Back

AI can be as fascinating as it is worrying. While some see it as a powerful lever for efficiency, competitiveness and innovation, others raise concerns about its environmental, societal and ethical implications. Striking the right balance between alarmist narratives and the excessive promises of tech giants is no easy task.
Romain BERROU (who graduated the Grande École Program in 2015), a sustainability specialist and Associate at NSW Conseil, takes a pragmatic view of the debate. In his opinion, AI is neither inherently good nor bad: it is a tool whose impact depends on human choices, regulatory frameworks and the ways in which it is used. He encourages us to move beyond preconceived ideas and to seek a fair balance between performance and responsibility.

You started using AI well before it became mainstream. What motivated you?

I discovered machine learning while working on my thesis between 2016 and 2020. At the time, it was mainly used to complement advanced statistical analysis. I began exploring the growing body of scientific literature on AI, which expanded significantly around 2019. I quickly started applying these tools in practice, particularly during my first consulting assignments, while always ensuring the protection of personal and sensitive data. AI undoubtedly helped me work more efficiently and produce sharper analyses. At the same time, I was concerned about its potential social, environmental and sovereignty-related impacts. I became interested in the open-source ecosystem and, drawing on my coding background, realized that it was possible to develop solutions that combine performance with sustainability.

Do you think AI and sustainability can truly go hand in hand?

This question reminds me of one I was asked while working on my thesis about sustainable finance. Like finance, AI is a tool – and it can be used either responsibly or irresponsibly. Economies need finance to function, yet it must be regulated and framed to limit its risks.

The same applies to AI: it brings both risks and opportunities, as well as significant potential benefits for society, the environment and the economy. AI can make education more accessible and accelerate innovation. At the same time, it may exacerbate inequalities and increase pressure on energy systems, as well as the of CO₂ emissions, water and other scarce resources. At my own level, I rely on open-source models that I run locally, ensuring full control over data. On a broader scale, I believe there is a need for a shared global definition of what “sustainable AI” truly means – even if the European AI Act already represents a major step forward in that direction.

What can you tell us about NSW Conseil, the consultancy where you are an associate?

We specialize in CSR and support SMEs and private equity funds in structuring and implementing their sustainability strategies. Our services cover a wide range of areas: certification processes, ESG assessments, CSR reporting, and the definition of CSR strategies that translate into concrete action plans over the first 100 days, as well as over three- and five-year horizons — all aligned with investors’ expectations. Our approach relies on AI-driven workflows that enhance our efficiency, responsiveness and competitiveness, while maintaining high quality standards.

To what extent do you think AI can make you more efficient?

It does – provided it is used as a complement to human expertise rather than a substitute for it. By combining human intelligence with AI’s efficiency, we can deliver higher-quality services, faster and with greater responsiveness. The time saved can then be reallocated to better understanding our clients’ challenges and building more strategic support solutions.

For instance, we have developed AI tools dedicated to the EcoVadis score, which assesses a company’s CSR performance on a 100-point scale. The combination of our expertise and these tools generally results in an increase of around 20 points in our clients’ scores, while reducing invoiced time by approximately 80%. This approach enables us to produce more comprehensive CSR reports, aligned with the company’s strategy and with the expectations of investors, financiers and major clients — while keeping teams’ workloads under control.

How do you address the challenge of assessing the environmental impact of a given technology or AI tool?

We need to rely on a transparency scale. It is not possible to accurately assess the overall environmental impact of large-scale models without access to reliable and publicly available data. However, when using open-source models run locally or hosted on French or European servers, the evaluation becomes more straightforward. We can measure the computational load, estimate the energy consumption and apply the corresponding energy mix, ideally in regions with low water stress. The use of more efficient algorithms also reduces the demand on hardware and limits pressure on critical resources. When AI is developed and deployed under these conditions — with models designed with sobriety in mind — its environmental impact can remain contained.

You have developed AI solutions hosted on eco-friendly servers: what does that mean? How do open-source solutions contribute to more ethical AI?

Today, we have access to many high-performing open-source AI models. When we use an open-source model locally, the data are stored on our own machines, and it is easier to measure energy consumption. If greater computing power is required, the model can be deployed to eco-friendly and sovereign servers located in France or elsewhere in Europe. This ensures exclusive control over the data and better optimization of the environmental footprint. With this approach, we maintain full control over incoming and outgoing data flows, as well as a clear understanding of the infrastructure being used.

By contrast, large-scale public models hosted on third-party servers often involve reduced control over how and where data are processed. For us, ethical and sustainable AI must therefore be grounded in transparency, sobriety and sovereignty in technical choices.

It seems difficult to reconcile the exponential growth of AI models with the reduction of our carbon footprint. Don’t you think it is already too late?

As with climate change, I believe it is never too late. Every ton of CO₂ avoided helps limit the severity of future scenarios. The pursuit of ever more powerful models is indeed a complex issue. There is currently a debate about whether AI may be reaching a technological plateau — in other words, whether the increase in computing power still leads to proportional improvements in output quality.

This discussion goes beyond purely technological considerations; it also affects financial markets, which are questioning which AI promises will ultimately be fulfilled. In the short term, one possible approach is to strike a balance between large-scale models and more lightweight ones. The latter are often sufficient for operational needs, consume less energy and are easier to manage. At the same time, continued innovation will be essential to reduce the environmental footprint of data centers. To conclude, I do not believe it is too late. We are at a critical juncture where the choices we make will have a decisive impact.

You regularly speak in companies to raise awareness
about sustainable AI. How do professionals react to the topic? What message do you want to convey?

Reactions are mostly enthusiastic. Many professionals feel reassured to see that experts are actively addressing these challenges and that substantial work has already been done to identify the main issues and propose concrete solutions. It provides a much-needed neutral perspective, somewhere between alarmist rhetoric and excessive optimism.

My goal is to promote a balanced view. There is no inherently pessimistic vision of AI, nor an entirely idealistic one. Like any other tool, AI reflects the intentions and choices of those who design and use it. The future of AI will not depend solely on technological progress, but also on human decisions. My ambition is to play an active role in shaping uses that create sustainable value for society.

How do you foresee the evolution of artificial intelligence in the next five to ten years?

I do not believe that we will see the emergence of artificial general intelligence within the next five to ten years. Nor do I think that the practical uses of AI will prove as revolutionary as major technology companies sometimes suggest. A number of decision-makers will remain reluctant to rely on AI for highly human or strategic matters. I also anticipate that the current AI bubble may burst in the short term – possibly as early as 2026 – paving the way for a more realistic and measured approach focused on tangible risks and opportunities.

That said, the job market is likely to experience significant disruption. Certain professions may disappear, particularly in sectors centered on visual creation, such as the audiovisual and gaming industries. More broadly, much of the production of standardized content – including in consulting – could increasingly be automated. Strategic decision-making, however, will remain a distinctly human responsibility.

What would you say to encourage the general public to engage with more sustainable AI?

First, I would encourage them to engage with sustainability more broadly and to understand its key challenges. While AI is a prominent and impactful topic, it remains only one issue among many when it comes to sustainability. It should be seen as a powerful lever to address the most urgent challenges we face: reducing CO₂ emissions across energy, industry, transport, construction and agriculture, as well as limiting water and raw material consumption.

At NSW Conseil, my associates and I have set ourselves a clear objective: to harness sustainable AI in order to help companies achieve stronger environmental performance, more rapidly and under optimal conditions. Striking the right balance between maximizing AI’s potential and minimizing its risks and impacts is – and will continue to be – at the heart of our daily work.

This article was written by Luna Créations for #IÉS, the IÉSEG Network magazine.