Neurocomputational mechanisms of approximative decision making based on forward planning using state abstraction

• Applicant: Professor Dr. Stefan Kiebel
• Subject Area: Biological Psychology and Cognitive Neuroscience
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 538763959

Forward planning is essential for making decisions that aim at achieving long-term goals through coordinated sequences of actions. In the fields of psychology and cognitive neuroscience, prevailing computational models assume exhaustive forward planning. These models simulate all conceivable future pathways, a methodology well-suited for standard experimental tasks. However, when applied to tasks of greater complexity, which are frequently encountered in our dynamic and uncertain environment, exhaustive forward planning would result in overly prolonged decision times due to the vast array of potential future trajectories. Research indicates that the human brain uses a more efficient method involving non-exhaustive planning, facilitating rapid decision-making. The exact process behind this rapid and approximate decision-making in humans remains unclear. The challenge lies in understanding how humans can effectively make forward planning decisions without resorting to exhaustive simulations of possible trajectories. This is crucial given that the brain's approach to forward planning shapes decision-making processes, potentially resulting in biases toward or against certain action sequences. This project proposes a new approach inspired by so-called state abstraction algorithms used in computer science to address this question. These algorithms were purposefully designed to enable artificial agents to make approximate and thereby rapid decisions in complex environments. Through the adaptation of these algorithms to cognitive models of human decision-making, the project endeavors to probe whether humans employ comparable strategies. With the resulting new models, we expect to better model approximate decision making of human participants. To demonstrate this, we plan to perform behavioural and functional magnetic resonance imaging experiments. The collected data will be analysed using probabilistic inference models, Bayesian model comparison, model-based analyses and representational similarity analysis. In summary, this project aims at providing a mechanistic explanation for one of the core features of human decision making, to make fast and approximate decisions in our complex and uncertain environment.

• DFG Programme: Research Grants