AI Conundrum Workshop
Follow-up

- Resources -

You participated in a GenAI Conundrum workshop with us,

where we exchanged a wealth of information!

This page aims to provide you with further inputs to support your sustainable AI journey.

Feel free to revisit it, as we will update it regularly.

We also welcome any feedback or suggestions you might have!

Definitions, Core AI Concepts & Overall Impacts

The International standard ISO/IEC 22989:2022 provides formal definitions and terminology for artificial intelligence systems, including scope, outputs, and human-defined objectives.
The latest study by the Green IT Association from 2025 analyses the environmental footprint of digital technologies, including emissions across production, use, and end-of-life.
The paper "The Climate & Sustainability Implications of Generative AI (2024)" published at MIT explores rebound effects, resource consumption, and systemic sustainability risks linked to generative AI.

Environmental Impacts of AI

The International Energy Agency (IEA) provides an a uthoritative global assessment of current and projected energy demand from AI and data centers, including efficiency scenarios.
The academic paper “Making AI Less Thirsty” analyses the hidden water footprint of AI models and proposes mitigation strategies.
Research by Cornell shows the projected water and energy consumption of AI-driven data center expansion in the US, which you can read about in the article "Environmental Impact of AI Data Center Boom".
Learn more about how cloud infrastructure contributes to water stress and material corrosion in this article by the American Geographical Society.
The research by Alex de Vries-Gao looks on water and energy trade-offs in large-scale AI infrastructure.
The Environmental and Energy Study Institute (EESI) provides a policy-oriented overview of water consumption linked to data centers.
This article by the Lincoln Institute of Land Policy investigates the land and water impacts from data center development.
Mistral's life-cycle analysis of AI systems covers hardware manufacturing, training, inference, energy use, and water consumption.

Best Practices from the GenAI Conundrum Workshop

PROTECT SENSITIVE DATA Each type of data requires a level of protection proportional to its sensitivity. In particular, it is crucial to avoid any possibility of personal identification, whether through explicit or inferred information (such as an email address, a job description, or a company name). Some GenAI tools allow users to disable history or data reuse through their settings, but this relies on trust in the provider. A safer approach is to use only data that is already publicly accessible or anonymized. Ideally, a local tool should be preferred to minimize risks.

RESPECT INTELLECTUAL PROPERTY It is recommended to avoid, to the best of one’s knowledge, using GenAI tools that rely on content which does not respect intellectual property rights. In addition, it is important to check the terms of use of a GenAI tool in order to determine its intellectual property rights over the created content. In some cases, settings can be adjusted by the user, but most often the free use of a tool implies waiving any rights to the generated content, whereas a paid subscription may provide certain additional guarantees regarding intellectual property.

VERIFY INFORMATION It is essential to identify the “hallucinations” and biases that a generative AI tool may produce. The final validation of any text, image, video, or other work or decision generated or assisted by a generative AI tool is the responsibility of the user. For information retrieval, it is preferable to use a tool capable of citing sources. When this is not possible, careful manual fact-checking remains indispensable. This step is crucial to preserve both one’s personal credibility and that of one’s institution.

BE TRANSPARENT As common digital tools increasingly integrate AI, it is difficult to ensure full transparency about its use. In the case of GenAI, although situations may vary, principles of integrity should remain a guiding compass in practice, taking into account the risks of plagiarism, errors, or fraud. In general, any AI-generated content intended for the public—for example, an image or a press release—should be clearly identified as such. For routine administrative or professional tasks, declaring overall use to one’s hierarchy is generally sufficient.

CHOOSE SOBRIETY Although GenAI offers immense potential for innovation and progress, it comes with a considerable ecological impact. Microsoft, in a sustainability report, admitted to a 30% increase in its CO₂ emissions in 2023, jeopardizing its carbon neutrality goals for 2030. The use of GenAI should therefore be thoughtful and limited to cases where it is truly relevant and provides significant added value.

PRESERVE YOUR SKILLS Every professional activity relies on a set of skills that form the core of the practitioner’s expertise. These skills vary across fields. Maintaining and developing them—rather than systematically delegating their use to GenAI—not only preserves the ability to intervene when automated systems reach their limits but also more broadly helps prevent skill erosion and dependence

PROMOTE COUNTERBALANCES The control of the value chain by a few dominant players makes it essential to support open and inclusive alternatives. At an individual level, people can choose open-source solutions, remain vigilant about terms of use, and voice their expectations regarding transparency and respect for digital rights. Collectively, involvement in citizen or civic-tech initiatives helps strengthen democratic participation and support public regulation. Such mobilizations foster a more diverse and resilient digital ecosystem, capable of reducing dependence on major platforms.

EXERCISE CRITICAL THINKING GenAI is versatile. The breadth of its applications often masks the fact that it is merely a statistical tool. Its rapid data processing and pattern reproduction do not provide any real understanding of context or intuition. The added value of GenAI is therefore not uniform. For genuine performance gains, it is essential to evaluate in which areas GenAI surpasses—or, on the contrary, falls short of—human capabilities. This critical assessment must be adapted to the individuals and professions concerned.

TAKE RESPONSIBILITY The performance, autonomy, and diversity of tasks handled by GenAI can overshadow the fact that it remains nothing more than a digital tool. The moral and professional responsibility of users is inescapable. Under no circumstances can GenAI replace individual responsibility or be held accountable for the consequences of decisions or actions taken. In short, one must always exercise critical and ethical judgment when using these tools.

Further inspiration and information to act (will be updated very soon!)

ArtificialAnalysis.ai is an independent benchmarking platform comparing AI models, performance, and
A tool for comparing different GenAI solutions based on their answer.
An energy benchmark of different AI models.
This consulting report by KPMG shows how organizations move from aspirational Green IT goals to operational AI sustainability practices.
You can learn more about the use of AI tools in sustainability reporting in this report by Reuters .

AI Conundrum Workshop Follow-up