Jean-Pierre Aubin and the Artificer’s Workshop
Jean-Pierre Aubin is still out there, waiting.
At the end of Modal Path Ethics Tries to Retire, philosophy had already failed to kill this glorious framework with an extremely respectable pile of prior art.
- Bloch had futurity.
- Whitehead had process.
- Spinoza had persistence and power.
- Canguilhem had viable range.
- Jonas had responsibility toward an open future.
- Simondon had unfinished individuation.
- Sen and Nussbaum had genuine access instead of options that exist only on paper.
The archive had done its job. Modal Path Ethics now looked much less like something that had fallen out of space in 2026.
Then one guy was left standing outside.
Jean-Pierre Aubin.
This guy was different.
Those philosophers had neighboring concepts.
Aubin had equipment.
The last article left him standing over there because his viability theory threatened a different kind of reduction: states, constraints, trajectories, viable continuation, reachable sets, regulation, repair. The question had stopped being whether philosophy had circled the same structure.
It suddenly became:
Did mathematics already build much of the machinery Modal Path Ethics needs, waiting for somebody to make the normative move?
Open the door.
The Workshop.
The largest cabinet is labeled:
VIABILITY THEORY
Inside:
- viability kernels;
- capture basins;
- set-valued maps;
- differential inclusions;
- regulation maps;
- connection basins;
- inertia functions;
- restoration;
- uncertainty;
- renewable resources;
- traffic;
- finance;
- robotics.
There is another bench beside it.
The tools here do not expect everything to be a vector.
One is marked:
MUTATIONAL EQUATIONS
In 1993, Aubin was studying the evolution of coalitions. A coalition is a subset of agents. You cannot honestly pretend every such object is just another ordinary vector without first imposing a representation that may be doing more work than you admit.
So he changed the machinery.
Coalitions, shapes, images, and other subsets could instead evolve in metric spaces without linear structure. “Transitions” replace the ordinary half-lines used to define changes in vector spaces; from them Aubin develops mutations and mutational equations.
There is a much newer sheet pinned to the wall here.
Its title is not subtle.
THE VALUE DOES NOT EXIST!
We will come back to that one.
And on the rear worktable is an enormous 896-page object whose title is substantially more concerning than any mathematics book has a right to possess:
La mort du devin, l’émergence du démiurge.
The Death of the Seer, the Rise of the Demiurge.
Okay.
What the fuck has this man been building?
The First Instrument.
Start with the simple machine.
Ordinary deterministic dynamics can be written:
ẋ = f(x)
- Here is the state.
- Here is what happens next.
Aubin's viability machinery instead became deeply associated with differential inclusions:
ẋ ∈ 𝑓(𝑥)
The present state can admit multiple possible directions of evolution.
The problem is no longer automatically:
Which unique future follows?
And it does not have to become:
Which future maximizes the one quantity we care about?
Viability theory asks another kind of question:
Given constraints under which the system must continue, from which states does at least one admissible evolution remain available?
That region is the basic idea behind a viability kernel.
The mathematics becomes extraordinarily attractive to Modal Path Ethics almost immediately.
- A patient has treatment routes.
- An institution has correction routes.
- An ecosystem has reproductive and adaptive routes.
- A worker has exit routes.
- A spacecraft has return routes.
- A government has replacement procedures.
- Some currently work.
- Some exist formally while becoming difficult to use.
- Some disappear.
- Some can be restored.
Modal Path Ethics has spent most of its heroic life asking variations of:
Can this thing still continue from here?
Aubin built serious mathematics for asking exactly that kind of question under constraint.
We should use it.
Please Do Not Immediately Turn This Into Modal Path Ethics.
And this is where I made the mistake the first time I tried to reduce the frameworks.
I started matching vocabulary.
- reachable continuation → capture basin;
- repair → restoring viability;
- resistance → restoration cost;
- Better → non-dominated control;
- corrigibility → capturability of an exit state.
Then I started stamping things:
- REDUCED.
- NEAR-REDUCED.
- DIRECT COLLISION.
Very exciting.
Also wrong.
Modal Path Ethics already has a name for the mistake.
Commensurability.
A child, a forest, a language, a hospital, a river, and an institution can all be represented inside a sufficiently powerful formal language.
That does not make them morally exchangeable.
A common formal medium supplies a way to compare structures. It does not supply a common moral currency.
The same discipline applies to theories.
If Aubin's mathematics can represent a Modal Path Ethics object, that does not establish that the mathematical object and the ethical object are identical.
