Characterising human cognitive strategies in decision-making

Human decision-making often involves solving complex computational problems, like selecting the best financial product or optimising a grocery shop subject to a budget. Research shows people use simplifying mental shortcuts (heuristics) which can lead to suboptimal choices. However, there is currently no robust way to characterise how individuals approach these difficult problems based on their observed behaviour.

This research aims to develop a data-driven methodology to characterise the decision-making strategies people use. By employing graph neural networks (GNN), our approach addresses the limitations of existing methods, generalising across different decision scenarios without needing to pre-define potential strategies.

Instead, we fit this model using choice process data, such as mouse clicks . This data-driven model accounts for individual variability and randomness in strategies, offering more accurate insights into the cognitive processes involved in decision-making. We also explore how factors like age, working memory, and processing speed influence strategies.

The research aims to transform understanding of decision-making through an innovative, readily implementable open-source methodology. An accurate description of cognitive strategies can help us understand when and how decision strategies fail, and how they can be enhanced to improve decision-making outcomes. Potential applications include designing tailored digital tools to support better financial planning and decision-making, especially for vulnerable groups like older adults.

MDAP's expertise in data analysis and advanced machine learning techniques will be vital for developing and implementing the GNN models to accurately predict human search patterns from the behavioural data.