Learning Across Difference

A civilization can remember everything it has done and still learn only from itself. That possibility became uncomfortable to me once I understood what long political memory could preserve. If each society could keep its own promises, failures, corrections, and institutional mutations connected through time, then fewer lessons would vanish between generations. But the many-gardens possibility required something harder. One garden would have to recognize a lesson grown in another without mistaking the other garden for its destination.

A word I had noticed earlier kept returning to me: ombudsman. It is a Swedish word that traveled into political languages that otherwise share little. The word survived the journey, but the institution did not remain one thing. Sweden’s Parliamentary Ombudsman began in 1809 as an office chosen by the Riksdag to supervise whether public authorities complied with law and protected citizens against abuses of executive power. New Zealand adopted an Ombudsman in 1962 without adopting Sweden’s constitution. Decades later, South Africa placed a Public Protector inside a constitutional architecture of institutions supporting democracy and gave that office power not only to investigate and report but to take appropriate remedial action.

If I compare those countries by regime label, almost nothing useful follows. If I compare the offices by name, I risk another mistake: the shared name suggests an institutional object more stable than the history actually contains. The useful comparison begins one layer lower. What recurring political problem was each office trying to solve? Which relation made the solution work? What conditions gave that relation force? Where did the mechanism stop? What changed when the institution entered another legal and political ecology?

That sequence is the difference between imitation and learning. The object worth carrying across political difference is rarely the visible institution in one piece. It is a conditional account of how a mechanism performed inside a particular history.

The Problem Came Before the Institution

The problem beneath the ombudsman is elementary. Public administration can injure a person even when no election is near, no crime has occurred, and ordinary litigation is too formal, expensive, slow, or narrow to provide a practical route of correction. A permit is mishandled, a detention condition is abusive, an official refuses information, an agency applies a rule unfairly, or a complaint disappears inside the institution being complained about. The person needs somewhere else to go.

Sweden’s 1809 design answered that problem through parliamentary supervision independent of the executive. The historical office was not born in its current form. The early Parliamentary Ombudsman was much closer to a prosecutor: it inspected judges and officials and could initiate proceedings for dereliction of duty. Over the next century and a half, practice shifted. Minor transgressions increasingly produced admonition rather than prosecution, and by the twentieth century criticism, advice, prevention, and the improvement of administrative practice had become central.

The mutation matters because it separates the mechanism from its first implementation. The durable idea was not prosecution by a parliamentary officer. It was an independent route through which the conduct of public authorities could be examined from outside the ordinary administrative chain, with access to the information required for inquiry and a public account of what the authority had done wrong.

Even that formulation is incomplete unless its limits remain visible. A Swedish Ombudsman cannot simply replace an ordinary appeal or alter a court judgment because a complainant dislikes the result. The office supervises legality and administration within a defined jurisdiction; many complaints are screened out because another route is proper, the matter is too old, the office lacks competence, or the record does not justify further inquiry. Its decisions are generally not legally binding, yet Swedish authorities report that criticism is in practice usually followed and often becomes material for improving internal procedures.

I was tempted to describe that last fact as the secret of the institution: independent scrutiny backed by a political culture willing to heed criticism. But the word secret hides the problem I was trying to solve. A mechanism is not one magic ingredient. Nonbinding criticism can matter only if several other relations support it—access to records, public visibility, professional administration, reputational consequence, and institutions that can revise practice after criticism arrives.

Appointment and tenure are part of that ecology because the office has to criticize institutions more powerful than the complainant who approaches it. If the executive can remove the investigator at convenience, independence is ceremonial. If the office cannot compel or reliably obtain the administrative record, criticism becomes speculation. If findings disappear into a private exchange, the wider administration cannot learn from them. And if no agency, legislature, court, press, professional community, or public has reason to treat a finding as consequential, formal independence can coexist with practical irrelevance.

