Artificial Intelligence: A New Geostrategic Weapon

Artificial intelligence (AI), once seen as a far-off fantasy reserved for science-fiction genre authors, has now become an inescapable reality in contemporary armed conflicts. Far from being a mere technological project, AI is now making its presence felt on the battlefield, redefining the very notion of military power. This transformation is not just the work of governments, but is the result of an ecosystem of hybrid players, combining traditional defense companies with new technology giants. As in the wars in Ukraine and Gaza, AI is playing a central role, upsetting geopolitical and economic balances by changing the way conflicts are fought and perceived. These theaters of war illustrate two distinct facets of the use of AI: in one case, as a strategic tool for turning the balance of power against a more powerful adversary, and in the other, as a technology for perfecting already formidable weapons systems. However, this unbridled digital arms race raises crucial questions: can AI really make warfare “cleaner“? Or, on the contrary, does it exacerbate the risks of error and disempowerment in military decision-making?

These issues invite us to question not only the operational impacts of AI, but also its influence on the legal and ethical standards that frame international humanitarian law.

I. History of AI
 

Artificial intelligence (AI) is a computer system for sorting, processing and storing large quantities of data. It relies on algorithms, which are precise sequences of instructions designed to accomplish a specific task. Algorithms work like “recipes“: a list of steps which, when followed in a defined order, produce a result. In simple terms, an algorithm combines raw data with a series of instructions to produce a targeted result.

For example, if a non-AI program is asked to find the image of a celebrity like Obama, it will scour its database, rather like a digital library, in an attempt to locate an image associated with Fred's name. If it finds a photo correctly tagged “Obama“, it will be able to respond to the request. On the other hand, if it comes across an image without a corresponding label, the program won't be able to show it. This reflects the rigidity of traditional programs, which can only process information that is already known in a straightforward way.

AI, however, takes a far more sophisticated approach. Through machine learning techniques, the program is able to adapt its search processes according to the data it is presented with. To learn to recognize Obama, for example, it will first analyze numerous images of him, spotting characteristic features that don't change, such as the shape of the eyes, the size of the nose or the mouth. At the same time, it will identify elements that are likely to vary, such as the color of clothing or the background. Thanks to this learning phase, the program creates a kind of digital imprint of Fred, enabling it to recognize him even in new situations. This ability to adapt is one of the pillars of modern artificial intelligence.
However, the history of AI is far from linear. Since its inception, it has alternated between periods of dazzling progress, marked by high-profile successes, and phases of stagnation or regression, often due to disillusionment with its limitations. We are now in a phase of renaissance, where AI technologies are once again arousing enthusiasm, but with a heightened need for discernment in the face of the challenges they pose.
 

To understand the challenges of AI, we first need to understand what biological intelligence (BI) is. In 2024, biological intelligence is defined as the study of cognitive science, which focuses on the mechanisms at work in human thought. These mechanisms enable us to perceive, speak, move, memorize, reason, plan, and even be creative and abstract. The idea of simulating human intelligence has prompted researchers to develop systems capable of mimicking certain cognitive functions. Two main approaches can be distinguished here: the symbolic approach and the connectionist approach.

The symbolic approach seeks to reproduce human logical reasoning, using a set of rules, procedures and knowledge. It is similar to the way knowledge is transmitted in schools: step by step, according to a structured process. As INRIA researcher Thalita Firmo explains, this approach is based on the transmission of explicit knowledge. For example, if you teach a pupil how to add a number, you can formalize this process into a series of steps that the machine can carry out much more quickly. This is the approach from which expert systems emerge, capable of reproducing the reasoning of a specialist to solve a specific problem. In the 1980s, this method enjoyed considerable success, before showing its limitations in more complex, less formalized situations.

The connectionist approach, also known as deep learning, focuses on imitating the way humans learn through experience. It is inspired by the way human neurons work, and enables machines to learn by trial and error, from a multitude of examples. It's like learning to ride a bike: you progress by practicing, failing and correcting your mistakes. Today, this method is at the heart of AI developments, as it enables us to process vast datasets to find complex patterns that symbolic approaches cannot capture.

For Frédéric Alexandre, a researcher in computational neuroscience, being intelligent means knowing how to adapt rapidly to complex situations, finding not the optimal solution, but the most satisfactory response. This implies the ability to distinguish what is important in a flood of information, to react to the unexpected and to adapt if conditions change. Biological intelligence possesses this flexibility, fuelled by emotion, motivation and creativity - aspects that are still lacking in today's AIs, which remain confined to the specific tasks for which they have been programmed.
 

Algorithms, the engines of artificial intelligence, are now omnipresent in our daily lives. They manage content recommendations on social networks like Facebook, optimize journeys thanks to GPS location, warn of traffic jams, and even hide behind facial recognition systems used in security and mobile applications. Every action, every choice, is influenced by an algorithm that has learned, analyzed and adjusted its behavior according to our previous interactions.

Computing, by its very nature, aims to process information automatically. So, once an algorithm has been translated into computer language (coding), it becomes a program capable of executing these instructions without human intervention. Whether in autonomous cars, intelligent household appliances or the optimization of our online navigation, AI is quietly shaping our environment.

AI is not limited to our personal uses: it profoundly influences geopolitical dynamics and cultural battles. Tech giants like Sam Altman (OpenAI) and Elon Musk are playing a key role in this new era. While Musk, through X (ex-Twitter), seeks to promote a vision of direct democracy, OpenAI, backed by Microsoft, is at the heart of an ideological struggle over the future of AI. Silicon Valley, once united, is now divided, with part of its elite tilting towards the ideology of the American far right, influencing the algorithms that structure our online interactions.

These algorithms are not neutral. They amplify certain content and hide others, shaping the international agenda. The great difficulty lies in the governance of these tools: how can we ensure that these technologies, in the hands of private players, serve the public interest rather than partisan political agendas?
Elon Musk, in particular, is seeking to turn X into a space of “direct democracy“, where power is supposed to return to the people. However, by destroying traditional checks and balances such as journalists, he is seeking to reshape the public space according to his vision, raising profound questions about the future of free speech and public debate.

...