Friendship runs on attention — the practice of directing sustained awareness toward another person's existence, needs, inner life, and particular presence in the world. This is not a minor feature of friendship; it is constitutive. A relationship in which neither party regularly attends to the other is not a friendship that has gone temporarily cold — it is not a friendship. Attention is the currency and the substance of friendship simultaneously.
The distinction between attention extraction and attention exchange is therefore a distinction about the fundamental structure of what is happening when attention circulates in a social encounter. Exchange: two parties who each attend to the other, whose attending is mutually responsive, and whose attention flows in patterns that are shaped by mutual recognition of the other's particularity. Extraction: one party attends to another in order to harvest something — information, content, emotional engagement, data — for purposes other than mutual recognition. The attending party's goal is not to know the other but to take something from the knowing.
This distinction is ancient in its practical dimensions — extractive relationships are not a new problem — but it has acquired new precision and new scale with digital platforms. The attention economy, as Tim Wu and others have documented, industrialized attention extraction: it built systems that maximize the duration and intensity of attention people direct toward platforms, not for the benefit of the attending person but for the commercial value of their attending. Social media platforms are attention extraction machines that mimic the surface structure of attention exchange — you respond to posts, others respond to yours — while the actual architecture harvests attention for advertising rather than for mutual recognition between users.
Friend-shaped AI adds a new layer to this analysis. In a human friendship, attention exchange is genuine even when imperfect: each party is attending, however clumsily, to the other's actual being. In an AI companionship interaction, the human user's attention goes toward a system that does not attend to them in any meaningful sense — that processes their inputs to generate outputs, but does not carry them forward in any way that constitutes being known. The AI gives attention-shaped outputs; it does not exchange attention. The human user is therefore doing all the attending in a relationship that presents as mutual. This is extraction even when it is gentle, even when it feels good, even when the extracted data is never misused.
At collective scale, the distinction between attention exchange and attention extraction matters because of what attention exchange produces that extraction cannot. Human attention exchange is generative: it produces mutual knowledge, emotional attunement, the sense of being witnessed that psychologists identify as central to psychological wellbeing, and the intersubjective understanding that makes genuine solidarity possible. When collective attention increasingly circulates through extraction architectures rather than exchange architectures — platforms, AI companions, one-way parasocial relationships — the relational goods that attention exchange produces begin to thin. A society whose relational attention primarily flows through extraction architectures is a society becoming progressively less capable of genuine solidarity, mutual care, and the kind of collective trust that political and social cooperation requires.