AI in the service of sustainable work

Artificial Intelligence (AI) has become the essential keyword for digital transformations. Yet, beneath the promise of a revolution that will boost productivity lies a much more nuanced reality: undeniable opportunities, but also technical, organizational, and social blind spots that, if ignored, could undermine the sustainability of work and employment.

The first necessity is to clarify what we mean by "AI." It is not a monolithic entity endowed with reason, but a vast array of algorithms and technical systems whose scope ranges from simple information search engines to agents capable of autonomously executing actions determined by probabilistic calculations. Within this landscape, we can distinguish five main categories: recognition and analysis, prediction and diagnosis, recommendation, content generation, and autonomous execution. Each relies on massive datasets, statistical learning, and a more or less opaque software architecture.

This lack of transparency is the first point of concern. When even the designers struggle to trace the internal logic of a model, how can the reliability of a medical recommendation or an HR decision be guaranteed? The absence of transparency complicates the detection of gender, racial, or social biases contained in training data, thus exposing organizations to costly or discriminatory errors. Hence the emergence of "explainable" AI: the issue is not only technical, but also touches on labor law, responsibility, and the legitimacy of decisions.

The second blind spot concerns how we measure the economic impact of these tools. Traditional approaches assume that a technology spreads like an exogenous shock, automating "routine tasks" and then mechanically redistributing the remaining work. This techno-deterministic view underestimates the role of organizational choices and overlooks a critical factor: the company's learning capacity, that is, its ability to transform field experience into actionable knowledge. European data shows that investing in this capacity protects jobs better than simply pursuing automation.

Based on these observations, four levers appear essential to put AI on a sustainable trajectory.

First, a renewed social dialogue: informing, debating, and negotiating the use of algorithms from their inception. Without this collective counter-expertise, there is a significant risk that AI will reinforce opaque decision-making chains or erode professional autonomy.

Next, the systematic development of organizations' learning capacity. This involves preserving the cognitive dimension of work, increasing opportunities for discussion, supporting supervised experimentation, and facilitating peer-to-peer knowledge sharing. These practices—eight key dimensions have been identified, ranging from autonomy in tasks to cooperative management—constitute a social buffer against the adverse effects of automation.

Third pillar: participatory project management focused on real-world activity. Too many deployments are still done "off the shelf," applying technical standards to work realities they don't understand. Involving professionals in defining needs, testing in situ, and continuously adjusting reduces hidden costs (loss of expertise, cognitive overload, value conflicts).

Fourth lever: document and share real-world experiments. Purely "laboratory" evaluations often show spectacular gains, quickly dispelled as soon as the tool encounters the complexity of customer service or the variability of a production line. Building an open and multidisciplinary database of feedback becomes vital to regulating the pace of innovation, avoiding perpetual hype, and anchoring AI in a logic of continuous improvement.

These four pillars outline a roadmap: moving from AI "to replace" to AI "to learn." The core issue is not the volume of code but the value created: work that is biocompatible, ergonomically compatible, and socially compatible, capable of preserving health, engagement, and the environment. It is under these conditions that the virtuous cycle can begin: an organization that learns better designs more relevant uses, which in turn foster new skills and open up avenues for inclusive innovation.

In practice, this means reconsidering success indicators. Automation rates or deployment speed are not enough; they must be complemented by metrics for skills development, quality of cooperation, reduction of psychosocial risks, and environmental impact. In many scenarios, AI is not a magic bullet: a technological transformation remains incomplete if it is not used to generate new knowledge.

The question of external regulation remains. A graduated approach, based on the level of risk and combining hard law, technical standards, and collective agreements, is developing. But it will only bear fruit if those on the ground actively engage with it: there can be no compliance without a shared understanding of the issues and without the power to act on the systems.

Ultimately, choosing AI means choosing the future of work. The debate cannot be reduced to a binary choice between "jobs destroyed or preserved." The real question is: what forms of work do we want tomorrow? The answers will be developed in co-design workshops, discussion forums, and social dialogue bodies, where technical expertise meets practical experience. It is in this crucible, and not solely through the power of algorithms, that AI will be forged that truly serves people and organizations, guaranteeing the creation of sustainable and shared value.

To learn more: discover our thoughts on AI and IH in harmony and our article Reconciling AI and Human Intelligence.