Pseudonymize your documents, and include the values in your AI deliverablesPseudonymize your documents consistently, and restore (de-pseudonymize) the original values in your AI assistants' outputs
Pseudonymize your documents, and include the values in your AI deliverablesPseudonymize your documents consistently, and restore (de-pseudonymize) the original values in your AI assistants' outputs
Marvin Systems CEO
At a conference organized by the Chinese government in early November, Chen Deli, senior researcher at DeepSeek, highlighted the paradox: technological optimism vs. social pessimism, or the fact that rapid progress toward AGI (Artificial General Intelligence) could destabilize our societies, despite immediate and measurable technical benefits.
One thing is clear: it seems difficult to assess the long-term risks to our social, economic, and political structures when the horizon is blurred by the euphoria of innovation and the fantasy of omnipotent super-intelligence.
Chen Deli calls on AI companies to be aware of the risks. They should, he argues, play the role of “guardians of humanity.”
One might imagine a responsibility shared among all citizens of the world, but that seems somewhat illusory.
The reality is that the keys to understanding the technical, economic, and governance issues are held by a handful of players. And the wider the gap becomes, the more society as a whole is reduced to the role of a large-scale beta tester, with no real say in architectural choices or ethical decisions.
Can we talk about organized abuse of weakness? Governments and regulators themselves seem reduced to the position of powerless spectators, condemned to react rather than anticipate.
Is humanity being held hostage by a technological race? Technological optimism is not a choice, but rather seems to be an imposed dogma.
AI companies set the pace, the rules, and the narratives. We talk about cognitive asymmetry (those who design the systems have an advantage in terms of expertise, infrastructure, and data) and institutional asymmetry (they directly influence norms and technical standards).
Everywhere, attempts at regulation are struggling to keep pace: the European Union is moving forward with the AI Act, but the texts often remain general and declarative, while companies operate on innovation cycles measured in months or even weeks.
Meanwhile, systemic risks (mass unemployment, social polarization, loss of sovereignty, and even loss of meaning) are becoming very real. They reveal the growing gap between the pace set by the tech giants and the ability of the rest of society to keep up, understand, and organize itself to adapt.
Today, we are able to quantify productivity gains and cost reductions to a certain extent. But how can we model the impact of AGI on social cohesion, democracy, or even the human psyche?
Exponential risks (algorithmic bias, technological dependence, loss of collective autonomy, cognitive inequalities) currently fall outside the scope of traditional analytical frameworks.
The solution may lie in establishing ethical safeguards for more transparent innovation.
It took decades of research, IPCC reports, international agreements, and public education for a collective awareness of climate change to begin to emerge.
In the case of AI, some actors, such as UNESCO, are beginning to propose competency frameworks to help students and teachers understand the challenges, opportunities, and limitations of these technologies. But we are still in the early stages: massive investment is needed in AI literacy that is accessible to all.
In addition to access to technological knowledge, it may be relevant to support the funding of independent R&D programs to counterbalance the narratives.
Let's start by rebalancing the power relationship between technology and society, giving more power to individuals and communities.
AI education is not a luxury: it is a minimum requirement to ensure that democratic debate is not hijacked by a few technology hubs.
Are we creating a tool to serve a collective project, or are we allowing the logic of perpetual optimization to redefine what it means to work, decide, create, and even exist in society?
AGI is not inevitable, but a societal choice (or rather a series of micro-choices): where we place public investment, what uses we authorize or prohibit, what forms of power we agree to delegate to opaque systems.
It is these choices, rather than the power of the models alone, that will dictate whether AI becomes a multiplier of resilience or an accelerator of fragility.
How do you perceive this dilemma? Do you think that the risks of AGI are sufficiently taken into account today, or do we need to radically rethink our approach to innovation?