The World of AI and WE. Part three

Chapter V. Measure: the indicator of the threshold

Three forecasts described a process already underway; the fourth chapter showed how this process is experienced from within. Here is neither a forecast nor an experience, but the condition under which forecasts cease to be reversible. The distinction matters: a forecast can be refuted by a future observation; a condition can only be met or not met.

The novel offered for this transition a concrete, bodily marker — hairy hands, the residual biological trace of a split that had already occurred within a single body. For the forecast about society there is no such ready image, and it would be dishonest to pretend there is. But the indicator can be specified more precisely than the image about diverging dialects — and it is verifiable. A caste is when the difference in outcome is explained by the difference in access: to data, to tools, to training. But an absolute test, conducted today and repeated in twenty-five years, will measure different things, because the tools themselves will have changed — what must be compared is not the outcome, but the share. The indicator is not a one-off comparison, but a quantity tracked over time: what part of the variance in outcomes between people is explained by the difference in access to tools, and what part is not explained by it — that is, remains even when access is equalized. This is tracked through long-term cohort studies, where what is measured is not academic result or productivity, but autonomy of goal-setting and the capacity for critical synthesis under conditions of equal access to AI tools — that is, not what a person produces, but whether they can set a task that no one has proposed to them, and assemble an answer that exists in no ready output. If the access-unexplained share grows over time, the difference moves from the external to the internal, and this is no longer a caste but a fixed skill of divergence that access no longer governs. The threshold is crossed not at the moment when a test shows something once, but at the moment when this share ceases to shrink from snapshot to snapshot.

Why the indicator is needed now. The 2018–2021 cohort of births is today aged five to eight. The window of plasticity for prefrontal functions closes by eighteen to twenty-five. This leaves ten to seventeen years before the moment when formatting becomes a fact of structure. But a cohort study capable of capturing a trajectory requires not a single measurement but a series: a baseline, intermediate cross-sections, a final comparison. To have data by 2032–2035, measurement must begin now — not because tomorrow will be too late, but because a longitudinal design requires five to nine years of observation before the first meaningful cross-section. If we start in 2030, the data will arrive in 2038 — when the window for the 2018–2021 cohort has already closed and there will be nothing left to measure except the fact. The indicator is not an academic exercise. It is the only instrument that allows us to learn whether the threshold has been crossed before its crossing becomes irreversible.

Chapter VI. Institution: what can be demanded, not only what can be practiced

If the disease is structural, a personal habit does not cure it — at best it slows it for one person, leaving the structure untouched. Therefore a separate list is needed here — as an independent demand on institutions, without fulfilling which personal practice settles nothing at the scale of society. The list is heterogeneous: the first four demands make the system accountable — they leave its scale unchanged, but compel it to explain and to allow decisions to be revised. The fifth and sixth make the system smaller — not more transparent, but less encompassing. These are different tasks, and both are needed, but they are not interchangeable.

Accountability:

• a legislatively fixed duty of explainability for any systems making publicly significant decisions — credit, insurance, hiring, access to services;
• clearly delineated areas closed to a final decision by an algorithm in principle, and not merely by default — justice, the declaration of war, the raising of children, questions of life and death;
• a fixed right to the review of an automatic decision by a human vested with the real authority to overturn it, not merely to hear a complaint;
• independent audit of algorithms, accessible not only to the regulator but to those whose data and decisions pass through these algorithms. This point requires explanation, because it is precisely this one that will be contested in the language described in forecast 2: reports on "threats" and "risks," whose elimination requires additional computational capacity and data, will be presented as care for safety — and paid for accordingly, even if the objective effect of such care is the expansion of the very system that detects these risks. The political language of risk and care will be used to expand infrastructure, not only to manage it. Independent audit is the only counterweight to this cycle, because it allows one to distinguish a risk that requires expansion from a risk constructed to justify expansion. Without audit, the language of care is indistinguishable from the language of capture — and it is precisely in this indistinguishability that forecast 1 operates: the operation is not felt as an operation.

