fostering discussions on sustainability. The Responsible AI Coach further enhances this experience, offering insights into AI’s environmental impact.

 

User Behavior Impact and Cultivating Culture (Experimental Study at LMU Munich)

As part of our commitment to empirical validation, our study with 200 students from the Educational Foundation Freiburg will assess how CO₂ tracking influences AI behavior. By providing visibility into emissions, we aim to encourage users to reduce redundant queries, optimize AI usage, and adopt more sustainable practices. Insights from our study will contribute to a larger goal: fostering a culture of sustainability in AI development. As global concerns about climate change intensify, responsible AI development is no longer optional—it is essential.

 

Future Vision: Beyond the CO₂ Tracker

The CO₂ Tracker is just the beginning. Our vision is to align our innovations with a future Sustainability Index for AI, where models are rated by their carbon footprint—similar to energy efficiency labels on appliances.

More than a tool, the CO₂ Tracker represents a commitment to AI’s sustainable future. At aihorizon R&D, we’re not just measuring emissions—we’re helping set new standards to ensure innovation and responsibility go hand in hand.

​

Contributors: Dr. Michael Jülich (Conceptual Lead), Moritz Goeke (Technical Lead), Leon Kollofrath (Design & Engineering), Silas Stulz (Engineering), Dr. Hans Lückhoff (Ethics & Law) and Kirstin Stiller (Ethics & Law)