The Rule Must Be Able to Lose

In January 2023, American transplant programmes acquired an unusual obligation: to repair the past. A new policy required them to review the waiting lists of Black kidney transplant candidates and seek adjustments for those disadvantaged by a race-inclusive estimate of kidney function. Changing the calculation for the next patient would leave earlier patients carrying the cost of the old one.

The distinction is easy to understand and expensive to honour. A corrected formula fits in a software release. Corrected waiting time requires finding people, recovering records and deciding what an institution owes them. The arithmetic can change overnight while its consequences remain lodged in a life.

Reza Negarestani gives this problem a philosophical severity that the language of continuous improvement usually avoids. In his 2014 essay, The Labor of the Inhuman, he argues that a commitment has to be elaborated through its consequences, then updated as those consequences become explicit. To undertake a commitment is to accept work that may change the commitment itself. Fidelity cannot be reduced to repeating the original words.

That is a demanding account of what it means to care for someone. It is a still more demanding account of intelligence.

Negarestani's Intelligence and Spirit builds on Hegel's account of mind as a social achievement. Language makes our judgments available to other people, who can ask what follows from them and whether we are entitled to make them. Thinking becomes answerable. An intelligence capable of representing the world must also be capable of treating its own representation as something to examine and transform.

This is far removed from the familiar picture of intelligence as a private quantity, accumulated inside a skull or increased by adding computational capacity. Negarestani is interested in the conditions under which a system can participate in a shared space of reasons. Accurate responses matter. So does the capacity to recognise an obligation created by a response, or a contradiction between commitments that had previously seemed compatible.

The kidney-function equation provides a particularly unforgiving example.

Estimated glomerular filtration rate, or eGFR, translates laboratory measurements and patient characteristics into an estimate of how well the kidneys filter blood. Older equations used a race variable that produced a higher estimate for patients classified as Black. The resulting number could affect the route into transplant evaluation and the accumulation of waiting time. A calculation presented as a description of the body helped determine what the institution would permit that body to receive.

The point is not that estimation should have been abandoned. Direct measurement is difficult, and useful medicine depends on models. The question is what happens when a model's assumptions acquire the authority of the thing being measured.

In 2021, the National Kidney Foundation and American Society of Nephrology task force recommended immediate use of a refitted creatinine equation without race. It also called for greater use of cystatin C to improve clinical decisions. This was a scientific reconstruction of the estimate, rather than a request that clinicians simply stop taking measurements seriously.

The transplant system then faced a different problem. A prospective change could prevent another delay without addressing a delay already imposed. The policy that took effect in January 2023 required programmes to assess affected candidates and submit evidence for waiting-time modifications. Recognition had to travel back through the institution's memory.

These reforms do not settle every inequity in transplantation. Nor does their existence prove that each patient received an adequate remedy. They establish a more exacting standard for correction: the system has to reconsider what it did under the authority of the judgment it has now revised.

Negarestani did not write a transplant-policy manual. The connection is an application of his argument, and its force lies in the obligations that revision creates. If the commitment is to allocate care fairly, an inherited equation cannot be exempt from scrutiny merely because the institution has become good at applying it. When the equation fails, defending the commitment may require overruling the procedure that once expressed it.

Institutional automation makes this distinction harder to avoid. A perfectly executed rule can distribute an error with extraordinary consistency. Every record may be complete. Every decision may have the required signature. The audit can establish that the institution obeyed itself while leaving unanswered whether it had reason to do so.

Imagine a clinical pathway that excludes a patient from a service. The patient challenges the decision, and the challenge reveals that the pathway relies on an assumption the institution can no longer defend. An appeals process may admit that patient as an exception. A learning institution must ask who else was excluded, which earlier decisions need review and whether the pathway should continue to exist. The exception repairs one encounter. Revision changes what the institution is authorised to do next.

This does not make every objection decisive. Expertise would be useless if a rule disappeared whenever someone disliked its conclusion. A serious correction needs evidence and an account of why the earlier judgment no longer holds. The difference is whether those reasons can defeat the rule, rather than merely earn permission to bypass it.

There is a practical architecture implied here. An institution needs to know which assumptions supported a decision and which version of a model produced it. Otherwise, changing the model cannot reliably identify the decisions that deserve another look. Records become a means of correction, with an obligation attached to their retrieval. Keeping evidence is only the beginning of being answerable for it.

The same standard should temper claims about intelligent systems. Fluent explanations and faster decisions may be useful capabilities. They offer little reassurance if a system cannot expose a mistaken premise to challenge, or if its operator treats yesterday's output as settled simply because it has entered the record. Intelligence requires a route from a reason to a changed judgment. Institutional intelligence requires that change to reach the people affected.

Negarestani's account refuses the comfort of an intelligence that merely confirms our present self-description. A commitment can oblige us to become different from the people who first made it. That possibility is the source of its seriousness.

A health institution committed to fair care must therefore permit its own rules to lose an argument. It must also be prepared to find the patients who lost under those rules. Anything less preserves the institution's innocence at their expense.