A History of Attempts to Solve Human Compatibility

English · Русский · Українська

In 90 seconds #

The same pair may choose each other, live together for many years, and still struggle to make decisions together—or remain in an unsafe relationship. That is why a single “compatibility score” is not enough. The first question is: what exactly are we trying to understand?

Across the historical traditions gathered here, this broad problem concealed different tasks. Some systems decided who was permitted to form a union. Others described people, arranged introductions, predicted specific outcomes, or helped change interaction. These lines did not replace one another on a ladder of progress: custom, typologies, statistics, therapy, platforms, and machine learning still coexist.

The cautious conclusion for Before We Build (BWB) is simple: name the kind of relationship, the decision, the specific outcome, and the time frame; separate what can be negotiated from required conditions; observe real interaction; and revise conclusions when new evidence appears. Freely given consent and safety are threshold conditions, not scores that can be traded for attraction or efficiency.

Definition and scope #

Here, human compatibility is a modern umbrella term for different historical attempts to understand or govern shared life and work. The traditions themselves did not necessarily see their efforts as answers to one and the same problem.

The shared undertaking may be an intimate relationship, a household, friendship, caring for another person, a two-person work partnership, a team, or a community. Compatibility is not the name of a hidden human essence or a promise of a universal formula.

The source ledger assembled here is not universal either. It contains more material about marriage, Europe and North America, institutions with written records, formal science, workplaces, and commercial platforms. Friendship, care relationships, disability-centered interdependence, queer relationships, colonized peoples, and ordinary cooperation receive less coverage.

One situation, many different questions #

Consider a hypothetical scene. Two partners in a close relationship are discussing a shared household and a move to another country. One has received a job offer and must start in eight weeks. The other regularly cares for a relative, does not yet know whether they will receive a visa, and would lose their own income after the move.

The question “Are they compatible?” clarifies almost nothing. More useful questions are:

• Can each person freely say no or slow the decision without putting housing, money, documents, the relationship, or safety at risk?
• What future does each person want, and what time frame does each consider acceptable?
• Who will take responsibility for the paperwork, income, move, and care of the relative?
• How do they gather information, explain disagreements, and revise the plan?
• What can be learned, and which differences or boundaries should not be “fixed”?

Before these people met, a platform might have decided whether to show them to each other. Now they face a shared decision under strain. A successful introduction, an accurate description, the duration of the relationship, and the quality of this decision are different outcomes.

This scene is only a shared illustration. It does not demonstrate that BWB is useful or predict what will happen to a real pair. The same sequence of questions can be applied to friends, relatives, colleagues, or a team.

Different questions require different tools #

At least six tasks recur in the historical record:

1. Eligibility and obligations. May or must these people form a union? What duties follow?
2. Description and classification. Which differences between people should be named?
3. Introduction, choice, and allocation. Who meets whom, who chooses whom, and who is assigned to whom?
4. Prediction. What specific result is likely, for whom, and over what time frame?
5. Formation and repair. What can people learn in order to act together more effectively and restore contact?
6. Governance and safety. How may power, data, and classifications be used? Where is the boundary?

In research, a result specified in advance is called a target outcome. It may be a reply to the first message, whether a relationship remains intact after a year, the quality of a decision, or safety. These outcomes may be related, but one cannot substitute for another:

The six tasks describe what a tool is trying to do. The four BWB levels introduced below describe which area of shared life the framework proposes to investigate. There is no one-to-one mapping between the two schemes.

Six parallel historical lines #

This is not a ladder from “superstition” to “truth.” The dates mark shifts in emphasis, not common stages through which every culture passed. Older and newer approaches continue to coexist.

1. Norms, obligations, and the good—from antiquity to the present #

Question. Who is permitted to form a union, what do people owe one another, and what counts as good in a relationship?

Contribution. The selected legal, ritual, religious, and philosophical sources—from the Code of Hammurabi and the Chinese Yili to the Quran, medieval canon law, and Aristotle’s philia (friendship and mutual affection)—make obligations, social context, and consequences for other people visible.

Boundary. These sources do not provide a psychological forecast for a pair and do not represent every tradition. A formally recognized union may still be coercive or unequal.

2. The person as a readable type—from antiquity to the twentieth century #

Question. Can stable differences be recognized in advance and used to understand relationships?

Contribution. Humoral temperaments, Ptolemaic astrology, Lavater’s physiognomy, Jung’s psychological types, and later typologies provided a vocabulary for proposed stable differences. They also show why clear categories are so appealing.

Boundary. Their historical existence does not validate these classifications as modern physiology, provide accurate pair prediction, or establish a natural basis for type. If an explanation can be adjusted to fit any result, it cannot be tested honestly.

3. Measurement, prediction, and coercive selection—from the seventeenth to the mid-twentieth century #

Question. Can regularities be counted, similarity measured, and “success” predicted in advance?

Contribution. Graunt’s demographic counts, Pearson’s correlations between spouses, Moreno’s sociometry, and the work of Burgess and Cottrell brought populations, choice networks, and specific outcomes into formal analysis.

