NHNAI white paper

Humanity is confronted with major challenges including climate change, inequalities, geostrategic tensions, weakening of democratic political organizations (notably with acute questions about the equilibrium between public and private powers as well as about collective intelligence and the threat of post-truth). Moreover, these multiple challenges all occur at the same time and with very rapid and brutal dynamics. As illustrated with the well-known analogy with the pharmakon, AI has considerable potential either to support mitigating these challenges or to amplify them.

T​he current era of AI—driven by advances like deep learning and Large Language Models (LLMs)—is marked by frenetic development and intense global competition. This rush is fueled by the perception of AI as a primary economic driver, leading to a “tyranny of tardiness” where attempts at regulation are protested as economic suicide. This hectic, market-driven environment, coupled with growing power asymmetries favoring tech giants who often adhere to techno-solutionist ideologies, makes ethical AI development extremely challenging.

This document constitutes a first version elaborated by NHNAI coordination team thanks to the direct the support and inputs of NHNAI partners. It is now published and opened for comments and reactions. Feedback will lead to the development of a second updated version.

To ensure effective and robust AI regulation, society must establish an inclusive and horizontal reflection process. While expert groups and political representatives are necessary, delegating the entire regulatory effort to them risks a technocratic or elitist setting, which may not yield the best outcomes.

Genuine governance requires mobilizing all concerned stakeholders. Broad participation is vital not only for the quality and content of the regulations but also for ensuring high social awareness of these rules. Without widespread understanding and involvement, the enforcement and long-term success of AI regulatory devices will be severely undermined.

With AI, ethical considerations begin well upstream of usage. For instance, concerns over biased classification tools, or recommendation engines driven by attention capture demonstrate that the ethical “quality” of outcomes is largely built into the system’s design, not just determined by user choice.

The background reflection society must conduct is one that recognizes the need for ethical deliberation to happen at the design and development stages of AI. This requires participatory input from citizens and stakeholders to embed values directly into the creation process through methodologies like Value Sensitive Design (VSD).

AI ethics and governance raise complex issues which are often ‘wicked problems’ necessitate social transformations and require transdisciplinary co-production of solutions where diverse experts support the collective reflection.

Experts are not expected to bring answers to issues at hand. Instead, their expertise must facilitate processes of self-criticism, mutual enrichment, and self-transformation within social groups.

This collective and transformative work (mapping issues and seeking answers) empowers citizens and stakeholders to contribute to AI guidance. This empowerment is crucial for several reasons:

Engaging in a collective and renewed exploration of humanism, by critically re engaging with fruitful ideas – like Kant’s central focus on freedom, responsibility, and the faculty of judgment—and coupling them with insights from AI and cognitive science.

This deepened, shared understanding of our own human nature is indispensable for developing reliable ethical guidance for AI.

AI is often misrepresented as a limitless, inexorable wave leading toward AGI (Artificial General Intelligence), fueling narratives of human obsolescence or lost control. These misleading representations ignore AI’s crucial dimension as a human artifact, screening off its political and ethical reality. It is essential to secure a not-too-abstract understanding of AI technologies within participatory communities.

• Demystifying machine learning
• Recalling the materiality of AI
• Highlighting the dependance on human intelligence across the pipeline

Resisting the injunction of “outcomism”- the prescription to restrict the discussion of AI’s human-like traits solely to comparing observable outcomes between AI systems and humans. 

Challenging the core assumption that inner, lived experiences are kinds of “black holes” we can say nothing objective or reliable about (as “outcomism” does). By taking lived experience, life and biology seriously, we can distinguish living beings from mere information technology artifacts. This perspective reinforces our basic intuitions regarding what is alive, conscious, and possesses autonomy or free will.

Critically addressing the common linguistic tendency to attribute human traits like “intelligence,” “decision-making,” and a “relationship to truth” to AI and digital systems. It risks eroding the unique ontological status of humans, encouraging us to see ourselves as mere machines. Ethical governance requires that we preserve the possibility that these terms signify a deeper, non-reducible reality rooted in life and human lived experience.

Human communities need to develop AI literacy in relation to such issues associated with knowledge and truth. This literacy is necessary to distinguish among currently existing systems which can deliver trustworthy pieces of knowledge (warranted by human experts) from those that are merely powerful tools for exploration. It is also indispensable for exploring the type of new systems we may develop to bring additional dedicated support in the various facets of our cognitive and epistemic lives. Strong reflection is required to assign AI its proper place for preserving, fostering and prolonging our collective efforts to relate to truth.

We must reflect upon the risks of deskilling. Overuse of AI can impede cognitive development. A thorough analysis is required to identify which skills and lived experiences are indispensable for maintaining the cognitive faculties necessary for human flourishing.

Moving beyond “fixing” or regulating current algorithms and fundamentally re-orient economic models away from attention capture, designing AI to serve as a fertile ground for genuine human collective intelligence and avoid cognitive bubbles.

A profound ethical threat to this agency is the temptation of cognitive offloading. Fueled by “Promethean shame” (a sense of inferiority to machines) and an aversion to risk, humans may delegate decisions to AI to avoid the burden of responsibility and the possibility of error. This surrender is incompatible with decision-making in the strong human sense, which relies on the capacity to sidestep past regularities, exercise creativity, and imagine alternative possibilities. Maybe more threatening than an abrupt takeover, we face the risk of a “gradual disempowerment”—an incremental erosion of human influence over societal systems.

To combat this, collective effort is required to resist the temptation to reduce oneself to an inertial object illusorily relieved from any responsibility or painful exercise of autonomy. We must design systems that empower rather than replace human judgment. We must discuss key questions: What price are we willing to pay to defend human autonomy? How do we ensure algorithms support and empower our strong decision-making abilities, rather than encouraging the illegitimate offloading of responsibility?

Resisting to techno-solutionist shortcuts that uncritically link technological innovation with genuine human progress, assuming every problem has a technological fix. This mindset risks reducing human issues to metrics of efficiency, ignoring root causes in favor of superficial technological patches.

This mechanistic mindset treats any limit, mistake, or vulnerability as a defect to be eliminated. Or, vulnerability cannot be reduced only to its negative aspects, it also corresponds to the fundamental possibility of being affected. Trying to eliminate it is a mistake. Absolute robustness is the negation of life; it is our vulnerability that allows us to love, feel joy, and connect with others.

We should shift from the misleading question about whether AI can make us all-powerful and invulnerable to: How can we develop and use AI systems to help us better tame and balance the ambivalent but essential vulnerability and fallibility that lies at the deepest heart of who we are?

Collective reflection to orientate AI development and use must confront with the important questions about how to foster and cultivate, in the age of AI and, if possible, with the support of AI, our sensitivity to life and to the presence of lived experience of genuine persons. More than that, we must cultivate our ability to assess the decisive value of genuine presence over mere appearance, and to discern where and how it is primary, and how it should be balanced with considerations about outcomes’ quality.