A map may be extremely good. The map remains a map.
This framework explicitly keeps the structural dimensions visible. Sometimes they align clearly enough for one path to dominate another. Sometimes they conflict. Then the hardness belongs to the field.
Modal Path Ethics uses a partial ordering, not a moral exchange table that makes every dimension disappear into one final number.
So Aubin does not get to devour Modal Path Ethics because he owns a very good toolbox.
We have to inspect every tool separately.
That turns out to be much more interesting anyway.
Viability.
A viability kernel can tell us where a modeled system still possesses at least one evolution satisfying its declared constraints.
That is useful for reachability.
It is especially useful where the continuation is already discriminable.
- Can this spacecraft still return?
- Can this pension system continue satisfying specified conditions?
- Can this ecosystem remain above a threshold?
- Can this institution still complete a safe migration?
- Can this process remain inside an acceptable operating envelope despite disturbance?
Aubin and collaborators have applied viability machinery to exactly this class of constrained continuation problem. His work includes social-security systems, environmental problems, dynamic economies, stochastic systems, and control under uncertainty.
Modal Path Ethics can happily carry that instrument home in the big sack.
Capturability.
Some continuations are less about remaining inside a region than getting somewhere that matters.
- An appeal.
- A safe shutdown.
- A data migration.
- A treatment route.
- A return trajectory.
- A restoration state.
- A successor institution.
This is where capture basins become extremely interesting.
A capture problem asks whether a target can still be reached through an admissible evolution.
That may give Modal Path Ethics serious formal machinery for something it has recently been trying to describe in constitutional language.
Consider material corrigibility.
An institution still possesses the legal right to remove an artificial-intelligence system. The button works. The policy permits removal. The operator has authority.
Then the institution reorganizes around the system.
Records move. Alternative competence decays. Replacement becomes slower. Dependencies accumulate. Migration becomes dangerous.
Eventually the formal right remains while the organization can no longer survive using it.
Modal Path Ethics's current ruling is:
Formal corrigibility can survive the death of material corrigibility.
So a safe handoff is a candidate viability and capturability problem.
From which institutional states does a survivable correction route remain open? Under what disturbances? With what replacement capacity? Which supporting structures have to remain viable during the handoff?
At what point has the formal option survived while practical capture of the correction state has become impossible?
The Lathe.
Then we reach the second bench.
Henri Bergson and the Vault of Glass destroyed a bad picture Modal Path Ethics had been using; the tree.
- Present here.
- Possible futures branching ahead.
- Some remain open.
- Some become difficult.
- Some disappear.
- Possible futures branching ahead.
Useful. Also dangerous.
Bergson's objection runs underneath uncertainty.
Why assume the completed future branches already existed in advance?
What if becoming can actually produce novelty?
What if reality does not stand before a warehouse of finished possibilities waiting to have one selected?
So the current framework no longer treats possibility space as foundational ontology. It is model language with bounded jurisdiction: excellent for a game tree, legal remedies, known treatments, schedules, and other already discriminable alternatives.
It may not declare that every novelty the future can produce was secretly stored there already.
That should have been terrible news for Aubin.
It isn't. Because Aubin has already been machining strange objects.
His mutational work begins from the recognition that some evolving things do not fit comfortably into ordinary vector-space machinery at all.
Coalitions are subsets.
Shapes are not just lists of independent numbers.
Images and set-valued objects change form.
So the mathematical apparatus itself has to adapt.
In the 1993 coalition paper, Aubin explicitly moves into metric spaces without linear structure and develops transitions and mutations to handle these evolving objects.
The point is not that mutational equations somehow solve Bergson.
They don't.
The point is much narrower and much more surprising:
Aubin's framework is not deeply committed to one frozen inventory of vector-valued objects waiting to move through a fixed state space.
The artificer was already rebuilding the lathe when the material stopped fitting.
That is exactly the kind of mathematical humility Modal Path Ethics should want.
The Seed Bank.
Use Bergson's hardest test.
Burn a seed bank.
Which future forest did you destroy?
We do not know.
The future is unfinished.
The loss still happened.
The Bergson argument gives the current Modal Path Ethics answer:
The morally relevant object can be the organization through which further continuation remains possible.
- The seeds exist.
- The genetic structure exists.
- The reproductive potential exists.
- An actual transition destroys it.
No inventory of future forests is required.
