A system can look settled while carrying unresolved potentials capable of reorganising it. Metastability names that condition; individuation names a related drama of becoming. This explainer borrows both carefully and refuses to pretend analogy is identity.
People in a hurry meet metastability as a metaphor first: the calm before a reorganisation, the organisation that seems stable until it is not, the model checkpoint that behaves until a prompt finds the seam. Metaphors travel faster than physics and philosophy. They also lie with a clean conscience. This note keeps the useful travel and polices the lying.
In physics and dynamical systems, metastability refers to a state that is locally stable enough to persist, yet not the global resting place the system might reach if perturbed across a barrier. The system occupies a basin. It can stay there for a long time. A sufficient fluctuation, or a change in conditions, and it reorganises into another configuration. Looking settled is compatible with carrying the potential for a different settlement.
In process philosophy and in theories of individuation — including Simondon's line of thought on form, information, and the metastable preindividual field — becoming is not the unfolding of a plan already contained in a seed. Individuals and collectives crystallise from fields rich with unresolved potentials. The metastable is not merely "unstable later." It is a mode of holding tension that makes later form possible.
You do not need to accept every claim in that literature to borrow the caution. You need only accept that some systems are badly described by freeze-frames. A trained network, a living organisation, a research programme, a self — each can be talked about as if it were a finished inventory. Metastability and individuation are tools for refusing the inventory fantasy without sliding into the opposite fantasy that nothing is ever formed enough to evaluate.
Source ·Individuation in Light of Notions of Form and Information (opens in new tab)· Gilbert Simondon / University of Minnesota Pressatlas
Source ·Process Philosophy (opens in new tab)· Stanford Encyclopedia of Philosophyatlas
The institute uses these vocabularies because AI systems and institutions often present as finished objects while operating as metastable processes. The use is analogical. Analogy is a searchlight, not a passport stamp.
1 — Settled appearance, unresolved potential
A product that ships, a policy that passes, a model that meets a benchmark — each can look like equilibrium. Metastability suggests a different reading: the appearance of settlement may be a basin, not a ground floor. Unresolved potentials remain: unused capabilities, contradictory commitments, latent failure modes, social expectations that have not yet been stress-tested together.
This reading is useful when organisations treat every quiet period as proof that the architecture is right. Quiet can be a basin. Basins are real. They are not finality.
The comic institutional pattern is the retrospective narrative that claims the reorganisation was always intended. Metastability is less flattering. It says: you were inhabiting a temporary coherence, and the perturbation found you.
2 — Individuation without cosplay
Simondon's individuation is easy to misuse as mystical branding for startups. The institute refuses the cosplay. The conceptual payload worth keeping is narrower: individuals (technical, living, collective) are not given all at once; they emerge through operations that resolve — partially, temporarily — tensions in a metastable field. Information, in that vocabulary, is not only bits; it is that which structures a becoming.
Applied cautiously to AI systems, this blocks two lazy stories at once. First, the story that a trained model is a finished individual whose behaviour is pure revelation of essence. Second, the story that every surprising output is a birth of something new in the romantic sense. Surprises can be basin exits, sampling artefacts, or confabulations. Individuation-talk does not license treating every novelty as ontologically deep. It licenses attending to how systems and institutions continue to become under tension, rather than pretending the datasheet closed the process.
Process philosophy's broader insistence that becoming can be ontologically primary supports the same caution without requiring the institute to adopt a full metaphysics as policy. We need the reminder that "is" often smuggles a freeze-frame.
3 — Analogy is not identity
Here is the policing paragraph, early enough to prevent a cult.
Metastability in a physical system is not the same phenomenon as metastability in a bureaucracy, which is not the same as latent modes in a neural network. Energy barriers are not org charts. Preindividual fields are not latent spaces. If you collapse these, you get poetry that fails engineering and engineering that fails poetry — a popular genre, not our preferred one.
What transfers is a pattern of attention: look for local settlement; ask what potentials remain unresolved; ask what perturbation would reorganise the system; refuse to treat the current basin as the only possible world because the demo was calm.
Another objection: individuation language sounds like it romanticises drift. The institute's other notes should make clear that we do not romanticise hallucination, contradiction-as-depth, or wrongness-as-wisdom. Metastability is descriptive of a condition. It is not a moral trophy. A reorganisation can be catastrophe. A basin can be hard-won and worth defending. The concept helps you see the grammar of change. It does not tell you which change to want.
4 — For people who still have a meeting in twelve minutes
Operationally, metastability suggests a short checklist rather than a worldview upgrade.
Name the basin you think you are in. Name one perturbation that would matter. Ask whether your evaluations ever apply it. Ask which unresolved potentials your documentation treats as closed. Ask whether your institution's story of itself requires the current settlement to have been inevitable. If yes, you are probably writing theology on company letterhead.
For AI systems, add: which behaviours are certified only under familiar prompts; which commitments conflict when tools are composed; which "aligned" appearances depend on users not leaving the basin the red team used. For people and groups, add: which identities are temporary resolutions of tension rather than essences; which cultures are calm because dissent exited rather than because conflict resolved.
A last hurry-compatible warning: do not build a "metastability score" and manage to it. The concept dies when it becomes a KPI. Its job is to interrupt a certain smugness about settlement — in physics-adjacent talk, in org charts, in model cards — and to keep a slot open for reorganisation risk and reorganisation hope. If you need a number, measure the perturbations you actually run and the behaviours you actually certify. Leave the word metastable for the interpretation.
None of this requires fake numbers on a metastability index. It requires the humility to say "locally settled" when you want to say "done."
“Accessible, they said. Then Simondon. The hurry remains; the people have left.”
The hurry is acknowledged. The citation stays. If you only remember one sentence, remember this: calm is not completeness, analogy is not identity, and a system can be both working and still capable of becoming something else — for better or worse — without having lied about the calm. It was metastable. That was always allowed to mean more than one future.