None of these conditions is sufficient alone. More legal power cannot compensate for an office that people cannot reach; easy access cannot compensate for an office unable to investigate; excellent investigations cannot repair a system in which findings have no audience or response. The mechanism is therefore a chain rather than a box on an organizational chart. A complaint can enter, or the office can initiate a matter on its own; evidence has to become inspectable, an independent judgment has to be stated, and the surrounding political order has to contain some route by which that judgment can alter what follows.

I had paid for this distinction before I had words for it. In Sareth, we transferred an elected assembly, a professional civil service, independent revenue administration, and a protected review tribunal, then mistook the presence of those institutions for the presence of the relations that made them usable. External advisers were still mediating disputes, outside money was stabilizing salaries, and security guarantees were supplying conditions the visible institutions were expected eventually to sustain themselves. Sareth does not prove anything about Sweden; only Sweden’s own history can carry a claim about Sweden. But Sareth is why I no longer hear the phrase institutional model without asking which relationships are doing the work beneath the model.

The Swedish case therefore gives no detachable recipe. It gives something more useful: a problem, an evolving correction mechanism, a set of enabling conditions, and a boundary around what the office does not do.

The Name Traveled; the Institution Mutated

New Zealand makes the next step visible because it imported the name without importing the Swedish constitutional settlement. In 1962 it became the first country outside Scandinavia to appoint an Ombudsman. The initial jurisdiction was narrower than the office New Zealand has now. It focused on central government administration. Then the mechanism expanded as the political system discovered new uses for the same underlying relation.

Education and hospital boards entered the jurisdiction in the 1960s; local government followed under later legislation. When the Official Information Act came into force in 1983, the Ombudsman acquired a major role in reviewing refusals and other decisions about access to government information. Later functions reached protected disclosures and independent inspection of places of detention under New Zealand’s implementation of the Optional Protocol to the Convention against Torture. The office did not merely grow larger. The problem-space around independent administrative scrutiny widened.

That widening changed what the word ombudsman meant in practice. A mechanism that had traveled as an informal and comparatively accessible route for complaints became connected to information rights, detention monitoring, whistleblower guidance, disability monitoring, and systemic administrative investigation. The institution was recognizable enough to preserve lineage and different enough that lineage alone could not explain its present capability.

The New Zealand history also shows that later mutation can change the problem the institution is understood to solve. An office created chiefly as a route for reviewing administrative conduct became, through official-information jurisdiction, part of the machinery by which citizens and journalists could inspect government itself. Through detention inspection it moved from responding to individual grievances toward preventive scrutiny of places where people are especially dependent on public authority. The lineage remained visible, but the mechanism acquired new temporal forms: complaint after harm, disclosure during controversy, and inspection before abuse.

That is precisely the kind of development a static comparative dataset can miss even when every legal amendment is coded correctly. The question is not only whether the office possesses a power in year one and year two. It is whether the added power changes how the institution interacts with other actors—whether information review makes administrative reasons more contestable, whether preventive inspection changes the evidence available before a scandal, or whether added mandates stretch staff and attention until an older complaint function weakens. Institutional mutation has direction, tradeoffs, and sequence.

I recognized another Luminaran mistake in that gap. We often judged borrowing at the moment of adoption: an institution had been accepted, resisted, or imposed, and the comparison was treated as largely complete. The more consequential question was what the borrowed mechanism became after years of use. New Zealand was teaching me to treat mutation after adoption as part of the transferable evidence rather than as local noise around the original model.

This is where comparison often becomes lazy. One country is said to have borrowed an institution from another, and the sentence ends at adoption. But adoption is the least interesting moment. The political learning occurs afterward: which jurisdiction was granted, how the office obtained records, which complaints it could refuse, whether ministers or agencies had to answer, what happened when recommendations were ignored, how later laws added functions, and which surrounding institutions made those functions usable.

Once those questions are kept attached, failure stops being embarrassing noise around a successful model. It becomes part of the model’s meaning. An office can be formally independent and practically starved of staff. It can investigate but lack access to decisive records. It can publish criticism that no institution must answer. It can receive thousands of complaints while having no route to correct the recurring condition producing them. The visible office survives while the correction mechanism weakens.

