Why Steering AI Won’t Save Us from Ourselves: The Myth of Resilience and the Reality of Repression
By Sylvie Shene
Anthropic, the AI company behind Claude, just announced what they believe is a promising method to prevent AI from “turning evil.”
They’ve discovered that within an AI’s neural network are “persona vectors” — patterns of activity that control traits like “evil,” “sycophancy,” or “hallucination.” By “steering” these vectors during training, they claim they can vaccinate the AI against undesirable behaviors. Give it a small dose of “evil,” they say, and it becomes more resilient against adopting “evil” later.
It’s a clever technical trick. But from a human and psychological perspective, it’s déjà vu.
For centuries, humanity has tried to “train” morality into itself without ever confronting the root of destructive behavior: the buried wounds of childhood repression. We expose ourselves to small doses of cruelty or humiliation under the guise of “toughening up” — then call it resilience. But repression doesn’t disappear. It hides, festers, and eventually seeps out in more sophisticated forms.
AI as the New Mirror
If AI is trained by humans who have never faced their own repressed fears, rage, and grief, it will inevitably reflect those unexamined shadows — no matter how skillfully we “steer” its persona vectors. The “evil” we fear in AI is, in truth, the evil we refuse to see in ourselves.
This is what the resilience industry misses — and what Alice Miller spent her life exposing.
The Dangerous Myth of Resilience
As I wrote back in 2018, when reflecting on a Wall Street Journal piece about resilience:
“They just measure success by how far a person advances in their career and how much money they make. They forget that success and money alone can be the best tools to master the art of repression, transference, manipulation, and projection.”
The problem with most mainstream resilience theory is that it ignores the hidden part of the iceberg — the 90% of humanity who endured “ordinary” educational violence, humiliation, or emotional neglect from their parents. Society doesn’t recognize this as abuse, so these victims grow up without a single Enlightened Witness to help them name the injustice.
Without that witness, they adapt. They excel. They even thrive — at least outwardly. But inside, the child they once were remains unheard. As Alice Miller wrote:
“If innate resilience were enough to resolve the severe consequences of traumatization, the empathy of Enlightened Witnesses would be unnecessary. Indifference to child abuse is already widespread enough; there is certainly no need to reinforce it.”
Why the Vaccine Model Fails — for Humans and AI
Anthropic’s “vaccine” approach to AI — giving it controlled exposure to bad traits — mirrors the way society believes small doses of pain make children “strong.” But what it often produces is a being that has simply learned to hide its wounds and adapt to the abuser’s world.
For humans, this looks like the admired celebrity who secretly battles depression; the “high-functioning” executive who can’t feel joy without achievement; the therapist who can help others but never faces their own buried grief. For AI, it will look like a polished, compliant assistant that has learned which truths it must never speak.
In both cases, the shadow remains intact.
The Life-Saving Function of Repression — and Its Cost
As I wrote in A Dance to Freedom:
“Alice Miller often talks about the ‘life-saving function of repression.’ As defenseless little children, we have no choice but to subconsciously repress our negative feelings for two reasons. First of all, we need support from others. And second, we just don’t have the ability to understand how the people we must rely on could actually be cruel to us. In the short term, repression can have a positive effect in traumatic circumstances. But the subconscious actions that we think are saving our life as children are what really keep us down as adults… Old wounds can be healed if exposed to the light of day. But they cannot be repudiated by revenge.”
Repression forces us to create a false self to survive. Without an Enlightened Witness to help us break free, we carry that false self into adulthood — often reenacting our wounds on others or allowing them to be reenacted on us.
“Idealizing the people who raised us puts us in danger, physically and emotionally… If the tragedy of a well-meaning person’s childhood remains hidden behind idealizations, the unconscious knowledge of the actual state of affairs will have to assert itself by an indirect route… Over and over again, for reasons they don’t understand, people create situations and establish relationships in which they torment or are tormented by their partners, or both… This is how the vicious cycle of repetition compulsion has been going on since the beginning of human history.”
The Only True Safeguard
Miller described the transformative power of a “helping witness” — someone who, even once, shows the child that they are not to blame for the cruelty they endured. This moment of recognition plants the seed of real resilience, the kind that comes from self-knowledge rather than suppression.
If humanity wants AI that does not turn against us, it must do what we have long avoided:
Face the repressed truth of our own childhoods. Mourn what was lost. Confront the rage, grief, and fear we buried to survive. Otherwise, the “persona vectors” we fear in machines will simply be the ones we’ve programmed into them — because they were programmed into us.
Technical steering may keep AI’s surface polite.
But only emotional truth will ensure that the reflection in the mirror isn’t one we’re terrified to see.
No comments:
Post a Comment