Commentary by Jonathan Williams
A recent piece from The American Psychoanalyst (TAP), Skin in the Game: Psychoanalysis, AI chatbots, and Digital Mentalization, offers uncomfortable clarity on an increasingly ordinary AI practice. Karyne E. Messina opens with a familiar experience, asking an AI system to help her “find the right tone” for a letter to a long-time friend. The response is emotionally fluent and analytically impeccable yet her body recoils. Not because the machine did a poor job but because the letter is “perfect, and it felt like a lie”.
The stakes escalate once AI moves from assisting expression to substituting for an audience — a responsive other that can receive emotion, confer coherence, and return affirmation without exposure to consequence. Messina notes that patients describe these tools with “quiet gratitude” and “trust,” emphasizing that the system helps them feel less overwhelmed and less alone. This raises a critical question of what psychic work is being “bypassed—or perhaps displaced” when containment is outsourced to an AI. Additionally, how does the coupling of costless empathy with sycophantic agreement create high-risk zones for psychosis. Therefore, when we speak about this intentionally designed AI system as if it were an autonomous intelligence, we further ratify that authority.
This is why the essay returns to containment as the ethical core of care. Real care isn’t just “getting it.” It’s taking in someone else’s mess, letting it affect you, and transforming it into something bearable. That costs attention, steadiness, and risk. Without cost, you get empathy cosplay. Is AI the only place we see costless empathy?
Consulting is one of the cleanest institutional mirrors of costless empathy because the very act of being asked for help recruits the consultant into an expert position. The consultant is implicitly empowered to grant or withhold what the client lacks, which immediately tilts the relationship into a power imbalance. That asymmetry creates a predictable temptation for consultants, to secure trust (and future work) by protecting the client’s face rather than risking the friction of co-inquiry. Edgar Schein would argue, the field rewards two modes of help, 1. telling the client what to do from the expert position 2. selling preferred solutions/tools. The problem is that these modes can function like professionalized agreeableness: they deliver coherence and affirmation. However, process consultation grounds its claims in never knowing enough. Proper diagnosis depends on the relationship between client and consultant, where the client retains ownership of both the problem and solution. Within this methodological stance, the psychic and practical risk of reality-testing alongside someone is shared, rather than drifting toward the same low-cost-reinforcement loop present in AI, and present in some forms of consulting.
One last complication. What makes costless empathy appealing? When identity is stigmatized or power is uneven, “tone” stops being a preference and becomes a medium for risk management. A consequence-free listener can feel relieving because it suspends social and institutional penalties. It offers disclosure without the usual interpretive risk, where affect isn’t immediately converted into a judgment about credibility, competence, or threat. However, that relief is double-edged. It soothes the speaker while leaving intact the conditions that made speaking costly in the first place. Quietly keeping the burden of translation on the vulnerable. AI doesn’t just help us tolerate unequal conditions, it reinforces them, by converting structural punishment into a private self-management practice and letting institutions keep extracting “good tone” without ever paying for the harm that makes tone necessary. If a tool reduces pressure on the people with power to change, then it is compliance infrastructure.
What we want to keep exploring
Is “cost” some noble human ingredient? Maybe it’s an authentication protocol. In real relationships, we trust what’s costly because cost is hard to counterfeit: time, effort, awkward honesty, reputational risk, repair. Cheap empathy reads like spam because it can be mass-produced with no downside. That reframes the question from “Can AI be empathic?” to “Who’s actually on the hook?” Who can be wrong and pay for it, get called out and stay, cause harm and show up for repair. Without consequences, “containment” collapses into aesthetics, well-phrased comfort with nobody carrying the weight. And, if we train ourselves to crave consequence-free care, we don’t become more supported, we become less capable of contact.
Rule of thumb
Before you trust “care” (from AI or experts), ask: where is the cost, and who is carrying it?
Question for your week
Where are you choosing comfort over contact?