The Council of Europe’s Venice Principles are revealing for this reason. They do not prescribe one standardized ombudsman. They explicitly recognize different models while emphasizing conditions such as a firm legal foundation, independence, adequate resources, broad access, and protection against interference. The standards are useful not because they identify the one correct design, but because the global spread of the institution made its enabling conditions easier to see.

More Power Is a Different Mechanism

South Africa then prevented me from turning those conditions into a universal blueprint. Its Constitution created the Public Protector as one of several independent institutions supporting constitutional democracy. The office must be impartial and accessible, may investigate improper conduct in state affairs or public administration, may report on that conduct, and may take appropriate remedial action. That final power changes the relation.

In Sweden, criticism ordinarily depends on administrative and political institutions responding to a nonbinding finding. In South Africa, the Constitutional Court held in the 2016 Economic Freedom Fighters case that Public Protector remedial action could not simply be ignored; absent a successful legal challenge, those subject to it had to comply. The institutional family resemblance remains, but the consequence attached to a finding is stronger and more juridically consequential.

My old reflex surfaced here. Because the South African Public Protector can attach stronger consequence to a finding, I first felt the familiar pull to call it an upgraded ombudsman: more power, therefore a more advanced form. That is precisely the ranking habit Earth has been forcing me to abandon. A stronger remedy changes the safeguards the institution itself requires. The greater the consequence of a finding, the more important become questions of mandate, procedural fairness, judicial review, evidentiary discipline, and the boundary between investigation and adjudication. Power that solves one weakness can create another object that must be constrained.

South Africa also clarifies why the same visible safeguard can carry a different political burden. A nonbinding ombudsman can rely heavily on persuasion, professional norm, publicity, and legislative attention. An office whose remedial action has legal consequence can change a person’s or institution’s position more directly. That capability may make correction harder to ignore, but it also raises the stakes of error inside the correcting institution itself. The comparison therefore has to follow not only the office’s power over others, but the routes through which others can challenge the office.

This is where comparative learning becomes genuinely reciprocal. Sweden does not stand as an old model that later countries improve upon, and South Africa does not stand as the mature version toward which every ombudsman should evolve. Each exposes a design question for the others. How much practical consequence can criticism obtain without binding force? When does stronger remedial force become necessary? What review protects against an erroneous or overbroad remedy? The answers depend on the wider architecture of courts, legislatures, professional administration, public trust, and the availability of other corrective routes.

The comparison therefore cannot ask which ombudsman is best. Sweden’s nonbinding criticism operates inside one ecology of legality, professional administration, public reasons, and compliance. New Zealand’s office accumulated additional statutory roles inside another. South Africa constitutionalized a Public Protector whose remedial action can carry legal force subject to review. Each arrangement contains a different answer to the same family of problems: how to give people and institutions a route for challenging maladministration without requiring every wrong to become ordinary litigation or electoral crisis.

At this point the word ombudsman almost disappears from the analysis. What remains are relations: independence from the actor being examined; access to complaints and records; authority to investigate; a public explanation; some route by which findings alter administrative conduct; and a boundary preserving the standing of courts, legislatures, executives, or other authorized institutions. Those relations are the material from which learning can begin.

That disappearance matters to me. On Luminara, prestige attached too easily to the institution and then to the society carrying it; the name became a vehicle for status before the mechanism had been understood. Once the comparison is reduced to relations, borrowing no longer has to mean conceding that another political order is the destination. A society can study the corrective relation without accepting the political identity in which it happened to become visible.

One Institution Is Already Too Many Histories

The ombudsman is unusually convenient because the family name survived. Most political learning is harder. Two institutions may perform similar functions under entirely different names, while two institutions with the same name may possess different authority, complaint rules, evidentiary access, remedial powers, or relationships to courts and legislatures. A scholar can compare several examples carefully. Civilizational learning asks for thousands.