Reduction of scale:

• the principle of subsidiarity — the transfer of the right to make a decision to the lowest, most local level possible, so that the precision of feedback does not remain the privilege of those who are already closer to the archive;
• the right to unreadability — the legislative fixing of zones, services, and institutions where the presence of a digital trace and of algorithmic assessment is forbidden, and not merely optional.

The last point requires two clarifications. The first: this is not about the right of an individual to exit the trace — such a right, in practice, protects only the one who has the resources to exit. The poor cannot afford to be unreadable, because readability is a credit history, access to services, the possibility of being hired. The right to an individual exit from the trace is a privilege dressed up as freedom. This is about something else: about spaces to which the trace is not admitted by definition, regardless of who is in them. Schools where assessment is not digitized and where the child is obliged to meet cognitive friction without an interface that removes it. Medical decisions that do not pass through scoring. Public spaces that generate no behavioral data. Unreadability here is not a property of the subject but a property of infrastructure: not "you can hide," but "there is nowhere to hide here, because there is nothing here to read." This is not Luddism. It is a structural constraint on the Integral: the circuit can connect a territory, but not any territory. And without unreadable spaces, in which the child passes through the formatting of the brain via the resistance of the environment rather than through its absence, the second step of biological cooling becomes irreversible for an entire generation.

The second clarification: unreadability does not cancel accountability. A space free of algorithmic assessment is not a space free of responsibility. A teacher in a non-digital school still reports — but to colleagues, parents, the professional community, and not to a scoring model. The difference is not in the presence of control but in its nature: human control admits exception, context, an unpredictable decision. Algorithmic control does not.

This is not an alternative to personal practice but its condition: without institutional constraints, the personal habit of formulating a position before the model remains an individual hygiene, useful for one person and powerless against a structure that stratifies millions.

Chapter VII. Addressee: personal practice, and to whom it is actually addressed

What can be done already now at the level of an individual, concretely: to preserve the habit of formulating a position before the model proposes it, and not instead of doing so; to notice the moment when a convenient ready solution removes not only the effort but the very possibility of noticing that there was a question at all.

But forecast 3 introduces an asymmetry that this text is obliged to voice. For an adult reader who formed their thinking in the pre-algorithmic era, personal practice is real self-help: the architecture was there, it weakens without load but does not disappear, and the load sustains it. For the 2018–2021 birth cohort, this text is not addressed to them. It is addressed to their parents, educators, designers of the school environment — to those who decide whether the child will have the experience of cognitive friction before the window of plasticity closes. The child cannot "preserve the habit of formulating a position before the model," because they have neither a position, nor a habit, nor a model yet — there is only the environment that adults build for them. Personal practice for this generation is not their practice, but the practice of those who build their environment. And this is, perhaps, the hardest form of responsibility: to make a decision for one who cannot yet make it themselves, knowing that the window in which this decision matters is biologically bounded.

Here an honest caveat is needed, not a brisk conclusion. The capacity to notice that you did not notice is precisely what is removed first when the reduction has already occurred: the recommendation "notice" is addressed to the one who can still hear it, and no text can guarantee that the reader is exactly this addressee, and not the one for whom anesthesia has already taken place — quietly, without any sense of loss. In the light of forecast 3, this caveat ceases to be rhetorical: for a person whose neural architecture was not formed under conditions of cognitive resistance, the recommendation "formulate the position yourself" is addressed to a substrate that is not there. This is not a contradiction within the argument, but the very fork about which it has been speaking all along: some words work for the one who can still hear; others only describe what is happening to the one who can no longer hear. A text can give both descriptions. No text can give a guarantee that the reader is the first and not the second.

The ending of the novel gives no answer to how the confrontation between the Benefactor and the rebels will end. This is an honest ending, not a flaw: Ashby's law does not conclude by itself; it continues to operate while someone either grows their own variety or consents to its reduction. Measurement of the indicator must begin this year — not because the text prescribes it, but because the longitudinal design does not wait, and the cohort for whom the window is still open will not be open forever. If measurement does not begin now, we will learn the result after the fact — not as a forecast, but as a fact with which nothing more can be done.