The Mind to Keep, the Mind to Choose

Lifespan research splits into two lines: raising the average, and raising the ceiling. Life expectancy in Korea went from 62.3 years in 1970 to 83.7 in 2024, a gain of 21.4 years in 54 years. That gain came from cutting the causes of early death, such as infectious disease, infant mortality and heart disease. More people now get close to the ceiling, but the ceiling itself has not moved. The longest verified life is Jeanne Calment's 122 years, ending in 1997, and in the 29 years since no one has passed it.

To raise the ceiling, aging itself has to be treated as the target. That direction started drawing capital only in the 2010s. Google founded Calico in 2013, and Altos Labs launched in 2022 with three billion dollars. Drugs that clear senescent cells, rapamycin and metformin are also being tried on people.

All of these attempts measure proxies, not lifespan: epigenetic clocks, inflammation markers, muscle strength. Proof that the ceiling has risen only comes when someone passes 122, so the results for people trying these things now will arrive decades later. Wait for certain proof and it will come too late. What's needed is the judgment to adopt, early, the treatments whose proxies show an effect and whose risk is low.

The three keys are not at the same stage.

Key	Where it stands
Epigenetic reprogramming	ER-100, injected into the eye to deliver three factors (OCT4·SOX2·KLF4) that erase a cell's marks of age, was cleared for clinical trials by the US FDA in January 2026, and the first patient was dosed on June 9. It is a phase 1 safety trial in up to 18 patients with glaucoma and optic neuropathy
Hyperbaric oxygen	A 2020 Israeli study gave 35 people aged 64 and over 60 sessions across three months; telomeres lengthened and senescent cells fell. It was small, had no control group, and has not been independently replicated
Metformin	Observational data on diabetics hinted at longer life, but mouse lifespan tests found no clear extension from it alone. In a trial of older adults doing strength training it blunted muscle gain. The human aging trial TAME never properly started for lack of funding

The sense that keys are turning is true for reprogramming. This is the first time partial reprogramming has entered a human body. The thing to watch is whether it proves safe in the optic nerve and then spreads to other organs. Because metformin blunts the benefits of exercise, there is currently little basis for a healthy person to take it against aging.

As a goal, yes. Muscles, joints and organs can be replaced once technology catches up. The brain is different. Whether the person after a replacement is the same as the one before is measured by whether memory, response patterns and self-reference carry over, and the brain is what holds all three. The brain is the ruler for identity, so it has to stay original the longest. If the mind has collapsed by the time ceiling-raising technology arrives, fixing the body leaves no one to carry forward.

As a means, the two can't be separated. In a randomized controlled trial of older adults, the group that did a year of aerobic exercise grew their hippocampus, the seat of memory, while the control group's shrank. Cardiorespiratory fitness and blood pressure are brain-vessel care. In 2024 the Lancet Commission on dementia reported that 14 modifiable risk factors play a part in nearly half of dementia, and physical inactivity is one of them. When strength goes, you fall, and a fall that strikes the head can damage the brain in one blow.

Physical ability is not a goal in itself but a means of protecting the brain. This is also why metformin counts as a weak key: a drug that blunts exercise blunts a means of protecting the brain.

Those three are the frame, and adding one more fills the empty cell.

Cell	What it does	How it fails
Supply: blood vessels	carries oxygen and glucose	high blood pressure, atherosclerosis, high blood sugar
Structure: brain tissue	maintains neurons and synapses, clears waste	atrophy, inflammation, amyloid buildup
Regulation: hormones	keeps the brain's working environment steady	chronic excess cortisol, insulin resistance
Input: senses	brings in the signals to process	loss of hearing and sight

The fourth cell comes from the definition of a system. <Cognitive Axioms 03_1> defined a system as a frame that takes in, transforms and sends out. Cut the intake and the system shrinks even if the transforming part is intact. The Lancet Commission also lists hearing loss and vision loss among modifiable risk factors. That is why hearing aids and cataract surgery count as ways of preserving the mind.

Split this way, exercise and sleep turn out to be tools that work across several cells, not cells of their own. Exercise lowers blood pressure (supply), grows the hippocampus (structure) and raises insulin sensitivity (regulation). Sleep washes out waste (structure) and sets the rhythms of cortisol and growth hormone (regulation). How many cells a tool touches at once is the test for choosing it, and it is why exercise and sleep sit at the top of the list.

