Since the arrival of ChatGPT in late 2022, Artificial Intelligence has embedded itself into our personal and professional lives at an unprecedented pace. No technology had ever achieved such rapid adoption on a global scale. Within just a few months, what still belonged to the realm of specialized research became a tool accessible to the many — and a major strategic priority for organizations.
Well beyond the generative applications popularized in recent years, artificial intelligence technologies are today transforming a wide range of sectors and functions: healthcare, industry, cybersecurity, customer relations, decision support, and operational optimization. Across companies and public institutions, AI projects are expanding rapidly, driven by significant financial investment and ambitious policy commitments. Across companies and public institutions, AI projects are expanding rapidly, driven by significant financial investment and ambitious policy commitments.
This acceleration is opening new horizons for organizations: productivity gains, automation of certain tasks, improved employee experience, operational optimization, faster innovation, and enhanced decision support.
At the same time, this explosion in usage is not without consequences. While AI is profoundly transforming working methods and organizations, it also raises major environmental and social issues: increasing energy consumption, pressure on resources, algorithmic biases, digital sovereignty, and the evolution of professions.
In this context, the challenge is no longer simply to adopt AI, but to develop uses that are at once relevant, responsible, and sustainable.
A technological revolution with tangible environmental impacts
AI may seem intangible. Yet it relies on physical infrastructure and equipment that is particularly energy-intensive: data centers, networks, servers, and electronic components that provide the computing power needed to train and run models.
Accurately assessing AI’s environmental footprint remains a complex undertaking. The lack of transparency from some industry players, the absence of a shared reference methodology, and the sheer number of variables involved (AI models, energy mix, infrastructure location, nature of queries, hardware used…) all make comparisons difficult.
One thing is certain, however: energy consumption linked to AI is surging. As usage becomes widespread, it is no longer only the model training phase that carries a heavy footprint, it is now their everyday use, known as the “inference” phase.
In Ireland, for instance, data centers now account for nearly 22% of national electricity consumption, up from 5% a decade ago. A growth trajectory that is straining energy grids and may lead to increased reliance on fossil fuels.
Beyond energy consumption, AI-related infrastructure also requires substantial volumes of water for server cooling purposes. Several US states already affected by water stress, including Arizona and Texas, are witnessing the emergence of local tensions surrounding the establishment of new data center facilities.
Finally, the production of digital equipment mobilizes minerals and rare earth elements deemed critical, giving rise to significant environmental, geopolitical, and strategic dependency concerns.
Societal impacts to anticipate
AI is bringing about profound changes in organizations and professional roles. While it offers the prospect of freeing up time through the automation of certain repetitive tasks, it may also generate cognitive overload, a perceived loss of control, or a reassessment of specific areas of expertise.
Contrary to popular belief, creative, analytical, and intellectual professions are not spared. Generative AI is already disrupting content production, advisory activities, customer relations, and a number of support functions.
Furthermore, AI depends on a form of human labor that largely goes unacknowledged. L’entraînement et la modération des modèles mobilisent des milliers de travailleurs à travers le monde, parfois dans des conditions sociales difficiles et peu encadrées.
Another major issue is algorithmic bias. Far from being neutral, AI systems reproduce the biases present in the data on which they are trained. Le cas de l’algorithme de recrutement d’Amazon, qui défavorisait les candidatures féminines, illustre concrètement ces dérives possibles. Le cas de l’algorithme de recrutement d’Amazon, qui défavorisait les candidatures féminines, illustre concrètement ces dérives possibles.
Finally, the question of digital sovereignty is becoming central. The majority of AI solutions in use today come from American or Chinese players, placing organizations in the face of technological dependency, data security, and control over critical infrastructure.
On top of this come new cyber risks: sensitive data leakage, model manipulation, prompt injection, and the automation of cyberattacks.
Toward responsible AI: meeting the challenges ahead
Given the environmental and social challenges associated with AI, the question of responsible use is emerging as a central concern.
- Starting form a concrete need
Launching an AI project should not be driven by trend, but grounded in a clearly identified need. First and foremost, it is worth asking whether AI is the right tool for the job, and what level of technology is genuinely required.
Depending on the use case, a simpler solution can sometimes deliver relevant results with lower complexity and a more limited environmental footprint.
AI should not become an end in itself, but remain a lever for transformation in the service of business functions and organizations.
- Scaling AI to its intended use
A responsibly designed AI is above all an AI adapted to the uses it is meant to serve. This involves choosing the right type of AI (predictive, generative), the right models, the appropriate level of processing power, and suitable hosting solutions.
A specialized AI system designed to perform a defined task will generally prove more resource-efficient than a general-purpose generative AI capable of addressing a broad range of applications.
- Placing people at the center, from conception to use
To design a solution that is useful, adopted, and value-creating, people must be placed at the heart of the co-design process. AI should be conceived as an assistive tool in the service of business functions, not as an autonomous decision-maker.
Beyond the design phase, openness to dialogue, consideration of user feedback, and change management support are essential to foster adoption and bring stakeholders on board.
- Designing AI without bias or exclusion
By examining the discrimination mechanisms that may exist and incorporating diverse, representative training data, it becomes possible to design AI systems that are more inclusive and less biased. In healthcare, for instance, a melanoma detection AI will perform better if it has been trained across a wide range of skin tones. Inclusivity also means taking accessibility into account: ensuring that services developed are accessible to people with disabilities is essential.
- Opening up the black box
Rather than functioning as an opaque system producing decisions that are difficult to interpret, AI must be transparent in order to inspire trust. Transparency regarding the algorithms deployed, operating rules, source traceability, and the explainability of outputs contributes to the development of a well-governed AI — reducing exposure to legal, financial, and reputational risks.
In the face of these challenges, the question is not about slowing down innovation, but about building an AI that is truly responsible, value-creating, and aligned with the needs of organizations.
At Willing, we support organizations in developing more responsible AI applications, striking the right balance between performance, environmental impact, and human considerations. An approach that enables the design of solutions that are more relevant, more resource-efficient, and better suited to real business needs.