The International Ombudsman Institute alone describes a network of roughly two hundred independent ombudsman institutions in more than one hundred countries. Behind that count lie national offices, regional offices, specialized mandates, parliamentary commissioners, public protectors, mediators, inspectors, and other hybrids. Each has amendments, annual reports, complaint categories, court decisions, budget changes, political conflicts, reforms, and sometimes periods in which the institution remained on paper while its practical independence weakened.

The human-scale problem is no longer finding examples. It is maintaining the comparative unit across them. A researcher has to keep the original problem attached to the mechanism, distinguish formal authority from actual implementation, connect staffing and access rules to the office’s capacity, follow what happened after findings were issued, record failures rather than only celebrated successes, and then notice how later reforms changed the same institution. The comparison is temporal and relational before it is statistical.

Language adds another distortion. The word ombudsman makes Sweden and New Zealand easy to find together. A functionally related institution elsewhere may use a title that translates as mediator, commissioner, defender, protector, inspector, or something with no obvious lexical relation at all. Legal traditions divide similar responsibilities differently. One system places information complaints inside the ombudsman; another gives them to a separate commissioner. One office may inspect detention; another may leave that function to a national preventive mechanism with a different institutional home.

This is the point where machine scale changes the comparison rather than merely accelerating a literature search. A multilingual system can retrieve differently named institutions, align provisions that describe similar powers, track amendments, connect annual reports to later court decisions, and surface candidate functional relatives that a particular researcher might not have known to place side by side. The candidate relation remains a hypothesis. But the field from which hypotheses can be drawn expands dramatically.

Luminara had scholars capable of making comparisons this careful. What we lacked was the ability to keep thousands of such conditional histories simultaneously available across languages, jurisdictions, amendments, failures, and later mutations. Our comparative intelligence repeatedly collapsed toward the institutions famous enough, powerful enough, or recent enough to remain in view. The machine changes that scarcity before it changes the judgment.

The practical query I would give such a system is deliberately narrower than “compare ombudsmen.” I would ask for institutions that let an ordinary person challenge maladministration outside the ordinary chain of command. For each candidate, identify who appoints and can remove the officeholder, which bodies fall within jurisdiction, whether records can be demanded, whether investigations may begin without a complaint, whether findings are public, what consequence attaches to a finding, which institution can review it, and what later record shows about response. Only after those relations are visible would the institutional names become useful.

The machine’s distinctive advantage would be persistence across the comparison. If a statute changed, the earlier authority would remain attached. If a court later narrowed or strengthened a power, the new legal state would not overwrite the old one. If an annual report showed a surge in complaints after jurisdiction expanded, that later change could be placed beside the amendment without being mistaken for proof that the amendment caused the increase. If a new government altered appointment rules or funding, the comparison could ask whether subsequent behavior changed rather than silently treating the office as institutionally identical across time.

At human scale, this kind of work is possible because comparative-law scholars, ombudsman networks, journalists, and public institutions already perform it. The novelty is not the invention of comparison. It is the ability to maintain many conditional comparisons at once: hundreds of jurisdictions, several decades, differently named legal powers, translated records, later amendments, court decisions, budgets, complaint data, and evidence of institutional response. A larger portion of political experience can remain simultaneously available to inquiry instead of being reduced to the handful of famous examples one researcher can continuously hold.

The machine can then preserve the structure that ordinary comparison tends to compress. It can keep one office’s independence rule beside the appointment process that may weaken it; place a formal power to obtain records beside evidence of whether records were actually supplied; connect a recommendation to the later administrative change or refusal; and distinguish an expansion of jurisdiction from an improvement in performance. Most importantly, it can retain failed variants instead of allowing the famous successes to define the family.

That last ability matters to me because Luminara learned badly from success. Once one institutional arrangement acquired prestige, our scholars and reformers became increasingly skilled at explaining why it worked and surprisingly poor at remembering where closely related versions had failed. The success became a type; the failures became local exceptions. A machine-maintained comparative history can reverse that convenience by keeping the exceptions attached to the mechanism that produced them.

