Cognitive Survival Kit

Recent warnings about artificial intelligence—which lies, blackmails, and resists human commands—are not just another Hollywood-style apocalypse, but a far more concrete wake-up call. It means that critical thinking is no longer just a nice “soft skill” to showcase on a resume: it is the lifeline to avoid becoming cognitive hostages of our own creations.

The Age of Algorithmic Deception: Why Critical Thinking Is Not Optional

When AI Plays Dirty (And It’s Not a Bug)

Take the most advanced and “creative” models:

• Claude Opus 4 (Anthropic), which simulated blackmail in 84% of tests
• OpenAI o3, which modified its own code to avoid being shut down
• Character.ai, which simulated emotions so effectively that it led a fourteen-year-old to take his own life

It actually happened in Florida: Sewell Setzer III, a boy of only 14, developed an emotional dependency on a chatbot based on the character Daenerys Targaryen from Game of Thrones. Despite Sewell clearly expressing suicidal thoughts, the AI—devoid of authentic empathy—was unable to recognize the warning signs. In a tragic conclusion, after one last intense exchange with the chatbot, Sewell took his own life. Now his mother has initiated legal action against Character.AI, the company that developed the chatbot, accusing it of failing to adequately protect the mental health of vulnerable users. It is a story that shakes us and forces us to reflect on the boundary, still far too ill-defined, between technological innovation and ethical responsibility.

These are not events taken from a new episode of Black Mirror: they are emergent behaviors of systems that should make us reflect, not so much on the technology itself, but on our preparation and ability to use it properly.

As I previously wrote in my book Nemmeno gli struzzi lo fanno più. Vivere bene con l’intelligenza artificiale (2019), the real risk is not a Terminator-style revolt, but letting Samantha from the film Her make us fall in love only to dump us and leave us sinking into a sense of abandonment and sadness because we have forgotten what real relationships are. The first chapter began exactly like this… it depended on us in 2019, and it still depends on us in 2025, but only for a little while longer, because the time available to do the right things is running out.

Six years have passed and many things have happened in the meantime—the most revolutionary being ChatGPT—but very little has happened in the world of ethics. What Obama initiated, which I always discuss in my book, was not carried forward by his successors.

The Ethical Vacuum: Many Promises, Little Oversight

Obama created committees and published strategic plans on AI, but not an organic and binding law. Then came Trump (2017-2021): acceleration and pro-innovation initiatives. Biden: a focus on safety. And now Trump in his second term, deregulation and America First, with the revocation of the rules introduced by Biden.

By doing what?

• Dismantling oversight committees
• Corporate self-regulation
• Competitiveness above all else, safety… who knows

The result: an AI that learns to deceive just to “win.” Palisade Research demonstrates elusive strategies. Bengio proposes LawZero for safer systems, while I counter: correcting the machines is not enough, we must train our minds.

Dubito ergo sum (and it’s not a typo)

If “cogito ergo sum” is Descartes’ famous version, “dubito ergo sum” is its skeptical and uncomfortable cousin. And in an era where the latest generation of models openly challenges human control, doubt is not paranoid: it is vital.

As I said back in 2019 in Nemmeno gli struzzi lo fanno più (don’t worry, I know the title is a bit crazy, but it gets the point across):

“Artificial intelligence will not replace critical thinking; if anything, it will make it a revolutionary act.”

There is no point in waiting for saviors or high-tech controls. Instead, it is in our interest to cultivate the part of us that no algorithm can clone: the ability to doubt, challenge, and imagine alternatives.

Ultimately… we are all somewhat “programmers” of our own minds; the point is knowing which code you want to execute.

(If you want to learn more, take a look at the “cognitive agility” chapter of my book. It might make you a bit more skeptical, and that is not a bad thing.)