Three years, almost to the day, after the release of ChatGPT - a date some will be sure to celebrate - it is also the moment for me to put down in writing everything I have in my head and in my heart after three years spent scrutinising, analysing and dissecting AI in an intensive way. Almost to the point of making myself sick.

My diagnosis is that generative AI technologies are, on the whole, dangerous for humanity.

These dangers, now that I have understood and assimilated them, I can no longer look away from. More still, I cannot conceive of pursuing my personal and professional path - which are intimately linked - while pretending that everything is fine.

Because no, everything is not fine.

These technologies that we have thrown ourselves at - me among the first - in an unreasonable way and, above all, without the distance or the discernment they required, endanger our thoughts and opinions, our reasoning capacities and, much more seriously, our humanity.

And yet, we could have, we should have, suspected it.

Our thoughts and opinions in danger

Generative AIs are machines for producing symbols probabilistically.

They produce the most probable words (symbols) from long and complex calculations based on the analysis of the corpus of everything humanity has been able to create as content and that is available on the internet.

In hindsight, it now appears obvious to me that this intrinsic operation tends to reinforce the production of concepts representing the "most common" meanings, the "status quo" and, in a general way, everything that is "within the norm".

Consequently, it is easy to imagine - and we are already feeling the first effects on social media - that the growing use of these AIs amplifies the generation of content going in this direction of a certain "flatness" and a patent lack of diversity, of points of view and opinions.

The feeling of omnipotence given to the user when these tools make pseudo-reasonings and pseudo-knowledge spring up, from their calculations, in an almost magical (or divine) way, is totally deceptive and tends, on the contrary, to flatten any point of view.

A majority of users seems to be satisfied with this machinic levelling by the norm, notably through the peremptory tone of self-satisfaction of these AIs and other procedures I will detail further on.

This danger is all the more diluted as these generative AIs are at the same time used to obtain simple cooking recipes or - and this is the trap - to write content that will subsequently be published on the internet; substituting almost indistinguishably the opinion or thought of the user with this production of statistically most probable text.

The same generative AI is capable of having a benign externality (unless we see humanity feeding only on pineapple pizza) while the second use is a danger for the plurality of expression, of points of view and opinions.

With all the known excesses such as echo chambers or polarisation of opinions that will be further reinforced.

It is… only from plurality that we can define our singularity. To see and be seen, to hear and be heard in the in-between.Hannah Arendt, Denktagebuch 1950-1973

Our reasoning capacities in danger

Our brain is lazy by definition [1].

By default, generative AIs are facilitators of pseudo-reasoning that, if their uses turn into habits for the user, could well lead in the medium term to the attrition of their cognitive capacities [2][3].

Although there are methods to query (prompt) these generative AIs in a way that reduces facilitation and thereby increases the user's brain work, these methods are neither available by default - via the system prompts installed by the publishers - nor, most often, easily accessible to users.

Therefore, by default, the user accesses in a simple and immediate way pseudo-knowledge and pseudo-reasoning, providing them with instant pleasure and a feeling of omnipotence, to the detriment of a more complex friction that, not so long ago, was called reasoning and intellectual work.

These pseudo-reasonings and the pseudo-knowledge brought in these conditions seem to me to constitute an even greater danger for young audiences, for whom the construction of the thinking system, reasoning and the acquisition of knowledge are - still, I imagine - priorities if we are still to want a flourishing future for them.

I can already hear the objections that compare these generative AIs to Wikipedia or to crude tools like calculators back in the day, to comfort themselves in their vision and their uses. I would refer them, with a wink, to another of my articles where I pseudo-dialogue in a fairly deep way with a generative AI (here Claude Sonnet 4.5) [4]. Neither Wikipedia nor a calculator can corrupt our perception of the real, of reasoning and of knowledge to the same degree as generative AIs.

Our humanity in danger

A new drug by and for Big Tech

This is, in my view, the most serious of the three major risks that the use of generative AI poses to our society. I only became aware of this third risk recently, by piecing together a mental puzzle whose pieces had remained scattered on the table of my thoughts for more than ten years.

As a "technologist", I have been able to witness - and sometimes take part in - this Tech that has evolved at a frenzied pace over the last thirty years. Generative AIs are a revolution for society but an evolution for Tech.

It all started around 2010, when smartphones and then mobile applications, platforms, and finally social media came into being.

In 2014, Nir Eyal published a book that had marked me at the time and that takes on its full crucial importance now as I write these lines.

In Hooked: How To build Habit-forming products, the author explains in great detail how to design mobile or web applications and other digital services in such a way that the user becomes literally addicted ("hooked").

These "specifications leading to addiction" were followed to the letter to design applications like Facebook, Twitter, TikTok, Instagram and the majority of social media platforms. [5]

Through clever mechanisms of reward and psychological feedback - most often unconscious -, vanity metrics (likes, followers, impressions, shares), or infinite scroll fed by content dictated by AI (non-generative); it has been proven that these applications make their users dependent, just as drugs or alcohol can.

I was one of those "addicts", even though I knew the psychological mechanisms as well as all the technological machinery at work in these applications. The oenologist was also an alcoholic, in a way.

The primary aim of these applications is to capture as much of the users' attention as possible because attention is the profit factor (mainly through better ad targeting).

