Switching to online: Testing the validity of supervised remote testing for online reinforcement learning experiments.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Behavioral Research Methods

Background:

• Online experiments offer alternatives to traditional lab settings.
• Conducting complex, supervised behavioral research online presents challenges.
• The COVID-19 pandemic necessitated remote research methodologies.

Purpose of the Study:

• To assess the computational validity and feasibility of remote, synchronous reinforcement learning (RL) experiments.
• To detail the adaptation of an in-person behavioral experiment for online, supervised delivery.
• To demonstrate the replication of working memory (WM) load effects on RL performance in a remote setting.

Main Methods:

• Utilized open-source software for data collection, statistical analysis, and computational modeling.
• Developed Python code to simulate computational models of working memory (WM) load's impact on RL.
• Adapted an existing in-person behavioral task for supervised remote execution.

Main Results:

• Behavioral data successfully replicated previously observed effects of working memory (WM) load on reinforcement learning (RL) performance.
• Computational analyses using Python code confirmed the accurate simulation of WM load effects on RL.
• The study demonstrated the reliability of algorithms and optimization methods in reproducing observed behaviors.

Conclusions:

• Remote, supervised RL experiments are computationally valid and feasible for complex behavioral research.
• The methodology allows for the replication of in-person findings in online settings.
• This approach provides a valuable resource for researchers adapting long and complex experiments for online delivery.