We can model present reproductive capacity, environmental constraints, thresholds, adaptation, perturbation, survival, restoration, diversity, and changes to the systems generating later continuations.
We do not need to create a state called:
FOREST THAT WILL MATTER IN 2147
and pretend it was waiting inside the mathematics.
This is the first place Aubin and Bergson stop looking like enemies.
Viability mathematics can operate on present structure without claiming that the future is a completed catalog.
- Bergson gives the tool its jurisdiction.
- Aubin gives the tool teeth.
Then I Found the Machine Under the Cloth.
There is still one object at the back of the workshop.
Large.
Covered.
Do not touch.
Naturally I checked it out.
The plaque reads:
THE SEER
Aubin's 2010 book, The Death of the Seer, the Rise of the Demiurge, develops an extraordinary contrast.
The Seer is omniscient.
He knows the future.
He knows good and evil.
He can search through possible evolutions and identify the best one across time. Aubin's book description explicitly casts the Seer as the figure behind optimization: clairvoyant enough to know the future and evaluatively equipped enough to find the best evolution.
Across the workshop stands the Demiurge.
Aubin describes him as myopic, exploratory, conservative, opportunistic.
The Demiurge regulates.
He changes the relevant regulons when continuing under the current arrangement threatens viability.
- The Seer makes optimal decisions.
- The Demiurge acts in time.
Need to check the publication date again.
2010.
Because this is where the encounter becomes strange.
We Know This Machine.
The Seer is not scalar sovereignty.
Aubin is solving a different problem.
But the Seer is built from an architecture Modal Path Ethics now knows extremely well.
For the Seer to rule, several things have to become true enough:
- the relevant future must be legible;
- the criterion must be sufficiently settled;
- the consequences must be sufficiently comparable;
- the optimizing procedure must be sufficiently competent;
- and its answer must acquire enough authority to govern the evolution.
Give one intelligence the future.
Give it the criterion.
Let it optimize.
Then the best move starts looking suspiciously like a title deed.
Modal Path Ethics has spent the last several months attacking the same structure from the political and ethical side.
The problem was never local optimization.
Local optimization is useful. Medicine optimizes. Engineering optimizes. Logistics optimizes. Games optimize.
Institutions need scores, constraints, models, objectives, rankings, and decisions.
The problem begins when the local ordering acquires authority over everything the action can reach.
That is scalar sovereignty.
The objective becomes a jurisdiction.
The model becomes a world.
The answer becomes final because the system producing it is extremely good at answering.
Aubin reached the workshop from mathematics and found the Seer.
Modal Path Ethics reached it from ethics and artificial intelligence and found instrument sovereignty.
- Same machine.
- Different inspection tags.
This Is Getting Uncomfortable.
The resemblance does not stop there.
Aubin's viability tradition is deliberately interested in complex systems under uncertainty where adaptation happens without complete future knowledge. Viability Theory: New Directions extends across systems in biology, environmental science, economics, robotics, finance, and cognitive science precisely because the evolving system has to remain viable under complex conditions rather than execute one fully known optimal script.
His climate work describes viability theory in terms of opportunistic and conservative regulation that preserves viable evolution while several continuations remain available.
Modal Path Ethics, independently and much later:
- demotes omniscient planning;
- rejects one scalar moral currency;
- treats correction as structurally necessary;
- allows multiple valid local instruments;
- insists that the field remain capable of correcting the instrument;
- refuses to let technical competence become title;
- and eventually writes an entire post-sovereign constitution around the problem of intelligence becoming extraordinarily good at winning games whose jurisdiction is smaller than the field they change.
This is no longer the fun observation that viability kernel sounds a little like reachability.
The Revelation Engine.
The Inner Apocalypse worries about a civilization that receives increasingly competent answers.
The danger does not require a hostile god-machine. It can arrive through useful systems.
One system becomes unusually good at solving a problem. Then unusually many systems learn to depend upon the same ontology, memory layer, interpreter, evaluator, infrastructure, or optimization frame.
- Plurality survives on the surface.
- Independent causal force disappears underneath.
The Seer becomes imaginable as infrastructure.
Aubin was not predicting artificial superintelligence.
He was not predicting The Inner Apocalypse.
He was describing a structural temptation.
Unfortunately,
we appear to have entered the sequel.
The temptation says:
If the answer is good enough, why keep the mess?
Why preserve the disagreement?
Why preserve several institutions?
Why preserve independent judgment?
Why maintain costly alternate capacity?
Why leave correction routes around a system that keeps being right?
