The New Cognitive Aristocracy

This is Not the End of Thinking, Not for Everyone at Least

Very short summary: In this essay, I address the growing concern that AI could herald the “end of thinking” in the near future. While this concern is probably exaggerated, I argue that a plausible scenario is the emergence of a new cognitive aristocracy. As the majority of the population gradually loses literacy and thinking skills, this cognitive aristocracy will strengthen theirs. The emergence of this new aristocracy and its consequences should be compared with those of the liberal aristocracy that many critics of liberalism have targeted.

---

Concerns that technology might transform society for the worse are as old as humanity itself. These concerns extend beyond the economic fear that technological innovation may destroy human jobs. More than two thousand years ago, Plato feared that the invention of writing would undermine cognitive skills like memory and critical thinking. Closer to our time, the invention of television quickly sparked claims that watching it would make us intellectually lazy. Nowadays, the internet and its many applications (social networks, online search, email) have triggered similar warnings that we may lose cognitive capacities such as our ability to focus. The latest avatar in this long list is artificial intelligence (AI), particularly the creation of large language models (LLMs).

An increasing number of people, including myself, have been speculating that AI could have adverse cultural effects that are difficult to assess but nonetheless potentially significant. For instance, in a previous essay, I suggested that the widespread adoption of LLMs could lead to two distinct effects with important political consequences: uniformization and disempowerment. Uniformization is the risk that, because we rely on similar tools to write, think, and even make decisions, we could end up all doing the same things. Because AI may help identify “best” practices and because we are increasingly prone to following the machine’s advice, the diversity of practices and thoughts may quickly shrink. This is bad not only because diversity is valuable in itself, but also because diversity is a desirable property of any “complex adaptive system.” Complexity science has long demonstrated that the more diverse a system is, the more robust it is against exogenous shocks. For instance, when facing a pandemic, it may be advantageous to have diversity in health care systems and containment strategies, simply because this increases the probability that someone will have an effective response to offer.

Disempowerment is an even more serious threat. The use of LLMs is likely to weaken our sense of responsibility. At the individual level, if I ask Claude to improve my draft and the AI produces substantive revisions, it’s no longer clear that the revised text is my own. The difference between using a standard correction software and an LLM is that the latter introduces a form of intentionality that is not mine. The LLM is more than a tool; it’s a quasi-autonomous intentional machine, like a co-author. We can generalize this and speculate that an increasing number of our thoughts and decisions will be “mixed” with the work of intentional machines. A possible outcome is that we will be increasingly discharged (or view ourselves as being discharged) from the responsibility for the results generated by our interactions with AIs. This might be particularly significant at the political level, with political leaders relinquishing control to intelligent machines.[1] The disappearance of the sense of responsibility would undermine human agency, turning us into passive spectators rather than active actors in our destiny.

These scenarios are admittedly speculative. These tendencies may be counteracted by psychological and social mechanisms that we cannot identify today. Perhaps LLMs will not trigger any mechanism of uniformization or disempowerment. However, there is increasing evidence that the systematic use of AI, especially generative AI, may have cognitive impacts that could produce cultural effects of a similar nature. In a recent essay, Derek Thompson reviews some of this evidence and concludes that the “rise of thinking machines” could lead to the “decline of thinking people.”

While this may sound dramatic, data suggest that cognitive performance in various tasks has been steadily declining in many countries over the past two decades. One plausible cause of this cognitive decline is that people are reading and writing less and less as their use of social media and LLMs increases. There are clear indications that literacy (being able to read and write and regularly using those skills) accounts for a large part of human cognitive abilities, beyond the obvious fact that it increases our capacity to disseminate and store knowledge. By reading increasingly fewer complex texts and by delegating the task of writing to machines, we are undermining the sources that make us exceptionally intelligent living beings. Thus, the decay of literacy could signal the decay of human intelligence.

As an aside, Thompson’s account of how LLM use affects medical practice closely describes what I call disempowerment. Practitioners are becoming accustomed to relying on AI for their diagnoses and decision-making, to the point where some are starting to question the authenticity of their thoughts. A decline in cognitive abilities coupled with a sense of disempowerment forms a deadly recipe that undermines not only modernity but the very foundations of what distinguishes human societies from the rest of the animal kingdom.

