The Philosophy Baccalaureate, a key moment for hundreds of thousands of students, took an unexpected turn in 2025. Among the 530,000 candidates who tackled four different subjects:
- “Does our future depend on technology?“,
- “Are we free in all circumstances?“,
- “Is truth always convincing?“,
- “Do we need art?“.
Programmed to write a philosophical essay, this AI took on the challenge of the subject “Is truth always convincing?“.
But did it manage to convince a human grader? A look back at a bold experiment and its surprising results!
An AI facing philosophy: a bold gamble
For this unprecedented experiment, the France 3 Hauts-de-France team submitted a generative artificial intelligence to the philosophy baccalaureate exam, asking it to write a complete essay, with an introduction, development, and conclusion, just as a high school student would.
The objective? To evaluate whether a machine, fed with data and algorithms, could produce philosophical reflection worthy of a human candidate. The chosen subject, “Is truth always convincing?“, invites exploration of the tension between the rigor of truth and its ability to gain acceptance.
The AI thus set about the task, structuring its argument like an exam paper.
Its essay was then entrusted to a professor qualified to grade the baccalaureate, who examined the text with the same rigor she would apply to a student’s paper.
The verdict? Mixed, but rich in lessons!
The AI taking the Baccalaureate: A mixed paper, between ambition and shortcomings!
From the first reading, the grader noted significant flaws…
First pitfall: the structure of the essay. While the AI attempted to organize its argument, it failed by having a plan that was too apparent, almost mechanical. “In philosophy, a good argument doesn’t reveal itself like a bulleted list,” explains the professor for France3 Région.
A successful essay must allow the main thread to emerge subtly, without brutally exposing its articulations. The AI, however, delivered a rigid construction that betrayed its algorithmic approach.
Even more problematic, the content of the paper revealed conceptual weaknesses.
The artificial intelligence reformulated the initial subject as “Is truth enough to convince?“, a nuance which, although appealing, moved it away from the question asked. This semantic shift prevented an in-depth analysis of key terms, such as “truth” or “convincing“.
“The AI proposes an interesting plan, but it doesn’t justify it,” notes the grader.
The arguments, though coherent on the surface, lack clear definitions and precise examples to support the reflection. Where a student would have mobilized philosophical references, such as Socrates, Kant, or Nietzsche, the AI contented itself with general ideas, without theoretical grounding.
Also read on this topic: Will artificial intelligence replace teachers in a school in the United States?
A harsh, but revealing, score?
Faced with this paper, the professor found herself in a delicate position. The essay, though original, suffered from a lack of depth and a certain inconsistency in its argumentation. “It’s almost ungradable,” she confides, emphasizing that the paper did not fully meet the expectations of a philosophy exam.
However, to give an idea, she would award an 8/20 to this work, a score that reflects both the effort of structuring and the glaring shortcomings.
“A student capable of such a plan would have been able to fill these gaps and produce a much more accomplished reflection,” he adds to the newspaper.
This result highlights a fundamental limit of AI in a field like philosophy, where nuance, sensitivity, and the ability to mobilize cultural knowledge are essential. While artificial intelligence can imitate the form of an essay, it still struggles to grasp its spirit.
Also read: Here are the techniques students use to outsmart AI detection
Could the AI have done better?
Undoubtedly, the AI’s performance on the 2025 Philosophy Baccalaureate subject depends closely on the quality of the prompt and the information provided to it.
An artificial intelligence, even an advanced one, is not autonomous in its thinking: it relies on the instructions and data it is given. By refining the prompt, specifying the importance of defining key concepts like “truth” or “convincing,” or requesting explicit philosophical references, it would be possible to significantly improve the paper.
Moreover, on platforms like Yiaho, where usage is unlimited, there’s nothing to stop users from endlessly reworking the text, adjusting arguments, strengthening the structure, and enriching the content!
With several iterations, a well-guided AI could theoretically produce a philosophical essay that is not only correct but worthy of an excellent score on the baccalaureate.
The key lies in human guidance to orient and polish the machine’s reflection!
An experience that raises questions?
This initiative raises questions about the role of artificial intelligence in education, as recently illustrated by the framework for the use of AI in education. It also prompts reflection on the nature of human thought.
Can we truly expect a machine to equal the creativity and depth of a mind trained in philosophical reflection? At this stage, this experiment demonstrates that, although AI is capable of producing structured text, it remains far from the analytical finesse of a well-prepared high school student.
AI excels at synthesizing general ideas, but it struggles to grasp the subtlety of philosophical concepts.
The AI may have scored an 8/20, but it pushes us to reflect on what makes thought unique. And what if, ultimately, that was its greatest lesson?
One might also regret that this year’s Philosophy Baccalaureate questions did not address the theme of artificial intelligence. But that’s another debate!
Source: France3-Region