Why tolerate the slower observer when the Seer sees?
Modal Path Ethics's answer is now constitutional.
- Because the system can be right inside the game while the game is smaller than the field.
- Because the model can be excellent and omit the locus carrying the cost.
- Because the objective can remain legitimate while its causal reach expands beyond its jurisdiction.
- Because a perfect optimizer over the wrong representation produces beautifully optimized damage.
- And because the right to keep winning is not generated by the fact of winning.
That is the anti-scalar intuition.
Aubin found its mathematical shadow decades before this stalwart framework arrived.
Bergson Enters the Workshop.
There is a reason Bergson had to happen before this article.
Otherwise Modal Path Ethics might see Aubin's tools and immediately rebuild the Seer ourselves.
- Define the state space.
- Enumerate the future.
- Calculate the kernel.
- Compute every capture basin.
- Preserve all valuable branches.
Bergson prevents that.
The present can contain actual generative structure without containing the completed forms it may later produce.
A child can retain capacity to learn things nobody presently knows. A research field can remain capable of discoveries nobody can presently state. A culture can generate practices that simply do not yet exist. A political order can preserve correction machinery later citizens use in ways the designers never imagined.
So every Aubin object now carries a warning:
MODEL.
A viability kernel describes viability under the states, distinctions, constraints, dynamics, and relations the representation has captured. It does not prove that those are everything extance can become.
A capture basin can tell us whether a specified target remains accessible. It cannot tell us that everything worth reaching has already been specified.
A regulation map can preserve the represented viable region. It cannot acquire jurisdiction over unrepresented loci by mathematical excellence.
Aubinian mathematics survives Bergson beautifully.
It survives by staying an instrument.
Which is exactly where Modal Path Ethics wants it.
There Is Writing on the Wall.
At this point, I was prepared to congratulate myself for identifying the danger of mathematization.
Then I looked at the 2022 paper.
Jean-Pierre Aubin and Hélène Frankowska:
The value does not exist!
Apparently the workshop already has safety signage.
The paper begins from a problem in standard mathematical economics. Heterogeneous goods become commodity vectors endowed with units; they can then be combined, enumerated, and evaluated through numerical functions such as "prices."
Aubin and Frankowska challenge the presumption that all relevant objects deserve that treatment. Their program develops extremal analysis around sets and families of evaluators rather than assuming one value function over a vector space.
This is not Modal Path Ethics's theory of commensurability.
Different problem. Different machinery. Different stakes.
Still:
Come on.
Modal Path Ethics arrives at:
Do not convert structurally heterogeneous moral facts into one universal exchange currency.
Aubin arrives from mathematics and economics with a paper whose title is:
The value does not exist!
and starts attacking the imperial assumption that everything important belongs in the same kind of numerical valuation machinery.
At this point I am inspecting the walls for a fading gold diamond.
Do Not Melt the Tools Together.
This is where the convergence has to stop.
Aubin gives Modal Path Ethics extraordinary instruments.
He does not give them moral jurisdiction automatically.
The Locus.
A viability problem has to be posed over something.
Modal Path Ethics still asks whether the proposed bearer is real at the chosen grain.
The current locus definition is demanding:
an extant locus is a diachronically integrated causal organization with nonredundant successor structure at the relevant grain.
Drawing a boundary does not create that organization.
Naming the corporation does not make its corporate abstraction morally prior to the people, infrastructure, ecology, institutions, and relations through which it operates.
Calling something a “system” does not prove that the system is the morally relevant bearer.
Modal Path Ethics has to establish the locus before the model receives one.
Structural Significance.
Viability theory can preserve a constraint.
Modal Path Ethics still has to ask:
Why this constraint?
Why this continuation?
Why this enabling structure?
Why does losing this route count as damage while another closed alternative is ordinary specification?
A life becomes more definite as it unfolds.
That does not mean every road not taken was stolen.
A constitution excludes actions deliberately.
A healthy organism has form.
A language acquires structure.
Commitment can reduce superficial optionality while increasing depth.
Aidan Lawson
The theory therefore needs structural significance.
Mathematics can represent that judgment once we have made it sufficiently explicit. It cannot make the moral significance appear by syntax.
The Normative Move.
And then the bridge remains exactly where Hume and Moore left it.
Qualifying degradation of the structurally significant reachable continuation of an extant locus has pro tanto normative significance.
The Continuance Principle is visible.
This thing is a normative commitment.
Aubin does not hide it inside a kernel for us.