“The Wandered above the Sea of Fog,” David Caspar Friedrich (1818)

As I said, that may be too dramatic. Regarding the cognitive impact of LLM use, we simply don’t have enough historical insight to make any informed assessment. It’s plausible that people will read and write less because they can delegate these tasks to generative AI. It’s also plausible that this will undermine literacy and other cognitive skills that build on this competence. However, there may also be positive side effects. By delegating repetitive tasks that use cognitive resources to machines, we may be able to allocate more of those resources to unique and more difficult cognitive problems. Machines are already very good—better than most of us—at surveying and summarizing the literature on any specific topic. For those of us whose jobs require us to regularly update our knowledge or acquire new knowledge, AI is incredibly useful. It gives us time to think about really hard issues, to create new ideas, and explore new intellectual pathways. In other words, the principle of comparative advantage still applies, and LLMs may enable us to deepen the division of (cognitive) labor to an extent never seen before.

The problem with doomsayers who claim AI announces the end of thinking is that they generally only consider the aggregate. Literacy will recede and, consequently, people will become less intelligent. That may be true on average, but it doesn’t mean it’s true for everyone, nor that it’s necessarily bad for society. In the most plausible scenarios, (generative) AI will not eliminate all intellectual jobs. Many intellectual roles will still be fulfilled by humans, if only because of comparative advantage. Moreover, not everyone will stop reading and writing. Quite the contrary: in a world where literacy becomes a rarer skill, its value may increase insofar as it’s still needed to correct thinking machines, to feed them with new inputs, and because some humans will still grant significance to human intellectual output.

What is more likely to happen is the emergence of what I would call a new cognitive aristocracy: a subgroup of the general population that maintains or even improves their literacy skills, possibly with the help of AI, to fulfill relatively scarce but highly important intellectual functions in society. Hints of this scenario are already evident; as Thompson notes, some studies point out that the cognitive gap between higher- and lower-performing students is increasing. This reminds us of the well-known fact that we’re not all equal when facing technological change. The economic, social, and cultural effects of technology on populations are not uniform. They are mediated by complex psychological and social mechanisms that may be hard to decipher and even harder to predict.

Contemporary critics of liberalism often argue that the ultimate failure of liberal society has been to abolish the traditional aristocracy only to lead to the emergence of a new one. This liberal aristocracy consists of people who thrive in a cosmopolitan society where personal and professional success depends on the ability to live autonomously and where traditional social bonds are weakened and replaced by more impersonal and formal relations. Not everyone is fit to flourish in this kind of society economic and cultural backlash underlying contemporary populism may stem from the fact that many people resent living in a society where they don’t feel at home. If this analysis is correct, liberal society has created new privileges that are less tied to formal status than to acquired social capacities and skills.

Something similar may happen with the technological, social, and economic evolution triggered by generative AI. Literacy and thinking will not disappear; however, they may become more exclusive. The cognitive gap between the minority that continues to train their literacy skills and the rest of the population that may suffer cognitive decline will increase. New forms of inequality will appear, far more concerning than economic inequality because they directly affect the very ability to be an autonomous agent.

Is this bad for society in general? At first glance, it surely is. The cognitive aristocracy will dominate the rest of the population on every imaginable dimension: political, economic, cultural. But such a judgment depends on one’s implicit “social welfare function.” We cannot exclude the possibility that intelligence will increase so dramatically among the new aristocracy that its positive effects in terms of material wealth and health will trickle down to the rest of the population. That may more than compensate for the loss of autonomy. This mirrors the argument that some liberals who acknowledge the rise of the liberal aristocracy are prone to make. Yes, the liberal society has undermined traditional ways of life and institutions, resulting in many people no longer feeling at home.” The compensation has been an unprecedented improvement in living conditions. Surely this is an acceptable tradeoff, isn’t it?

The success of populist politics strongly suggests that many think differently – or perhaps they just don’t perceive the tradeoff. If AI happens to really undermine most people’s thinking capacities, the rebellion may take a less articulated form, or may never happen at all.

---

[1] Henry A. Kissinger et al., Genesis: Artificial Intelligence, Hope, and the Human Spirit (Little, Brown and Company, 2024), especially pp. 97-8.