Transitive responding in animals and humans: Exaptation rather than adaptation?

In order to survive and reproduce, individual animals need to navigate through a multidimensional utility landscape in a near-optimal way. There is little doubt that the behaviourally more advanced species can bring cognitive competencies to bear on this difficult task. Among the cognitive abilities that are helpful in this context is transitive inference. This is typically the competency to derive the conclusion B>D from the premises A>B, B>C, C>D and D>E that imply the series A>B>C>D>E. In transitive inference tests used with humans, the letters stand for verbal items and the inequality symbols stand for a relational expression. To investigate analogous competencies in non-human animals a non-verbal form of the task is used. The premise pairs are converted into a multiple instrumental discrimination task A+B-, B+C-, C+D- and D+E-, where the letters stand for non-verbal stimuli and the plus and minus symbols indicate that choices of the corresponding stimuli either lead to a reward or to a penalty. When these training pairs are adequately discriminated, transitive responding is tested with intermittent presentations of the novel pair B∘D∘, where the circles indicate that responses to the stimuli are not reinforced. Using variants of this basic conditioning task it has been shown that pigeons, rats, squirrel-monkeys, macaques, chimpanzees, young children, older children and adult humans commonly reveal transitive preferences for B over D. Several theories have been proposed to explain this transitive behaviour. The evidence supporting these various models is reviewed. It is shown that the learning of the premises normally brings about a choice and reinforcement biasing and balancing process that can account for transitive responding. It is argued that a very simple algebraic learning model can satisfactorily simulate many of the results obtained in transitivity experiments, including some produced by human subjects who in principle, could have been applying rational logical rules. It is demonstrated that a value transfer mechanism also assumed to explain transitive responding, is in fact, a real phenomenon based on classical conditioning. However, it is argued that it mostly plays a minor role in transitive responding. It is shown that the algebraic learning model can be easily converted into a neural network model exhibiting an equivalent performance. The model can also be modified to cope with the surprising finding that a proportion of human individuals and a few animals subjects learn to discriminate the premise pairs, but nevertheless fail to respond transitively to the conclusion pair. This modification can simulate the results of experiments using non-linear, in particular circular, relational structures. The evolution of transitive responding is considered within the framework of ecosocial demands and neurobiological constraints. It is concluded that, in agreement with a preadaptation (exaptation) evolutionary origin, it seems to involve little beyond the capacity to learn multiple stimulus discriminations.