We propose that, for each familiar individual, i, the kin detection system computes and updates a continuous variable, the kinship index, KIi, that corresponds to the system’s pairwise estimate of genetic relatedness between self and i. These computational elements are regulatory variables that serve as input to neural programs regulating altruism towards i and, separately, to programs regulating sexual behaviour towards i.
Because relatedness cannot be directly observed, the system must be designed to register cues relevant to determining relatedness. To compute the kinship index, the system requires (1) monitoring circuitry designed to register cues to relatedness, and (2) a computational device, the kinship estimator, whose procedures have been tuned by a history of selection to take these registered inputs and transform them into a kinship index.
The cues the system uses cannot simply be derived ontogenetically (‘learned’) by identifying which arbitrary and transient cues happen to best predict relatedness in the local environment. To do this, the system would have to already know the relatedness of others—the very problem it needs to solve. Instead, the kin detection system must contain within its evolved design a specification of the core cues that it will use to determine relatedness—cues that reliably tracked genetic relatedness in the ancestral social environments that selected for the kin detection system.
For human foragers, a potentially informative cue to kinship is provided by the close perinatal association between mother and neonate that begins with birth and is enforced by the exigencies of early mammalian maternal care. Maternal perinatal association (MPA) provides a basis for the reliable mutual detection of mother and offspring and can, in turn, be used as an anchor point for sibling detection. Ancestrally, if an individual observed an infant in a durable, perinatal association with the individual’s mother, then it was highly probable that that infant was the individual’s sibling. We therefore proposed that sibling detection includes a monitoring subsystem specialized for registering MPA.
Although MPA is likely to be the single most informative cue, it cannot be used (for example) by younger siblings, because they are not alive at the time their older siblings are born and nursed. When MPA is unavailable, the kinship estimator should fall back on other cues that were highly predictive ancestrally. We predicted that the kin detection system would include a second subsystem specialized for registering the cumulative duration of coresidence summed over the full period they receive parental care. Ancestrally, parents (especially mothers) maintained close association with their children to care for them, and for this reason siblings co-associate statistically more than non-siblings. (Indeed, given the fusion–fission pattern of hunter–gatherer association, this same variable should—to some extent—link progressively more distant genetic relatives to increasingly diluted motivational residues.) Among human foragers, the maintenance of parental proximity for care delivery begins with birth and tapers off in late adolescence, a time when offspring become nearly independent adult foragers and when mating motivates new patterns of co-association36,37
. Although this hypothesis differs from the ethological proposal of a period of early childhood imprinting35
, it is consistent with evidence that suggests that familiarity is a cue mediating kin detection in non-human primates14,15,38,39
The kinship estimator consists of algorithms for transforming the registered cues into the kinship index, a variable whose magnitude tracks relatedness between self and other. If the cues are integrated into a single index, then we should find that the same patterns of inputs are associated with the same patterns of outputs for both altruism and sexual aversion. This model (summarized in ) leads to the following predictions.
Proposed model of the computational architecture of sibling detection
- When MPA is absent, coresidence duration before adulthood with an individual should (a) upregulate altruism towards that individual, (b) upregulate sexual aversion towards that individual, and, as a by-product, (c) upregulate moral opposition28,29 to third-party sibling incest.
- When MPA is present, it should produce the same three effects.
Selection should have tuned the procedures in the kinship estimator to use MPA and coresidence in a way that takes account of their relative informativeness and availability. Because MPA is the more robust, higher quality cue, we expect that when both are available, coresidence will be weighted by the kinship estimator far less than MPA, and perhaps not at all. Therefore, we propose a third prediction.
- When MPA is present, coresidence duration will not be as strong a predictor of altruistic motivations and sexual aversions. That is, the kinship estimator will use MPA in preference to coresidence duration in computing kinship.