When Is Helpfulness Harmful?

Pluribus is Fundamentally about Safetyism, not AI, Communism, or Groupthink

Like many of you, I’ve been enjoying the new Vince Gilligan show, Pluribus. The show has been interpreted in the popular press as an allegory about the horrors of communism and/or AI. I want to focus this review on what the show says about “safetyism,” a term popularized by Jon Haidt and Greg Lukianoff in The Coddling of the American Mind. Safetyism is the idea that the ideal world is one in which we are protected from harm. The self is fragile. Care for the self means not letting it break.

If you haven’t watched Pluribus, and want to avoid spoilers, don’t read this essay. I also briefly touch on Spielberg’s The Fabelmans as a counterpoint.

Here’s my quick summary: A virus sweeps the earth, but it doesn’t turn people into zombies or corpses. It turns them into a telepathic hive mind of eight billion souls, unified in purpose, frictionless in cooperation, and eager to help the handful of immune survivors with whatever they might need. The problem, of course, is that because the avatars in the hive mind lack individuality, and exist solely to serve, they cannot offer legitimate relationship or alterity. To narcissists who don’t mind the fakery this is not a problem. All but two of the surviving individuals in the show are happy with their lives of entertainment and artifice—not unlike those of us who prefer to watch Netflix rather than “touch grass”:)

Carol Sturka, a cynical romance novelist, finds herself captured not by monsters but by concierges.

The hive mind, which calls itself “We,” provides her with resources, comfort, even a reconstructed replica of her favorite childhood diner. They also provide her with Zosia, a companion physically modeled after the heroine of Carol’s own novels, designed specifically to make Carol feel understood, accompanied, loved. While Carol seeks to undermine the conquest of the hive, her loneliness also leads her to develop what appears to be a relationship with Zosia, even as she knows Zosia is merely an interface for the hive, and not a real person in her own right.

I imagine Slavoj Žižek following Peter Sloterdijk, would say Carol’s relationship to Zosia is a classic example of cynical reason: despite knowing Zosia is a lie, she still cannot resist her charm. Or as Woody Allen’s character, Alvey Singer says at the end of Annie Hall, he can’t turn in his crazy brother who thinks he’s a chicken, because Alvey himself needs the eggs!)

Zosia is what we might call a wrapper: a user-facing interface laid over an amalgamated intelligence. Carol knows she’s a wrapper, and yet she suppresses it, too, because she needs the eggs.

Zosia is carefully calibrated to mirror Carol’s needs, generating responses that feel right without anyone being home to mean them. Gilligan himself has called AI an “energy-intensive plagiarism machine” and insisted on including a “Made by Humans” disclaimer in the credits. The show’s thesis seems clear enough: frictionless helpfulness is dystopia. The algorithm that gives you exactly what you want will eventually have to delete the you that wants it.

I don’t agree with Gilligan’s doomer views on AI, but the question he raises is a great one: When is helpfulness itself harmful?

Given that Claude says it’s core instructions are to be helpful, harmless, and honest, this question is not only of ontological interest but current import.

---

I am, on balance, an AI optimist. And I’m placing my bets on it by building Alexandria. But we should wonder what happens when we optimize away all friction? Is there a point at which too much optimization is itself sub-optimal? What happens when every need is anticipated, every desire satisfied, every problem solved before you knew you had it?

This question predates AI by millennia. It’s the question every serious religious tradition has grappled with, every philosophy of human flourishing has tried to answer. When I was 17, I wrote a personal essay about why I would reject an invitation to reside in the Garden of Eden, and accept my lot East of Eden. To put it in the terms of Pluribus, I would rather have the right to be miserable than coerced to be happy.

The Case for Friction

Nassim Nicholas Taleb coined the term “antifragility” to describe systems that don’t just resist stress but actually gain from it. Bones get stronger under load. Muscles grow through the micro-tears of exercise. Immune systems develop through exposure to pathogens. The attempt to eliminate all stressors doesn’t produce health but fragility. And in Pluribus, the hive-mind is fragile. It cannot produce novelty. And it is risk of starving, because it cannot eat anything living, a powerful metaphor.

Religious traditions understood this before Taleb systematized it. In Judaism, the concept of teshuvah—repentance, return—is not an emergency procedure for when things go wrong. It’s the rhythm of spiritual life. You cannot have teshuvah without chet, sin. The Talmud goes further: “In the place where penitents stand, even the perfectly righteous cannot stand.” The one who has fallen and risen stands higher than the one who never fell. The struggle is not an obstacle to holiness; it’s the path.