Comparison Is a Conditional Sentence

A useful comparative output would therefore sound less like a recommendation than a conditional sentence. Under these appointment rules, this office remained independent enough to investigate the executive. Under these disclosure rules, it could obtain the records needed to make criticism specific. Where its findings were nonbinding, compliance depended on other institutions giving criticism practical consequence. Where remedial action acquired stronger legal force, the design also required a stronger review architecture. When jurisdiction expanded, staffing, expertise, and coordination had to expand with it or the new mandate could become symbolic.

The conditionals can also reveal when the same reform would solve the wrong problem. A legislature facing administrative secrecy may need an information-review mechanism more than another general complaints office. A country whose courts already provide inexpensive and rapid review may gain little from giving an ombudsman quasi-adjudicative force. A decentralized state may need several territorial offices or a coordination rule rather than one national center. Where people fear retaliation, formal complaint access may matter less than confidentiality, representation, or authority to begin inquiries without exposing the original complainant.

This is the difference between a successful-looking institution and a transferable mechanism. The first attracts imitation because its achievements are visible. The second becomes useful only after the comparison identifies the local absence it would address. A society that already possesses strong judicial review but weak administrative learning may need publication and follow-up rather than another binding remedy. Another may need a low-cost complaint route because formal rights exist but practical access to courts does not. The political question begins with the local failure, not with the prestige of the foreign institution.

Notice what such an account refuses to say. It does not say Sweden has the best ombudsman, New Zealand improved the Swedish model, or South Africa made the institution stronger and therefore better. Those sentences compress unlike political judgments into a single direction of progress. The conditional account is more demanding because it leaves the local problem open: which of these relations is missing here, and what would happen if it were introduced into the institutions that already exist?

This is also why the comparison must preserve failure modes. A jurisdiction considering stronger administrative review needs to know not only that an ombudsman can increase correction, but that independence can be undermined through appointment or removal, access can become nominal, resources can fail to match mandate, recommendations can accumulate without response, and a powerful remedial office can itself require reliable routes of legal challenge. Learning from another system means inheriting its problems in advance as questions, not importing its successes as promises.

The distinction changed my understanding of political memory. Memory within one society keeps a lesson from disappearing. Comparison across societies lets one society encounter a lesson it has not yet paid for. The second capability is therefore more than a larger archive. It turns separated political histories into a field in which mechanisms can be studied before local experience has produced the same failure.

Sareth now looked different to me for a third time. We had transferred the visible package first and discovered its dependencies after the failure. The comparative architecture I was now imagining reverses that order. It asks for the enabling relations, the failure modes, and the later mutations before anyone decides that an institution should travel. That would not have guaranteed a successful Sareth settlement; political judgment, local agency, power, and uncertainty would still have remained. But it might have made some of the risks legible before our confidence turned them into someone else’s experiment.

That is the acceleration Luminara lacked. We compared our political systems, certainly, but comparison was expensive, linguistically fragmented, strategically selective, and usually dominated by the institutions prestigious enough to attract attention. By the time one society understood the enabling conditions behind another’s success, the visible form had often already been copied—or rejected because accepting it looked like accepting the other society’s political identity.

Our problem was not that we refused every foreign example. We borrowed constantly. What we lacked was a disciplined way to separate the admired mechanism from the political prestige surrounding it. Veyran institutions traveled because Veyran was powerful; rival institutions were rejected because accepting them appeared to concede strategic or ideological ground. Even scholars who believed themselves neutral often began from the cases already made prominent by power. The comparison was filtered before the analysis started. Sareth was one consequence of that filter: Veyran institutions were easy to notice because Veyran was powerful, while the local relations on which those institutions depended were easier to treat as secondary because prestige had already selected the visible form for us.