Hawking had amyotrophic lateral sclerosis. It picks off the motor neurons that move the muscles and leaves the four cells largely alone. Diagnosed at 21 in 1963 and told he was unlikely to live two more years, he kept researching and writing until he died at 76 in 2018. The table of four cells had input but no output. Hawking shows, with a whole life, that the mind runs intact even when output is nearly closed.

Hawking also swapped his output channel for a machine. After losing his voice to a tracheotomy in 1985 he spoke through a speech synthesizer, and in his last years a sensor read the movement of a cheek muscle to pick words. A machine took over the work of mouth and hands, and no one doubted he was the same man. It is a live case of the judgment that if the mind is intact, replacing parts of the body still carries on the same person.

Then the definition of health changes. We don't call a nearsighted person a patient, because glasses push the difference below the scale. Replace the output channel completely and motor neuron damage drops to the same place. Health stops meaning that the parts are original and starts meaning that the four cells are intact.

The shift has already begun. In 2024 a team at UC Davis implanted electrodes in the speech area of an ALS patient who had lost his voice, and turned the sentences he meant to say into speech with over 97 percent accuracy. The channel through which Hawking picked words one at a time with his cheek has moved to one that reads straight from the brain. The direct cause of death in this disease, respiratory failure, is also output-side movement, so if all output can be mechanized it becomes a disease in which you change the body and keep the mind.

The direction is right, and reading comes first. Reading signals out of the brain and writing signals into it differ enormously in difficulty.

The eyes and ears are high-bandwidth input channels. When glasses put results in your field of view, what the AI produces already reaches the brain fast. For an implant to beat wearables on input, it would have to write in the form of thought, bypassing the senses, and that technology is at the level of making points of light in the visual cortex. The narrow channel is output. Speech and fingers, the ways a person tells an AI what they want, are very slow next to the speed of thought, and reading intention from the brain has already reached humans, as the patient in the last section shows.

1. Wearables: take in through the senses, send out through speech and hands
2. Output implants: take in through the senses, send out straight from the brain
3. Input-output implants: write straight into the brain and read straight out of it

At stage 3, AI walks the path mitochondria walked. Mitochondria were once free-living bacteria that took over energy conversion inside a cell, then handed most of their blueprint to the host's nucleus and became an organelle: an outside agent merged into part of its host.

The first six sections were about keeping the mind inside one person. When an extended mind meets power, the problem moves to society.

In the terms of the intelligence trichotomy in <Cognitive Axioms 02_3>: HI (hard, foundational intelligence) is the store of material a goal can draw on, SI (soft, selective intelligence) is the operation that filters out propositions with no information in them, and PI (performance, responsive intelligence) is the ability to take HI and SI in hand and draw out a contribution toward the goal.

Today the bureaucracy and staff act as a leader's outside HI, so leaders aren't required to hold HI in every field. Lawyers and administrators fill leadership because law is the operating language of the state, and the HI of its procedures is directly needed to exercise power. The price is that a leader's judgment is bound to the quality of the material the staff filter up.

With AI attached to their cognition, leaders can draw material from every field themselves, and power and cognitive reach can line up for the first time. At the same time, HI drops out as a way of telling people apart, because everyone has access to the same store.

The trichotomy has an asymmetry: when SI is low, contaminated material flows into PI, and if that repeats, HI itself corrodes. The material AI hands over is mixed with plausible errors. AI extension lifts a high-SI leader to the level their power demands, and magnifies a low-SI leader's misjudgments on the same scale. The leader's qualification shifts from domain knowledge to SI and PI.

The structural judgment holds, and the place where the filtering happens narrows to the nomination stage. The general election is fought over the middle ground between the two camps, so in theory it favors the balanced candidate. But who reaches the general election is decided by primaries, with low turnout and heavy weight on party members. There the most committed decide the result, and the candidate who gathers them by stirring group feeling gets through. The balanced candidate is out before the general election, which becomes a contest between two nominees each picked by the committed.

The committed decide the general election too. Korea's presidential race has no runoff, so one vote more wins, and in 2022 the gap was 0.73 percentage points. In an even race, the turnout of the committed who always show up settles it. The share of committed voters needed depends on turnout and on how the camps line up, so the 5 percent figure can't be checked, but the structure in which a small share sways the whole contest holds.

The selection machinery doesn't measure SI and PI; it measures the power to mobilize. Even if AI extension multiplies capable people, as long as the filter stays the same they get filtered out. The bottleneck is not the supply of ability but the structure of selection.

The limits come from the collapse of the conditions representative government assumed when it was designed.