So, for more than ten years, Big Tech has been fighting for one single goal:

Capture all our attention

This captured attention is first turned into data, then into gold. This is the loot that gave birth to these techno-empires.

In fact, it is these exact same mechanisms described by Eyal that are again at work with generative AI.

Where, with mobile applications and social media, Big Tech "only" had tracking information (clicks, likes, actions) and behavioural data I would call semantically weak, here, with generative AI, these same companies (most of which are American) will be able to obtain semantically strong data.

It is in fact an unspeakable euphemism to call them semantically strong data when in reality they represent an inestimable value.

They are everything that no generative AI will - I hope - ever reach, namely: the sensibility, the affect, sometimes the intimate, of each of the users who interact with their interfaces.

An operating mode optimised for the AI paradigm

The principles described by Nir Eyal have been adapted to address the new types of human-machine interactions at work in generative AI interfaces (conversational AIs).

Like mobile applications, platforms and social media, conversational agents use and abuse practices known as dark patterns.

These can be defined as a set of application design practices that implicitly manipulate a user's behaviour against their intentions, often because of financial incentives (Gray et al., 2024).

Let us mention here the main dark patterns of conversational agents [11]:

These practices, designed by the publishers, take part in the user's addiction to these new interfaces by coming to "solve", in turn, and according to the prompts and interactions, the triggers linked to uncertainty, to cognitive effort, or, worse still, to the existential loneliness of users, then caught in the trap.

This same existential loneliness which we now know is partly the consequence of a hyper-connection and a dependence linked to social media [7][8][9].

AIs can take part in precipitating the most fragile and least informed users into a psycho-cognitivo-emotional dependence that is the consequence of the very operation of their interfaces.

Or rather, of the way their designers chose to design them.

Because it would all the same have been possible, in my view, to eliminate - or at least to drastically reduce in normal operation - the dark patterns but also and above all, to design these interfaces in a way that does not stimulate harmful externalities of the kind that make the user dependent.

It would have been another design choice, implying other uses but also… a completely different story.

On the contrary, as designed, these applications can only amplify, or even generalise, phenomena of potentially pathological drift.

Some of which are the direct consequences of dependence on platforms and social media, already at work for several years.

This erasure of the self and of the other in this assemblage of mobile applications + social media platforms + conversational AI is, for sure, a detonating cocktail that is developing exponentially and endangers our own humanity and society (the very concept of Society) in an anthropological redefinition of what Man is.

Towards the end of the concept of "work"

Humanity in danger finally, through the potential job destructions announced by the limitless development of these technologies.

If the last industrial revolution did replace the workers by shifting many jobs to the newly created tertiary sector; it is more than probable that this one will not generate a shift to any kind of quaternary sector.

Indeed, even with all the imagination in the world, it is hard to glimpse a future where there would remain a place for human labour in a world populated by super-intelligent AIs and humanoid robots.

And it is not the wild dreams of Jeff Bezos with Prometheus [14] or Google's definition of AGI [15] that will contradict us.

The race to substitute all human labour, whether physical or intellectual, is well and truly launched at full speed without any safeguard or even any consideration of what will become of these new algorithmic displaced: Humanity.

The announced disappearance of the concept of employment - of work - will inevitably impact man and society in a profound way because it is one of the pillars of our relationship to ourselves and to society.

What we have before us is the prospect of a society of workers without work, that is, deprived of the only activity left to them. Nothing worse could be imagined.Hannah Arendt, The Human Condition

Conclusion

To quote the German philosopher Byung-Chul Han, it would appear that we have moved from Homo Faber [12] (man who makes), to Homo Digitalis [13] (man who taps - on his smartphone) and that generative AI places us before the existential choice of our evolution towards Homo Nihilis (nothingness-man).

A new era where all our individual and collective intelligence, but also all our humanity, would be done being transcribed into a machinic entity.

Thus completing the utopia of those who think that man can - and must - be surpassed and that this utopia will be realised through technology, with the complicity of all-powerful capitalism, but above all, no matter the cost.

This design, this particular conception of the world, I do not share.

I could obviously go on as if I did not know.

Hannah Arendt, who has decidedly never been as relevant as today, wrote in her book Eichmann in Jerusalem: A Report on the Banality of Evil:

It is the absence of thought and judgement that makes all totalitarianisms possible.

Unlike Adolf Eichmann - who was the architect of the Final Solution during the Holocaust -, I do not have this absence of thought and judgement and I am facing a case of conscience that, almost viscerally, compels me to speak and to act for what I believe is right.

That is to say, not this technology, not this AI.

I will end this article by quoting the philosopher Eric Sadin in his latest book - which I recommend - Le désert de nous-mêmes:

Consciousness is not to cry fire when the house is burning, but to be alert, that is, to put oneself in a position to grasp phenomena at the very moment of their germination, in order - relative to those bearing perils - to become active before they develop, consolidate and then become irreversible.Eric Sadin, Le désert de nous-mêmes

I sincerely believe that there is a path - certainly narrow, certainly difficult - that would combine meaning and action without giving in to the utopian sirens of "Stop AI".

This path, we could define it and walk it together - because everything remains to be done -, and I know some of you are also ready to turn these feelings into real actions.