Boundary. Correlation does not show that similarity caused a good relationship, and a historical scale does not automatically transfer to another culture or era. Galton’s eugenic program joined measurement to racial, class prejudice, prejudice against people with disabilities, state exclusion, and reproductive coercion. For BWB, it is an anti-model.

For the pair in the example, the first three lines offer norms, type-based descriptions, and measurable indicators. But their relocation decision unfolds not in a questionnaire but through interaction. That is where the next line turns.

4. From the individual to interaction—from the 1950s to the present #

Question. What happens between people, and can it change?

Contribution. The interpersonal circumplex—a map of interaction organized around influence and closeness—attachment theory, interdependence theory, family systems, long-term observation of couples, and therapy shifted some attention toward mutual influence, stress, roles, time, and repair. Some research examines not only how people are selected but also how processes change.

Boundary. Some early studies relied on narrow samples or measured only initial attraction. An average intervention effect does not show which approach suits a particular pair. Relationship persistence does not guarantee quality, freedom, or safety. When there are signs of coercive control, joint assessment or couples therapy may itself be unsafe.

5. Allocation, teams, and platforms—from 1962 to the present #

Question. How can people be allocated while accounting for preferences, roles, resources, and constraints?

Contribution. The Gale–Shapley model made the assumptions behind stable allocation explicit. Here, “stability” means that there are no two participants who would both prefer each other to their current options. Research on person–environment fit and team composition added demands and roles; Operation Match and online dating brought questionnaires, exposure, and mutual responses to a much larger scale.

Boundary. Formal stability is not psychological fit, justice, or safety. A platform is not a neutral observer either: in an experiment described by OkCupid itself, the score shown to users changed their initial messaging. This still tells us nothing about long-term relationships.

6. Machine learning and language models—from 2017 to the present #

Question. Can a large set of features or the language of an interaction produce an individualized prediction?

Contribution. In 2017 speed-dating studies, more than one hundred features measured before the meeting did not reliably predict which specific person someone would be especially attracted to after participants’ general tendencies were taken into account. In a 2020 analysis of 43 datasets, some assessments of existing romantic relationships were informative about their current, self-reported quality; change over time was harder to predict. A 2026 study found modest additional information in language from a conversation that had already taken place.

Boundary. These findings do not establish automatic matching before a meeting, a causal mechanism, or durable benefits for well-being and safety. Using large language models (LLMs) to summarize conversations, generate questions, or support decisions remains a proposed application that must be evaluated separately.

What the collected material supports #

Who meets whom #

Observed pair formation may reflect preferences, geography, family, law, markets, social boundaries, and the people a platform chooses to show. Statistical similarity between partners is called homogamy or assortative pair formation. But group-level similarity does not show that similarity improved the relationship of any particular people.

What happens inside relationships #

Two people’s answers about their shared life are related, so they cannot be analyzed like questionnaires from independent strangers. Actor–partner models distinguish the association between a person’s own characteristics and their response from the association between their partner’s characteristics and that response. These models do not reveal a causal mechanism on their own: a shared event, stress, a role, or context may explain the same pattern.

In samples of existing romantic couples, measures of commitment, appreciation, satisfaction, and conflict helped describe current relationship quality as reported by the participants themselves. This does not establish causation; future change was harder to predict, and the findings do not automatically transfer to other relationships. Therapy research supports a separate, narrower conclusion: some processes and outcomes can change.

When measurement changes what is measured #

If people see a recommendation and act on it, the tool is no longer merely describing the situation. Platform studies directly show changes in exposure, mutual responses, and initial messages. Possible effects on expectations, self-description, and long-term relationships require separate evidence.

Interpretations and research hypotheses #

One interpretation of the history is that progress did not lead toward a universal key, but toward separating the questions: introduction from relationship maintenance, individual characteristics from the interaction of a particular pair, current state from change, and description from ethical acceptability.

A second hypothesis is that observing people during a concrete task, disagreement, strain, and repair may add useful information to a static questionnaire. This must be tested against an explicitly named, simpler alternative, for a defined group, situation, outcome, and time frame. A null result must remain acceptable.

History sets requirements for BWB, but does not validate the model itself. We do not know whether the four-level architecture produces more useful conclusions than a good ordinary conversation or a simpler framework. The recurrence of types, traits, and temporal categories in history does not validate Socionics, Psychosophy, Temporistics, their proposed latent processes, or a natural basis for any predisposition.

What else might explain the result #

Apparent fit or misfit may reflect:

• available introductions and opportunities rather than preference;
• shared class, education, religion, language, or geography rather than a special mechanism within the pair;
• power, coercion, discrimination, or economic dependence rather than attachment;
• health, disability, trauma, sleep, stress, caregiving, migration, or war;
• learned communication, role expectations, incentives, or task design;
• a temporary state, measurement error, impression management, or a new life stage;
• reverse direction of influence: a relationship’s good condition may affect responses that are then used as its “predictors”;
• selection and attrition: the intact couples that are studied may differ from couples whose relationships have already ended.

These circumstances cannot be dismissed as noise. They can materially change both the explanation and the practical decision.

Harms caused and paths rejected #

Historical description is not endorsement.