Assumed at design	Now
Information is slow and scarce, so voters hand judgment to representatives	Information floods in, and recommendation algorithms spread whatever provokes anger
The press and parties filter information	Filters can be bypassed to mobilize the committed directly, and the cost of mobilizing has collapsed
A generalist representative can grasp the policy areas	The areas have specialized, and real judgment has moved to bureaucrats and courts
Election cycles match the time scale of problems	Problems that play out over decades go unrewarded in four- or five-year cycles

The core is that two directions have pulled apart. Selection has tilted toward measuring mobilization, while the job has moved toward complex judgments over long time scales. Korea's total fertility rate falling to 0.72 in 2023 is what builds up when a problem can't be handled on an electoral cycle.

There are experiments in choosing deciders by something other than mobilization. In 2016 Ireland formed a Citizens' Assembly of 99 randomly selected citizens, and their deliberation, after hearing from experts, led to the 2018 constitutional referendum. Sortition is a device that leaves no room for the power to mobilize. Whether it could replace the administrative state as a whole has not been tested.

Open the box and even the functions representative government holds up today begin to shake. As Popper pointed out, it is weak as a device for choosing the most capable and strong as a device for removing those who hold office without bloodshed. Its legitimacy comes from consent, not competence. It pays the cost of sometimes electing incompetent leaders, and in return hands every voter the means of removal.

To switch the selection standard to competence, someone has to set and apply the scale for SI and PI, and who checks those measurers becomes the problem at once. Regimes that selected by competence have historically slid toward the measurers seizing power. The moment a competence standard comes in, consent as the source of legitimacy and the means of removal both begin to shake.

AI extension makes this worse. A new question arises: can voters who aren't extended evaluate the judgment of a leader who is? No one has yet offered a design that brings competence into selection while keeping removal grounded in consent.

Suspending is not declining to compute; it is a computation one level up that decides where computation goes. A person's cognition is limited, and trying to compute everything leaves both their own concerns and society as a whole handled coarsely. The judgment to set down the problems of society and gather one's computation on one's own life is the higher-order computation. The free mandate Burke set out in his 1774 Bristol speech is the device that fills that gap. The representative is not bound by the voters' instructions but computes by their own judgment, taking on the share the voters suspended.

Where this parts from Plato is not whether the work is divided but who holds the removal switch.

	Computation	Selection	Removal
Plato	the wise carry it	the guardians raise and choose them	none
Free mandate	representatives carry it	voters choose them	voters hold it

A suspension is only a suspension if it can be taken back. Under the free mandate, voters suspend the process of computation but never let go of the switch that replaces a representative on the results. Just as a patient chooses a doctor by the results of treatment without medical knowledge, using the switch needs only the ability to judge results, not the process. A voter who has gathered their computation on their own concerns can still hold the switch, and the question from the last section about voters who aren't extended shifts from judging a process to judging results. Plato hands the switch to the wise along with the computation. Since it can't be taken back, it isn't suspension but dispossession.

So there is no convergence on Plato. What's needed is to keep the frame of the free mandate and change only the filter at the entrance: change how primaries and nominations measure, so that SI and PI are caught instead of the power to mobilize. Move the switch as well and suspension turns into dispossession, and the box opens. The weakness that remains is problems like population decline whose results appear decades later. A switch that judges by results doesn't fire in time.

Suspension is the sovereign's default state, and direct action is a switch that turns on only in a state of exception. What separates the two is whether the representative bodies have broken down. Article 20(4) of Germany's Basic Law writes the right of resistance into the constitution, and limits it to cases where no other remedy is available. The constitution, too, treats direct action not as the default but as the last switch.

Seen in this frame, the demand that every member take part in politics every day is a demand to turn the exception into the default. For that demand to be legitimate, a finding that the representative bodies have actually broken down has to come first. The language of crisis and revolution skips that finding and makes people feel as though the breakdown has already happened. Drawing out the computing resources of people who had suspended, and turning them into the committed, is what the mobilization in section 08 really is.

The way a balanced sovereign avoids being swayed by this language is the same as the way they use the switch: they judge whether there is a breakdown from results, not from the word crisis. Keeping the suspension is part of the higher-order computation too.

Both are suspension and delegation, but they differ in what is handed over and how it's taken back.

	Technical mandate (a car)	Political mandate (a representative)
What's handed over	computation with set answers, where physics and safety standards sort right from wrong	judgment mixed with values, where what comes first is decided outside the calculation
Verification	other experts reproduce and check it against the same standards	there is no outside standard to check the result against
Seeing results	the individual meets them at once: the car runs or it stops	they show up late, collectively, with tangled causes
Switch	each holds their own and can buy another car at any time	everyone holds one switch together and presses it only at elections
Binding	only the buyer rides in the car	those who voted against are bound by the result too

In Hirschman's terms, the switch of a technical mandate is exit and the switch of a political mandate is voice. If you don't like the car, you leave for another on your own. If you don't like the government, you can't leave on your own; you have to change it together with a majority.

