Modeling transitive-inference-related decision making using the drift diffusion model reveals generalizable latent cognitive mechanisms across human and non-human primates

Decision-making often relies on mental models built by linking experienced facts or stimuli. This allows individuals to decide between options by evaluating information that is not directly accessible from perceptual evidence alone. This form of decision-making can be studied through transitive inference (TI) tasks, in which subjects learn the rank order of a set of abstract, arbitrarily defined symbols (e.g., A>B>C>D>E>F) and use this structure to infer the rank relation between any pair of items in the set - for example, choosing the higher between B and E. In performing this task, both humans and animals exhibit the symbolic distance (SD) effect: faster and more accurate decisions when the distance between the items’ ranks is greater. This effect emerges because items with similar ranks have closer representations in the hierarchical structure of the set. The Drift Diffusion Model (DDM) is a computational approach developed to evaluate the variables underlying decision-making processes. By conceptualizing decision-making as a ramping-to-threshold process, the DDM can determine whether decision difficulty depends on the accumulation rate (v) or on the amount of evidence required to reach a decision, defined as decision boundary (α). Previous studies on perceptual decision-making have highlighted variation in v as the main factor accounting for different levels of decision difficulty. In this study, we assessed if the DDM could reliably account for decision-making in the TI task and which variables best explained decision difficulty in this task. Fitting the data from 83 humans and two rhesus macaques on a six-item TI task, the DDM reproduced both accuracy and reaction times with a high goodness of fit. However, the analysis of the model parameters revealed that v was the main factor accounting for task difficulty in humans. In monkeys, task performance was influenced by both v and a time-dependent collapse of α, reflecting increased urgency in the monkeys’ responses and a reduced level of decision caution as time progressed. To further explore the mechanism underlying the decision processes, we used the DDM to model the monkeys' performance after the injection of subanesthetic dosage of ketamine, an NMDA-receptor antagonist that we hypothesized would interfere with the ramping process, as it has been shown to impair rule-based decisions. After pharmacological treatment, we observed a reduced v along with a more pronounced collapse of α, indicating more impulsive responses. These results highlight that the DDM can be consistently applied across diverse decision-making processes to estimate the latent variables underlying cognitive decision mechanisms.