Value-based decision making via sequential sampling with hierarchical competition and attentional modulation

Active reinforcement learning versus action bias and hysteresis: control with a mixture of experts and nonexperts

Active reinforcement learning enables dynamic prediction and control, where one should not only maximize rewards but also minimize costs such as of inference, decisions, actions, and time. For an embodied agent such as a human, decisions are also shaped by physical aspects of actions. Beyond the effects of reward outcomes on learning processes, to what extent can modeling of behavior in a reinforcement-learning task be complicated by other sources of variance in sequential action choices? What of the effects of action bias (for actions per se) and action hysteresis determined by the history of actions chosen previously? The present study addressed these questions with incremental assembly of models for the sequential choice data from a task with hierarchical structure for additional complexity in learning. With systematic comparison and falsification of computational models, human choices were tested for signatures of parallel modules representing not only an enhanced form of generalized reinforcement learning but also action bias and hysteresis. We found evidence for substantial differences in bias and hysteresis across participants-even comparable in magnitude to the individual differences in learning. Individuals who did not learn well revealed the greatest biases, but those who did learn accurately were also significantly biased. The direction of hysteresis varied among individuals as repetition or, more commonly, alternation biases persisting from multiple previous actions. Considering that these actions were button presses with trivial motor demands, the idiosyncratic forces biasing sequences of action choices were robust enough to suggest ubiquity across individuals and across tasks requiring various actions. In light of how bias and hysteresis function as a heuristic for efficient control that adapts to uncertainty or low motivation by minimizing the cost of effort, these phenomena broaden the consilient theory of a mixture of experts to encompass a mixture of expert and nonexpert controllers of behavior.

Decision heuristics in contexts integrating action selection and execution

Heuristics can inform human decision making in complex environments through a reduction of computational requirements (accuracy-resource trade-off) and a robustness to overparameterisation (less-is-more). However, tasks capturing the efficiency of heuristics typically ignore action proficiency in determining rewards. The requisite movement parameterisation in sensorimotor control questions whether heuristics preserve efficiency when actions are nontrivial. We developed a novel action selection-execution task requiring joint optimisation of action selection and spatio-temporal skillful execution. State-appropriate choices could be determined by a simple spatial heuristic, or by more complex planning. Computational models of action selection parsimoniously distinguished human participants who adopted the heuristic from those using a more complex planning strategy. Broader comparative analyses then revealed that participants using the heuristic showed combined decisional (selection) and skill (execution) advantages, consistent with a less-is-more framework. In addition, the skill advantage of the heuristic group was predominantly in the core spatial features that also shaped their decision policy, evidence that the dimensions of information guiding action selection might be yoked to salient features in skill learning.

Reinforcement learning with associative or discriminative generalization across states and actions: fMRI at 3 T and 7 T

The model-free algorithms of "reinforcement learning" (RL) have gained clout across disciplines, but so too have model-based alternatives. The present study emphasizes other dimensions of this model space in consideration of associative or discriminative generalization across states and actions. This "generalized reinforcement learning" (GRL) model, a frugal extension of RL, parsimoniously retains the single reward-prediction error (RPE), but the scope of learning goes beyond the experienced state and action. Instead, the generalized RPE is efficiently relayed for bidirectional counterfactual updating of value estimates for other representations. Aided by structural information but as an implicit rather than explicit cognitive map, GRL provided the most precise account of human behavior and individual differences in a reversal-learning task with hierarchical structure that encouraged inverse generalization across both states and actions. Reflecting inference that could be true, false (i.e., overgeneralization), or absent (i.e., undergeneralization), state generalization distinguished those who learned well more so than action generalization. With high-resolution high-field fMRI targeting the dopaminergic midbrain, the GRL model's RPE signals (alongside value and decision signals) were localized within not only the striatum but also the substantia nigra and the ventral tegmental area, including specific effects of generalization that also extend to the hippocampus. Factoring in generalization as a multidimensional process in value-based learning, these findings shed light on complexities that, while challenging classic RL, can still be resolved within the bounds of its core computations.

Computational animal welfare: towards cognitive architecture models of animal sentience, emotion and wellbeing

To understand animal wellbeing, we need to consider subjective phenomena and sentience. This is challenging, since these properties are private and cannot be observed directly. Certain motivations, emotions and related internal states can be inferred in animals through experiments that involve choice, learning, generalization and decision-making. Yet, even though there is significant progress in elucidating the neurobiology of human consciousness, animal consciousness is still a mystery. We propose that computational animal welfare science emerges at the intersection of animal behaviour, welfare and computational cognition. By using ideas from cognitive science, we develop a functional and generic definition of subjective phenomena as any process or state of the organism that exists from the first-person perspective and cannot be isolated from the animal subject. We then outline a general cognitive architecture to model simple forms of subjective processes and sentience. This includes evolutionary adaptation which contains top-down attention modulation, predictive processing and subjective simulation by re-entrant (recursive) computations. Thereafter, we show how this approach uses major characteristics of the subjective experience: elementary self-awareness, global workspace and qualia with unity and continuity. This provides a formal framework for process-based modelling of animal needs, subjective states, sentience and wellbeing.

Goal congruency dominates reward value in accounting for behavioral and neural correlates of value-based decision-making

When choosing between options, whether menu items or career paths, we can evaluate how rewarding each one will be, or how congruent it is with our current choice goal (e.g., to point out the best option or the worst one.). Past decision-making research interpreted findings through the former lens, but in these experiments the most rewarding option was always most congruent with the task goal (choosing the best option). It is therefore unclear to what extent expected reward vs. goal congruency can account for choice value findings. To deconfound these two variables, we performed three behavioral studies and an fMRI study in which the task goal varied between identifying the best vs. the worst option. Contrary to prevailing accounts, we find that goal congruency dominates choice behavior and neural activity. We separately identify dissociable signals of expected reward. Our findings call for a reinterpretation of previous research on value-based choice.

Decision-making research has confounded the reward value of options with their goal-congruency, as the task goal was always to pick the most rewarding option. Here, authors separately asked participants to select the least rewarding of a set of options, revealing a dominant role for goal congruency.