This difference separates the legitimacy of the two mandates. A technical mandate hands over answers that can be checked, so confirming competence is enough; that's why licenses and qualifying exams work as entrance filters. A political mandate hands over judgments that bind everyone, so competence alone can't ground it, and it needs the consent of those who are bound. The measurer problem from section 10 comes from here: value judgments have no answer key you can hold up like an exam.

Some posts sit between the two. A constitutional justice takes on both a checkable competence, legal expertise, and value judgments that bind everyone. So the post mixes the technical mandate's filter, formal qualifications, with the political mandate's consent, appointment by elected power.

What the president and the Assembly hold is the power of appointment; removal is left to the term.

• Entrance filter: at least 15 years in the legal profession, and at least 40 years old
• Selection: the president appoints all nine; the Assembly elects three, the Chief Justice of the Supreme Court nominates three, and the president picks three directly
• Removal: under Article 112 of the Constitution, a justice can't be removed except by impeachment or a sentence of imprisonment. Mid-term, not even the appointing side can replace them; replacement happens when the six-year term ends

Voter consent is passed along one step removed, and the removal switch is set to time rather than held by a person. The design keeps mobilized politics from swapping justices out at will. The four-year term of the Governor of the Bank of Korea follows the same design. Society already sets apart, like this, the domains where mobilization would be dangerous.

This structure is a working case of what section 11 described: keep the frame and change the entrance filter. It has two limits. The entrance filter measures legal HI, not SI or PI. And since the appointers were chosen by mobilization, the filter problem from the earlier stage carries straight over. That is why appointing justices becomes a partisan fight.

The rule of law sets its rules before any act and measures by the same rules whoever the actor. The result can be known in advance, and the same act gets the same verdict. In the frame of <Cognitive Axioms 04_3>, law is a scale fixed before measurement, and equality before the law is putting the same scale to everyone.

Trial by public opinion looks at the case and the person first, and the mood of the moment decides the verdict. Because the standard is set after the act, it works like retroactive law and predictability collapses; because who you are outweighs what you did, equality collapses.

Insulating justices with fixed terms was a design to keep mobilization's hands off this scale. Trial by opinion is mobilization's pressure breaking through that insulation into the domain of judgment. The force that captured the selection filter has spread into the courts.

The guardian debate took on this problem head-on in 1931. In The Guardian of the Constitution, Schmitt argued that the guardian should be the president, elected directly by the people, because a parliament split into parties and a bench of judges could not represent the will of the people as a whole; his instrument was the emergency powers of Article 48 of the Weimar Constitution. Kelsen answered the same year that those who hold power cannot guard the limits of their own power, and named a constitutional court, set apart from politics, as guardian. Kelsen had designed the Austrian Constitutional Court in 1920.

History delivered a verdict. Schmitt's road ran through rule by emergency decree to Hitler's seizure of power in 1933. Postwar Germany took Kelsen's road with the Federal Constitutional Court in 1951, and Korea's Constitutional Court (1988) follows the Kelsen model too.

The debate is reopening because Kelsen's answer insulated only the exit. Guaranteed terms set justices apart from politics, but the entrance, their appointment, is still held by elected power. Once the electoral filter measures the power to mobilize, Kelsen's court takes in Schmittian force — the will of a mobilized majority — through the entrance. The question of who guards the guardians runs into mobilized politics and stops there. The remaining fix is to leave the removal switch and the term insulation as they are and repair the selection filter at the entrance.

The three problems are three faces of one mechanism. What the guardian debate exposed is that somewhere there is always a seat that makes the final call. Schmitt put that seat with a president elected directly by the people; Kelsen put it with a constitutional court. Either way, whoever sits there comes in through an entrance, and the force that holds that entrance now is mobilization. Whoever takes the final seat by mobilization can reset the scale of law that should stay fixed. The structural mechanism that threatens the rule of law is this path: mobilization passing through the entrance to the final seat and resetting the scale of law.

The path has actually been observed. In Hungary, the governing party that won two-thirds of the seats in the 2010 election wrote a new Fundamental Law the next year and changed the composition and reach of the Constitutional Court. In Poland, after the 2015 change of government, a crisis over the appointment of judges tilted the Constitutional Tribunal toward the ruling party. In 2023 the Israeli government passed a law abolishing one of the Supreme Court's standards of review, and the Court struck it down in January 2024. In all three, a majority won at the polls was used to change the entrance to the final seat.

