Confidence bias in algorithms

Paradox: faced with AI, we doubt our human judgment while blindly trusting the algorithms supposed to detect them. This fear weakens us, whereas the real issue is our relationship to our humanity, regardless of the tools used.

The unconscious excess of trust in machines

During a meeting bringing together the examiners for the entrance exam of a major French school in which I participated, the question of artificial intelligence was raised, along with ways to detect potential uses of AI in the application files submitted by candidates.

One person intervened to assert that, generally, one could sense whether a text had been written by artificial intelligence or by a human. Others reacted by highlighting the usefulness of software and algorithms that could detect whether AI had been used or not. These algorithms were presented as reliable tools, providing certainty that the text had been generated by artificial intelligence, a certainty that humans themselves could not achieve.

I then spoke up to point out what seemed to me to be a contradiction. In this supposed critical spirit regarding the use of artificial intelligence, we ended up trusting an algorithm more than our own capacity for discernment to detect the presence of algorithms in writing!

I found this both ironic and somewhat sad: those very people who advocate vigilance against the mechanization of thought are sometimes the first to fall into the trap of technological seduction, convinced that the machine is more competent than humans.

Reinventing our anthropology

What I actually think is that the real issue does not lie in the use or non-use of artificial intelligence, but in our relationship to our own humanity, regardless of the tool we use.

Lautréamont, in certain pages of Les Chants de Maldoror, copied extracts from the dictionary. And yet, his use of this cold and technical material is remarkable: he integrated it into his project of an epic poem of death, giving it unprecedented power.

I believe that what unconsciously drives many of our reactions to AI is fear. The fear of being dominated. And paradoxically, it is precisely this fear that makes us fragile and impressionable. By losing confidence in our own capacity for judgment, we open wide the door for others, including machines, to think in our place.

Fear anesthetizes us, it freezes us. That’s why we must first fight against this fear, so as not to lose our grounding. Because if we give in to this fear, we allow ourselves to be trapped.

One might object that there are areas where AI is objectively more effective than humans. For example, in the medical field, an artificial intelligence with an immense database will undeniably be more efficient at establishing a precise diagnosis. I don’t dispute this. Doctors should not fear being replaced either. On the contrary, they should see AI as a tool, a means to improve care and serve humanity. And even to develop it, for example by having more mental space to consider the patient as a person and not as an object.

Similarly, in the field of writing, if an author has been able, with talent, to use artificial intelligence to deepen their subject and produce a work greater than what they could have created alone, then so much the better. They will not have lost their humanity, they will simply have used a tool with lucidity, consciousness and creativity, for even more humanism.

Two biases to overcome in the relationship with AI

• The excessive trust in these software programs recalls the positivist illusion of the 19^th century, according to which science would provide absolute certainties about the world. However, as the philosophy of science has shown, scientific knowledge is never univocal or definitive; it is a process of interpretation, falsifiability and questioning (Popper, 1934). To believe that an algorithm can offer a flawless answer about the origin of a text is to adopt a scientistic posture that ignores the complexity of language, intention and style.
• The attitude of delegating to the machine the power to say what is true and what is not reveals a process well described by Hannah Arendt in The Crisis in Culture (1961): the renunciation of one’s own judgment in favor of an instrumental rationality that claims to be self-sufficient. By seeking algorithmic assurance, we deprive ourselves of our own ability to discriminate, contextualize and interpret. We let the machine impose its own frame of reference, even though this frame is a human construction, dependent on programming choices, statistical biases and cultural conventions. Thus, we choose to abandon our critical thinking, that is to say, to think in an informed way for ourselves.