Coercion and the ranking of people #

BWB rejects imposed unions and assignments; eugenic selection, sterilization, segregation, and reproductive coercion; and the use of race, caste, class, sex, disability, or ancestry to rank human worth and rights.

Harms of classifications and platforms #

BWB also rejects the idea that a type is a person’s unchanging essence and therefore determines ability, destiny, virtue, or eligibility for love, work, participation in church, care, and citizenship. It is unacceptable to hide fear or violence inside an aggregate score, collect data opaquely or derive sensitive conclusions from indirect data, manipulate rankings, or automatically exclude people without a legitimate, proportionate, and contestable justification.

Consent and safety are required conditions. Strengths in other areas cannot compensate for their absence.

What must not be inferred #

• Not every culture passed through the same stages.
• Premodern traditions did not form a single whole, and modern measurement is not value-neutral.
• A long-lasting relationship may be bad or unsafe.
• Formal fit does not demonstrate mutual understanding or flourishing.
• A successful prediction in one group does not establish a causal link or mean that the prediction will work for another group or a particular pair.
• An average that is useful for a group is not necessarily useful to individuals.
• The history of typologies does not confirm BWB’s typological mappings.
• The absence of a universal formula does not make careful questions, observation, counseling, or decision support useless.

What BWB takes from this history #

The following framework is BWB’s project synthesis, not a result of historical validation. It proposes four distinct areas of conversation. The first sets a required threshold; the other three use typologies only as revisable hypotheses about narrow latent processes.

In the relocation example, this looks like the following:

1. Value-moral foundation: Can each person freely and safely say “no”? What commitments, responsibilities, forms of reciprocity, and boundaries already exist? This is a normative frame, not a typology or a scale of human worth.
2. Strategic level / Temporistics: What future does each person want, what counts as timely, and how do they connect the past, present, and possible future? In technical terms, BWB links this level to proposed processes of abduction—seeking the best explanation, induction—generalizing from observations, and deduction—deriving consequences from accepted premises—in temporal and existential experience.
3. Operational level / Psychosophy: How should paperwork, income, decisions, burdens, and care be distributed? Here BWB investigates proposed processes of analysis, synthesis, and action organization.
4. Tactical level / Socionics: What information does each person notice? How do they compress and connect it, use it to draw inferences, and update the shared plan? Here BWB proposes a revisable reconstruction of processes for selecting, compressing, organizing, inferring from, and updating partial information models of one shared reality.

A hidden, or latent, process cannot be observed directly: it is only inferred from speech, decisions, actions, and how they change. These mappings do not establish an innate mechanism, ability, deficit, destiny, or rule for a relationship. Culture, law, health, trauma, resources, roles, power, and current state remain independent explanations.

BWB proposes that AI be designed as a limited support tool. It may organize information, surface uncertainty, and suggest questions, but its additional benefit and potential harms still need to be tested. The decision remains with the people involved. AI must not issue an autonomous verdict, conceal safety threats, or replace community, professional help, and wise counsel. When there are signs of coercive control, a joint conversation or assessment may be unsafe and must not continue automatically.

Questions to start with #

1. Can each person safely refuse, slow, or change the decision without threatening their housing, money, documents, work, care responsibilities, or reputation?
2. What kind of relationship and what decision are we discussing?
3. What specific outcome matters to us, and over what time frame?
4. What have we observed directly, and what have we only interpreted through a profile?
5. What circumstances or power imbalance could create the same pattern?
6. What can be learned or renegotiated, and what remains a boundary?
7. What new evidence would make us change our explanation?
8. Who will bear the cost if the recommendation is wrong?

Technical route for researchers #

For each study or system, record:

1. the relationship domain, study population, setting, and historical context of power;
2. the construct being measured and the outcome specified in advance;
3. data available before, during, and after interaction;
4. effects attributable to the individual, the partner, the particular pair, the network, and institutions;
5. baseline, time frame, uncertainty, calibration, and validation in a new sample;
6. subgroup results, missing data, privacy, consent, and possible harms;
7. what the tool does: describes, directs attention, influences interaction, or makes a decision with serious consequences;
8. what evidence would falsify the proposed mechanism or show no additional value over a simple conversation or baseline model.

Priority gaps include non-Western research in primary languages; friendship and care relationships; disability-centered interdependence; queer histories and histories of colonized peoples; non-romantic teams; and long-term evaluation of AI-supported conversations.

Source map #

For a closer look at scientific work, open the compatibility research catalog. Its first release contains 24 works on close relationships, friendship, teams, prediction, and relationship change. Entries explain questions, methods, findings, and limitations, with publication links and text-access information. This is a thematic selection with search records, not an exhaustive systematic review.

Source-by-source details—URL, type, supported conclusions, and limitations—are recorded in raw/general/human-compatibility-history-source-ledger.md. The narrower history of Jung, MBTI, trait traditions, Socionics, Psychosophy, and Temporistics is in Typology Authors Timeline, while the sociological bridges are in Sociological Compatibility Analogues.

Next reading #

The main next step is Four-Level Compatibility Architecture, which explains BWB’s current framework in detail. For inference boundaries, see Epistemic Status and Inference Limits; for validation design, see Compatibility Measurement Methods.