DayFR Euro

WHEN AI BLEACHES OBAMA | SenePlus

(SenePlus) – Could artificial intelligence be the amplified reflection of our own prejudices? This is the central question posed by Le Monde Diplomatique in its November 2024 edition, through a detailed investigation which deconstructs the myth of technological neutrality. “What is more neutral, they say, than a computer? Error: behind their cold verdicts, algorithms and automatons encapsulate all the biases of the humans who design them”, immediately warns the monthly.

The most striking example of these biases concerns Pulse AI, developed in 2020 by Duke University in the United States. Le Monde Diplomatique reveals how this system, supposed to depixelate images, “tended to whiten people of color to the point of generating a ‘white Obama'”. This failure is not trivial and finds its origin in a technological domino effect: Pulse used StyleGAN, another AI system developed by Nvidia, which “spontaneously over-represents white men due to its own learning.”

“If the Pulse algorithm contained no intrinsic bias, it indirectly integrates those of StyleGAN: when it depixelates the real face of Mr. Barack Obama, the program makes him a white man,” explains the newspaper, highlighting how prejudices are transmitted from one system to another.

The monthly emphasizes that these biases are not simple programming errors. “Without necessarily being aware of it, engineers transpose the discriminatory biases inherent to the conditions in which the data which feeds the machine are produced,” analyzes Le Monde Diplomatique. This situation is all the more worrying since “aren’t machines deemed objective and devoid of ideology?”

The concrete implications of these biases go far beyond the theoretical framework. “Certain litigants, victims of predictive policing algorithms which integrate discriminatory variables, will learn the hard way that this is not the case,” warns the newspaper.

A worrying opacity

Even more alarming, Le Monde Diplomatique reveals that the very understanding of these systems eludes experts. “‘Explain’ how AI models work, that is, translate the system’s response into a series of steps linked together by what a human being can sensibly interpret as causes or reasons” has become secondary. This explanation “is no longer one of the prerequisites for putting a model into operation, but rather serves as icing on the cake.”

The journal goes further by revealing that “even the most advanced researchers struggle to understand what is happening in the black box of algorithms.” This situation poses a fundamental democratic question: “How can the legislator establish standards for evaluating systems of which no one knows how they work, particularly in the sensitive sectors of health or education?”

Le Monde Diplomatique asks: “Is it up to private companies to decide alone the objectives pursued by these AI?” These “fundamental technical-political choices would nevertheless justify collective deliberation and closer public control.”

The newspaper recalls that these systems are developed by “a handful of actors whose financial capacities and expertise in industrial policy follow those of certain G20 countries.” This concentration of technological power poses fundamental democratic questions.

Towards a necessary redefinition

Faced with these challenges, Le Monde Diplomatique calls for a profound questioning of our approach to AI. The journal even suggests abandoning the term “artificial intelligence” in favor of “computational automata”, an expression “clearly less flattering but fairer since these machines achieve their objectives by calculating the best way to repeat past results.”

British writer James Bridle, quoted by the monthly, asks the essential question: “Can we imagine information and communication technologies that do not exploit us, deceive us or supplant us?” The answer, according to Le Monde Diplomatique, necessarily involves “a separation of the State and the market” and a democratic takeover of these technologies which shape our future.

The journal concludes that the aggregation of masses of data cannot replace democratic reflection and critical dialogue. A crucial reminder at a time when AI is intruding ever more deeply into our daily lives.

-

Related News :