🔗 Share this article

This recent season, I found myself navigating in gridlocked traffic on the sweltering streets of a southern French city. At a crossing, my companion in the front seat suggested a right turn toward a famous spot for fish soup. Yet, the navigation app on my phone commanded us to go forward. Fatigued and in a stifling car, I followed the app's advice. Minutes later, we were stuck at a roadwork site.

A seemingly trivial incident, maybe. But one that encapsulates a defining question of our time, where technology permeates nearly every facet of existence: who gets our trust more – other people and our personal instincts, or the machine?

The Enlightenment's Promise and A Contemporary Backslide

The renowned German philosopher Immanuel Kant famously described the Enlightenment as "humanity's emergence from its self-inflicted immaturity." This state, he argued, "represents the inability to use one's own understanding without guidance from another." For centuries, that guiding "other" for society was frequently the clergyman, the king, or the feudal lord – entities purporting to speak for divine will. To explain natural events like volcanic eruptions, people looked for explanations in religion. In organizing the social world, from economics to matters of the heart, faith acted as the primary guide.

“Sapere aude!” or “Have courage to use your reason!”

Kant argued that humans had the ability to think rationally. They simply lacked the confidence to employ it. With revolutions in America and France, a fresh era emerged: logic would supplant blind faith, and the intellect, liberated from authority, would become the engine of advancement and a more ethical world.

Now, two and a half centuries later, one might question if we are slipping back into a state of dependency. An app suggesting a direction is just one example. AI risks becoming our new "other" – a silent guide that influences our decisions and actions. We risk ceding the hard-won courage to reason for ourselves – and now, not to gods or kings, but to lines of code.

The Swift Adoption and Subtle Dangers of AI Dependence

ChatGPT was launched only in 2022, and yet a global survey found that an overwhelming majority of people had engaged with artificial intelligence in the preceding half-year. Whether deciding on a breakup or choosing a candidate, individuals are looking to algorithms for guidance. Data suggests a large majority of user prompts concern non-work topics. Even more intriguing than our use of AI for judgment is what happens when we allow it express for us. Composition is now among the top uses for generative AI, just behind everyday tasks. The celebrated American author Joan Didion famously said, “I write purely to find out what I am thinking.” What transpires when we cease writing? Do we cease discovering?

Worryingly, emerging research hints the answer may be affirmative. A study from the Massachusetts Institute of Technology used electroencephalography to observe the cognitive activity of participants who had could use AI, Google, or nothing. Those who could use AI showed the lowest cognitive engagement and had trouble accurately recalling their own work. Maybe most troubling was that after a few months, participants in the AI group became increasingly reliant, copying entire blocks of text.

“Inertia and fear,” Kant wrote, “are the reasons why so great a proportion of men … remain in perpetual nonage.”

Of course, AI's appeal lies in its efficiency. It is fast, minimizes work and – critically – offers a novel way to abdicate accountability. In his 1941 book, Escape from Freedom, the German psychoanalyst Erich Fromm argued that the rise of fascism could be understood by a preference to surrender personal freedom in exchange for the reassuring certainty of subordination. AI presents a digital method of surrendering the weight of having to think and choose.

The Black Box Problem: Trust Without Understanding

AI's greatest allure is its ability to perform tasks beyond our minds – sifting through oceans of data at unprecedented speed. Stuck in the car in Marseille, this was, ultimately, why I opted to trust the machine over my friend (a decision she took as an affront). With access to real-time information, surely the app had superior insight – or so I thought.

The core issue is that AI operates as a opaque system. It generates answers, but not always deepening human understanding. We cannot fully grasp how AI reaches its conclusions – even its creators admit this. Nor can we verify its logic against clear, objective criteria. So when we heed AI's recommendation, we are not being led by logic. We are back in the domain of belief. In dubio pro machina: when in doubt, side with the machine – that could be our future guiding principle.

Using Without Losing: The Critical Balance

AI can be a formidable ally for humanity in scientific pursuit. It can help inventing drugs, free us from "bullshit jobs", or handle administrative chores – duties that demand little thought and are unfulfilling. All to the good. But Kant and his peers did not advocate for enlightenment just so humans could optimize chores or have extra free time. Critical thinking was not merely about efficiency – it was a discipline of freedom and human emancipation.

Human thought is inherently messy and error-prone, but it compels us to debate, to doubt, to test ideas – and to recognize the boundaries of our own understanding. It builds confidence, both individually and collectively. For Kant, the exercise of reason was never solely about information; it was about enabling people to become agents of their own destinies, and to oppose domination. It was about creating a ethical society grounded in the shared principle of rational discourse, rather than blind belief.

With all the advantages AI brings, the key question is this: how can we leverage its potential of superhuman intelligence without undermining human reasoning, the bedrock of the Enlightenment and of liberal democracy themselves? That may be one of the central dilemmas of our century. It is a question we would do well not to outsource to the algorithm.