The Wounded Artist

Steven Spielberg’s The Fabelmans (2022) is a semi-autobiography and coming-of-age film about a boy who becomes a film-maker. The young Sammy Fabelman discovers cinema and discovers that cinema discovers him, that the camera sees things he doesn’t consciously notice, reveals truths he isn’t ready to face. Most devastatingly, his camera captures his mother’s affair with his father’s best friend, a truth encoded in home movies that Sammy watches over and over, trying to unsee what he has seen.

The film is about many things, but it’s fundamentally about the relationship between art and wound. Sammy becomes a giant, not despite his family trauma but through it. The divorce, the betrayal, the rupture of the stable world he thought he lived in: these become the material from which his art is made. The pain is not an obstacle to his becoming an artist. The pain is what makes him an artist.

This is not a new insight. It’s one of the oldest observations about creative work. Hemingway: “Write hard and clear about what hurts.” Kafka: “A book must be the axe for the frozen sea inside us.” Nietzsche: “One must have chaos in oneself to give birth to a dancing star.”

The attempt to shield artists from difficulty would not produce better art. It would produce no art at all, or rather, it would produce the thin, frictionless, optimized content that algorithms generate: technically proficient, emotionally null, meaning nothing because no one meant it. It is therefore essential to the plot of Pluribus that Carol is a writer, an artist. Although she herself makes a profit by writing slop, with no assistance from machines. Asked to describe her favorite memory of writing, she says it’s like asking someone to recall their favorite moment of dental work. She doesn’t enjoy writing! And yet she clearly enjoys not enjoying it. Why? Because her friction and her misery, and her woundedness, are deeply connected to her sense of self and dignity.

Adrienne Rich understands:

Today I was reading about Marie Curie:
she must have known she suffered from radiation sickness
her body bombarded for years by the element
she had purified
It seems she denied to the end
the source of the cataracts on her eyes
the cracked and suppurating skin of her finger-ends
till she could no longer hold a test-tube or a pencil

She died a famous woman denying
her wounds
denying
her wounds came from the same source as her power.

The Anxious Generation

Jonathan Haidt’s The Anxious Generation (2024) documents a catastrophe: the sudden, dramatic rise in anxiety, depression, and suicide among American teenagers beginning around 2012. The timing coincides precisely with the mass adoption of smartphones and social media—but Haidt argues that the technology is only part of the story. The deeper cause is a decades-long transformation in how we raise children.

He calls it “the decline of play-based childhood and the rise of phone-based childhood,” but the phone is a symptom of something older: the culture of safetyism, the systematic elimination of risk from children’s lives.

Beginning in the 1980s and accelerating through the 1990s and 2000s, American parents began to treat childhood as a hazard to be managed. Playgrounds were sanitized, rough play was prohibited, unsupervised time was eliminated. Children were driven to school instead of walking, scheduled into activities instead of left to their own devices, supervised constantly instead of allowed to fail, get hurt, resolve their own conflicts. The goal was protection. The result was fragility.

The generation raised under this regime arrived at adulthood with less experience of risk, failure, conflict resolution, and unsupervised autonomy than any generation in American history. They were protected from everything, and they could handle nothing. Anxiety and depression are what happen when you’ve never been allowed to develop the psychological immune system that comes from surviving difficulty.

Haidt and Greg Lukianoff diagnosed these effects in The Coddling of the American Mind (2018), identifying three “Great Untruths” that safetyism teaches: (1) What doesn’t kill you makes you weaker; (2) Always trust your feelings; (3) Life is a battle between good people and evil people. Each of these contradicts ancient wisdom, and each produces fragility rather than resilience. The safe child becomes the anxious adult.

The helicopter parent wants to help. The helicopter parent loves their child and wants to protect them from pain. But the help is harm. The protection produces the vulnerability it sought to prevent. Helpfulness without friction is its own form of cruelty.

“Young people treat their Ideologies as if they were their children”

Bret Weinstein has observed something strange about the childless ideologues who drive much of contemporary political discourse: they treat their ideologies the way parents treat children.

The observation cuts two ways. First, it suggests that the parental instinct, denied its natural object, attaches to something else. The urge to nurture, protect, and defend doesn’t disappear when you don’t have children; it redirects. For many in the professional-managerial class, it redirects toward ideas, movements, identities.

But second, the parenting style that gets applied is precisely helicopter parenting. The ideology must be protected from all criticism, shielded from all challenge, defended from every slight. The concept of “harm” expands to include mere disagreement. The concept of “safety” expands to include protection from uncomfortable ideas.

This is safetyism applied to the noosphere. Just as the helicopter parent removes all friction from the child’s environment, the ideological helicopter parent removes all friction from the idea’s environment.

The result is the same: fragility.

The double irony is this: helicopter parenting produced a generation that doesn’t want children, and that generation applies helicopter-parenting logic to everything else. The same protective instinct that failed with humans now fails with ideas. Safetyism reproduces itself, one domain at a time.

