Skip to main content

Neurocomputational mechanisms underlying age-related performance changes in goal-directed decisions from experience

Periodic Reporting for period 1 - AGERISK (Neurocomputational mechanisms underlying age-related performance changes in goal-directed decisions from experience)

Período documentado: 2016-09-01 hasta 2018-08-31

Goal-directed choices about health, finance, and general well-being are more vulnerable to inferior outcomes at later stages of life. This is an urgent and widespread societal problem especially because human life expectancy is increasing. It is thus important to find out how we can support older adults when making goal-directed decisions. One promising research direction aims to try to understand how and why the neurocomputational mechanisms of goal-directed decision-making change the way they do over the lifespan. Such research may be able to discover how and why the process of ageing results in changes to these mechanisms underlying complex goal-directed decision-making performance.
The research carried out in this project aimed to contribute insights related to the effects of the complexity of the choice environment to this research direction. Together, the overall objectives aimed to identify age-related changes in the neuro-computational mechanisms underlying goal-directed decision making by using experimental and modeling approaches. The ultimate goal was to help identifying insights that can be used to support older adults in their daily decision-making, thus contributing to their general well-being. The project had three overall objectives. The first objective aimed to establish variability in decision strategies between younger and older adults across a range of different choice environments. The second objective aimed to explain such variability mechanistically. As such, the second objective aimed to assess the effects of increasing cognitive limitations across the lifespan by identifying and comparing the cognitive demands of the strategies that younger and older adults are adopting across task demands. The third objective was similar to the first objective but aimed to establish variability in decision strategies across the entire lifespan instead.
Most of the work was focused on studying how older adults behave in the explore-exploit dilemma. The explore exploit dilemma is ubiquitous in daily life because there are often too many options available to explore exhaustively. The n-armed bandit task is a well-known task that is excellently suited for studying this question. We studied how older and younger adults solved four and eight option bandit problems and how performance relates to structural connectivity in cortico-striatal loops and performance on a number of psychometric tests. We found that older adults reliably chose the most rewarding options less often than younger adults, especially in more complex choice environments. The decline in decision-making performance over time is large for more complex choice environments and medium in less complex choice environments. Older adults needed more exploration than younger adults to be able to distinguish between options. Furthermore, causal inference showed performance was connected to a simple math test directly. However, age group was more strongly associated with cortico-striatal connectivity strength, fluid intelligence, working memory, IQ, and risk taking as compared to performance itself. Together, older adults seem to process uncertainty that is associated with options in more complex choice environments sub-optimally, suggesting more limited task representations or inadequate future reward representations. The stronger the demands of the choice environment, the larger the age-difference, indicating limited available cognitive resources in older adults. This result fits to multiple contexts in the complex cognitive aging literature, and in particular to the context of challenges in the maintenance of goal-directed learning mechanisms.
The work performed during the course of the project basically included the standard steps in the research cycle. New data was collected in an experimental setting and already existing data was re-analyzed as well. Part of the work was performed in the context of analysis of these data. The analyses included a number of techniques, including: mixed effects regression modelling, comparing different performance metrics such as choice proportions and regret, and comparing age-related differences in overall performance and performance differences over time across simple and more complex choice environments.
Another part of the work was performed in the context of the implementation of computational models for the task paradigm as well as assessing how well they were able to explain choice behavior of the participants. Models were compared with each other in terms of overall goodness of fit as well as in terms of one-step-ahead model predictions to observed actions. A measure of neural connectivity strength (as based on magnetic resonance imaging-based diffusion tensor imaging measurements) was also compared with the goodness of model fits and model parameters values using correlational analysis as well as generative modeling. Data from a number of questionnaires and psychometric tests were further added in order to work towards building a causal model using directed acyclic graphs. The applicant received training in state-of-the art reinforcement modelling at various tutorials and workshops.
Another part of the work focused on the gamification and rebuilding of the main experiments into web-ready experiments for large-scale data collection. A part of the work involved reprogramming the experiments in order to work on hand-held devices. Another part of the work was carried out to add application functionality and game elements. Initial work was carried out to anticipate the online deployment. The applicant extended his expertise in web-based tools by self-directed learning about designing and implementing web-based applications.
The results were presented at various international conferences and research departments. The results have so far been published in two conference proceedings papers and publications in international journals are being prepared as well. Outreach activities focused on interactive exhibition and presentation at a large yearly event for the general public. The public’s response to these outreach activities has been very strong and positive. Other work involved arranging for effective data protection arrangements and data management.
Scientifically, the project made original contributions by bringing together several fields and addressing one of the central problems of an aging society: how to ensure that older adults continue to lead independent lives and to make goal-directed choices. The main results should be of interest to scholars in aging, decision-making, and neural plasticity and learning. The project raised new research questions about ways to enhance the mental representation of complex choice environments.

Societally, the outcomes of the project may potentially help to empower aging decision makers navigating cognitively demanding choice environments. The insights that the project offers into the neuro-computational mechanisms underlying how these environments influence the process of goal-directed decision making in older adults should be able to help to inform how older adults can remain independent decision makers. Furthermore, because the project provides insights into learning processes relevant for computer-assisted education systems, its findings are of interest to both academia and industry.
Example of the task and example choice profiles