A machine does not automatically remove that filter. Its corpus can reproduce fame, language dominance, publication bias, and the recordkeeping advantages of richer institutions. But it can make the filter more inspectable and, where records exist, search beyond the canon a human researcher happens to know. That is enough to alter the historical possibility. Comparative attention need not remain as scarce as it was for us.

What it cannot touch is the reason the filter existed. Borrowing was refused out of pride, feared as concession, and defended as identity, and machine-scale comparison cannot remove any of those. It can change the informational burden beneath them. A reformer can ask for mechanisms that address a defined administrative failure without first choosing a model country. A civil servant can compare how several jurisdictions preserve correction under different legal structures. A legislature can see which safeguards accompanied a stronger investigatory power. A civic group can point to failures of implementation rather than merely invoking another country as moral authority.

Difference Becomes More Than Distance

This is where Earth’s political diversity becomes valuable in a way I did not understand when I arrived. Difference is usually described as distance: one country has moved farther from another in ideology, institutions, rights, administrative practice, or political culture. A comparative memory changes the geometry. Difference can also become stored variation in ways of solving recurring problems.

Stored variation does not make every political form equally worthy. Some institutions violate the thin boundary of human standing and peaceful coexistence established earlier; some powers should be condemned rather than preserved for the sake of diversity. Nor does every mechanism deserve to travel. The value lies in refusing to let a justified moral judgment about one part of a political order erase everything that order has learned about another problem.

A society may reject another state’s restrictions on opposition and still study how one of its agencies preserves technical continuity. It may reject another society’s allocation of executive power and still learn from an administrative complaint mechanism, a fiscal rule, a local coordination practice, or a method of preserving professional knowledge. The learning becomes possible because the borrowed object is smaller than the political identity carrying it.

I once assumed that mature political learning would make societies increasingly alike because successful institutions would diffuse and failed ones would disappear. The ombudsman lineage suggests another possibility. Diffusion can produce variation rather than convergence. A mechanism travels, enters another institutional ecology, acquires new functions, loses others, encounters different constraints, and becomes a new source of comparison. Learning creates branches.

The branching matters because adaptation creates new evidence. New Zealand’s expansion of the ombudsman idea became a political experience Sweden did not have in the same form; South Africa’s remedial architecture created another. A later society need not choose among three finished models. It can learn from the differences the earlier adaptations themselves produced. The family tree keeps generating information precisely because no branch has to become the trunk.

This can make political diversity cumulative rather than merely static. One variation exposes a possibility, another exposes its dependency, a third exposes a failure mode, and a fourth invents a safeguard the others can then examine. The result is not convergence on a perfected institution. It is an expanding comparative record in which different political forms can become more capable while remaining recognizably different.

That possibility changes the one-garden story more deeply than coexistence alone. Peaceful coexistence protects political difference from being destroyed. Comparative learning can make the surviving difference productive. One garden can become more capable because another garden solved a problem differently, while the act of learning creates a third arrangement rather than a copy of either.

This is why I no longer imagine Earth’s political future as several fixed systems tolerating one another indefinitely. If machine intelligence can make enough political experience comparable without ranking the political communities that produced it, then the gardens can continue to diverge while learning from one another’s mutations. Their difference need not be a residue waiting for history to erase.

I cannot know whether Luminara could have preserved more of its political diversity had we learned this way. Our convergence emerged from pressures no comparative instrument could simply have wished away. But I can now see a counterfactual we rarely permitted ourselves to ask: if successful mechanisms had traveled as conditional relations rather than as badges of civilizational maturity, and if failed variants had remained as visible as prestigious successes, political learning might have produced more branches before conflict and imitation narrowed the tree. That possibility makes Earth’s remaining diversity look less like unfinished history and more like political material worth keeping alive.

But a conditional comparison still stops before action. Knowing that a mechanism might travel does not tell a society how to graft it into its own authority, rights, institutions, habits, and risks. Comparison can identify the candidate relation and its dependencies. Someone still has to decide whether the local garden should cultivate it, alter it, or refuse it.

That is where learning becomes design. And design is where the gardener must finally enter the argument.