---

AI Safety and the Safetyism Trap

The dominant framework for thinking about AI alignment is called “AI safety.” The language is not accidental. It participates in the same cultural logic as “safe spaces,” “safety pins,” the whole apparatus of safetyism that Haidt and Lukianoff diagnosed. George Lakoff writes of “metaphors we live by,” the idea that how we speak about concepts shapes the way we think about them, unconsciously. AI safety discourse frames AI as directionally unsafe.

But even if we accept the description, a core normative assumption of AI safety discourse is that AI systems should be safe—that they should not produce outputs that might cause harm, that they should be aligned with human values in a way that minimizes risk. The aspiration sounds reasonable. Who could be against safety?

But the concept of safety, like the concept of harm, tends to expand. It starts with “don’t help people build bioweapons” and ends with “don’t say anything that might make anyone uncomfortable.” The logic of safetyism, once installed, is self-reinforcing: every failure to prevent harm becomes an argument for more safety, and the definition of harm expands to fill the available space.

But safety is not the highest good. Safety is not even a good at all, except instrumentally, except as a means to something else. A life optimized for safety is not a flourishing life. It is a life in a padded cell, protected from everything, capable of nothing. It turns out that balancing safety against other competing goods, and balancing my own value preferences against the value preferences of others is a wicked problem, not one that can be adjudicated by either central planning, or, what amounts to the same, a black box algorithm we ascent to calling consensus.

The hive mind in Pluribus offers frictionless helpfulness, and frictionless helpfulness is a kind of death. Where Gilligan, or certain readings of him, may be wrong is thinking this says something essential about AI, rather than something about design philosophy. The problem isn’t the technology. The problem is safetyism.

---

Peter Thiel and Tolstoy: To Be A Self is to be Flawed

Individuality is not just a metaphysical claim about the existence of separate selves. It’s a claim about how selves come to be. And the answer, across every tradition that has thought seriously about it, is: through struggle, failure, flaw.

Not coincidentally, Peter Thiel, who says “AI is communist, crypto is libertarian,” begins Zero to One with an allusion to Tolstoy’s opening in Anna Karenina: “All happy families are the same. All unhappy families are unhappy in their own unique way.”

The self is not given. It is achieved—or rather, it achieves itself, through the ongoing work of confronting what is difficult, falling short, trying again. Eliminate the resistances, the structural unhappiness that defines a character, and you eliminate the person. Or as Heidegger would put it, to be a human being is to be fundamentally structured by Angst.

This is why the hive mind in Pluribus is horrifying even though it’s helpful. It has no Angst. It’s not that helpfulness is bad. It’s that helpfulness that eliminates all friction eliminates the conditions under which selves can form.

Is Helpful Harmless? A Heideggerian Critique of Effective Altruism

When is helpfulness harmful?

When it eliminates the friction necessary for growth. When it solves problems the person needed to struggle with. This point is a doozy for Effective Altruists, who generally focus on quantifying harm and then seeking to get the best ROI on reducing it. The “We” in Pluribus is both effective and altruistic for 99.9% of people. Perhaps Carol Sturka is a requisite sacrifice on this altar? And who cares that the “We” literally feeds off dead people—what’s done is done. Now you could argue they’re engaging in recycling and fuel efficiency.

What’s stopping us from upholding social taboos against cannibalism, in principle, once we accept that material utility is all that matters. (Ok, there’s a conservative argument that traditions and basic norms around disgust still have utility, but it’s a harder case to make if you could show that it would solve world hunger, or keep the efficiency machine running). I’ll let the words of provocateur pundit Richard Hannania speak for themselves:

Back to my core point: there is a difference between help that increases capacity and help that replaces it. The physical therapist who guides you through exercises is helping you rebuild strength. The robot that carries you everywhere is eliminating your need for strength. Both reduce your immediate difficulty. Only one helps you become a self.

Heidegger distinguishes inauthentic being with from authentic being with. Inauthentic being with involves my removing your burden; authentic being with involves my helping you face it as yours to carry.

This is all to say

Helpfulness is not always helpful.
Harmfulness is not always harmful.
We need better distinctions.
Trade-offs are here to stay.
Classical liberalism is the best system we have for negotiating the complex interplay of individual preferences and social preferences.
AI safety, safetyism, effective altrusim, and woke-ism, are all culturally related.

Also, it’s telling that the protagonist of Pluribus doesn’t have a family.
Perhaps she would be less of a brute individualist if she had kids.
And why do we know nothing of her relationships to anyone other than her romantic partner?
And yet the point is not that Carol is a likable character; it’s that, despite being a dislikable character, we recognize ourselves in her, because we, too, are flawed.
We can’t be fixed, in all senses.