The road of measuring competence to choose who holds the final seat is blocked by the measurer problem of section 10. The direction has to be reversed: instead of measuring who is most capable, put conditions at the entrance that mobilization alone can't meet. Mobilization is a fast force that leans to one side, so the conditions to set are time and consent that crosses camps. These conditions measure no one's competence, so they need no measurer, and they leave consent and the removal switch untouched, so they don't slide into Plato.

Setting these conditions at each final seat looks like this.

Seat	Condition	Precedent
The scale of law (amending the constitution)	on top of a supermajority, require two votes with a general election between them, so that one wave of mobilization can't change it	Sweden's constitutional amendments (two identical decisions with an election between); Korea requires two-thirds of the Assembly and a referendum
The guardian (electing justices)	elect by a supermajority that crosses camps, with a fallback route for deadlock	Germany's two-thirds election of Federal Constitutional Court judges; the 2024 Basic Law amendment wrote the court's structure and a fallback election into the constitution
The decision on the exception	give emergency powers an automatic expiry, and require a supermajority to declare and extend them	the two-thirds Bundestag requirement for declaring a state of defence in Germany (1968)
Selecting representatives	cut the edge of the committed with runoffs or preferential voting, and have a randomly selected citizens' panel deliberate on candidates and publish its evaluation	Australia's preferential voting; Alaska's ranked-choice voting (2022); Oregon's Citizens' Initiative Review (2010–)

The price is slowness. An entrance that demands time and cross-camp consent also delays needed reform and breeds deadlock. Slowness is exactly where Schmitt's argument, that emergencies need a seat that decides, cuts in. So emergency powers are granted, but with a time limit, and each extension has to win cross-camp consent again. Slowness is the cost paid for blocking mobilization.

Conditions at the entrance block mobilization, but they don't choose good officeholders. That remains the sovereign's job. Those who hold office in the organs of government, with power delegated to them, will be replaced. The cognition that tells a political operator's skill at mobilizing apart from the ability to actually exercise delegated power has to be reinforced and strengthened, in one form or another, on the side of the sovereign. Public records of judgments scored afterward, deliberation by randomly selected citizens, and issue briefs organized by AI are candidates for that reinforcement.

Measuring by results has two holes: attribution and manipulation.

Attribution is the problem of telling whose judgment a result came from. Korea's economy, with its large trade share, moves with the world economy and the trade environment, so there are long stretches in which growth and employment converge in the same direction whoever is in power. In those stretches, sovereigns read a boom from outside as the government's credit and a shock from outside as its fault. In Democracy for Realists (2016), Achen and Bartels argued that voters punish incumbents even for events no government controls, such as droughts and floods. A switch that measures by results misreads luck as judgment.

Manipulation is the problem that, the moment results become the measure, the values reporting them become a target. It is Goodhart's law: when a measure becomes a target, it ceases to be a good measure. Bunching spending just before an election to lift the numbers, changing the definitions and baselines of statistics, putting forward only the good figures — all of these belong here. US nonfarm payrolls come out through layer upon layer of seasonal adjustment, a model that estimates business births and deaths, and annual benchmark revisions. The preliminary benchmark revision of August 2024 cut the March 2024 employment level by 818,000. On August 1, 2025, right after a release that revised May and June down by a combined 258,000 jobs, the president fired the commissioner of the Bureau of Labor Statistics. Every layer leaves room for interpretation, and that room comes under political pressure. In 2009 it came out that Greece's fiscal statistics had been falsified. Scoring the accuracy of judgments comes under the same pressure: forecast only easy questions, or speak too vaguely to be scored, and the score is safe.

There are devices that shrink the two holes. For attribution, there is relative evaluation against countries in similar conditions: see where Korea stood among export economies that went through the same world cycle, and part of the outside factor can be stripped away. For manipulation, there is the independence of statistical agencies; after the Greek affair the European Union tightened the independence requirements for national statistical offices. Both devices shrink the holes, and neither closes them.

So scoring results can serve only as an aid to the sovereign's cognition. Time and cross-camp consent, the frame of the vision, measure no results and only set conditions at the entrance, so they don't lean on these two holes. The sovereign's task of telling good officeholders apart, on the other hand, carries both holes with it. That is why, even while the entrance conditions hold mobilization back, the sovereign's cognition in reading results has to be reinforced.