|
Abstract
|
Many of the choices humans make are intrinsically difficult. Indeed, many decisions require solving computationally intractable optimisation problems. To date, however, there is no general theory to granularly characterise the complexity of optimisation problems and its effect on human decision quality. Here, we address this gap by characterising generic (task-independent) complexity metrics of optimisation problems employing Fitness Landscape Analysis (FLA), a framework from operations research. We test the applicability of these complexity metrics through an online experiment. Participants are asked to solve several cases of a canonical computational problem (the knapsack optimisation problem) with varying levels of hardness (as characterised by FLA). We find that these metrics can collectively explain a significant proportion of the variance in human performance, making FLA a promising generalisable framework to characterise the intrinsic complexity of cognitive tasks. This study sheds light on people's information-processing limitations by providing a modelling tool to predict behaviour that can be used to augment and improve the accuracy of current decision-making models.
|