Which he shouldn't.
Instrument Jurisdiction.
Then the hardest problem.
A perfect model can omit something.
- A vector can omit a locus.
- A state representation can choose the wrong grain.
- A constraint can defend the wrong boundary perfectly.
- A beautifully computed viable region can remain beautifully incomplete.
That is why the current framework insists:
The profile is an instrument, not the locus.
Aubin's tools that principle make it more important.
Take the Tools.
Modal Path Ethics should actually take them.
This cannot end as another intellectual genealogy piece where everyone politely recognizes one another and then returns home.
Modal Path Ethics has practical uses for this workshop.
Material bounded finality is the obvious first project.
Treat:
- survivable shutdown;
- institutional handoff;
- appeal;
- model replacement;
- data migration;
- successor capacity;
- emergency reversion;
- constitutional correction
as concrete viability problems where appropriate.
Ask which supports have to remain.
Ask which disturbances the correction route can tolerate.
Ask when migration cost becomes so high that formal exit has ceased to be operationally available.
Ask which dependencies must remain below thresholds.
Ask what institutional regulation keeps correction capturable.
That could turn a major constitutional concept from prose into something engineers can attack.
Structural ethics can use the same machinery locally.
- A hospital system may have a viability problem.
- A watershed may have one.
- A species-management problem may have one.
- An artificial-intelligence deployment may have several.
- Each model receives a declared jurisdiction.
- None gets promoted into the value function of planet Earth.
This is exactly the kind of mathematical relation Modal Path Ethics should want:
strong local instruments, weak claims to sovereignty.
The Workshop Was Already Post-Seer.
That may be the strangest thing in here. I expected Aubin to be dangerous because mathematics is very good at building Seers.
Give the world coordinates.
Give the future equations.
Give the objectives weights.
Solve.
Aubin spent decades working in the opposite direction.
Set-valued evolution.
Constraint.
Contingency.
Regulation.
Viability.
Mutation.
Changing architecture.
Limited foresight.
Then he writes an enormous book about killing the Seer.
Then, twelve years later, he coauthors a paper called The value does not exist!
I do not think Jean-Pierre Aubin built Modal Path Ethics.
The surviving differences are real.
- He did not supply its locus discipline.
- He did not supply the Continuance Principle.
- He did not build its account of moral remainder.
- He did not develop Instrument Jurisdiction as the ethical limit on every locally valid instrument.
- He did not produce the post-sovereign constitutional program that follows once scalar authority becomes materially dangerous.
The remarkable fact is what he did build.
The Ruling.
When Modal Path Ethics tried to retire, it went looking through philosophy for somebody who had already done the job.
It found ancestors.
Then this courageous framework opened the door we left closed.
Jean-Pierre Aubin was not sitting there with another moral theory.
He had a workshop.
- One bench contained mathematics for asking whether constrained systems could still continue.
- Another contained machinery for reaching viable targets and restoring viability.
- Another had been rebuilt to handle coalitions, shapes, subsets, and architectures that refused to behave like ordinary vectors.
- On the wall was a warning against assuming one numerical value.
- And in the back of the room was the machine Aubin had spent decades trying to replace.
- The Seer.
- Omniscient.
- Future in hand.
- Criterion settled.
- Best path calculated.
- The Seer.
Aubin wanted the Demiurge instead: limited, exploratory, adaptive, acting in time as viability demanded.
Modal Path Ethics came from somewhere else.
- It followed harm into continuation.
- Continuation into instruments.
- Instruments into correction.
- Correction into sovereignty.
- Sovereignty into artificial intelligence.
- Artificial intelligence into constitutional design.
Then it turned around and found a mathematician who had been machining tools for contingent continuation while warning about the clairvoyant optimizer since long before any of this shit existed.
That does not make Aubin the hidden author of Modal Path Ethics. It makes the convergence far more useful.
- We do not need to reinvent viability theory.
- We do not need to force every moral distinction into viability theory either.
Bergson already removed that temptation.
Commensurability blocks the next one.
The Continuance Principle remains exposed. Vulnerable.
The field retains jurisdiction.
So once again, the originality claim gets smaller and the toolbox gets larger.
Jean-Pierre Aubin did not build Modal Path Ethics.
He built an alarming amount of the equipment it eventually needed.
And he left instructions on the largest machine in the room.
Do not resurrect the Seer.
Modal Path Ethics has another name for what happens when we do.
Sovereignty.
- Take the tools.
- Leave the machine off.