Reward learning during prolonged social isolation

The desire for interpersonal connection is a fundamental human need, yet as many as one in four older people and 15% of adolescents are experiencing social isolation (World Health Organisation, 2021). Social isolation, defined as the objective lack of social contact or support (Gardiner et al., 2018), is associated with a range of physical and mental health outcomes, the severity of which correlates with the extent of isolation (Lee et al., 2021; Perissinotto et al., 2019). The modern human is uniquely challenged by social isolation due to the digitisation of the modern world, as the digital landscape transforms the way in which individuals connect and interact with others. Paradoxically, time on social media and the number of apps an individual uses correlate both with interpersonal connection and isolation across different studies (James et al., 2017; Meshi et al., 2019; Primack et al., 2017, 2019). The underlying mechanisms linking social media use, social isolation, and mental health outcomes could help resolve this apparent contradiction, yet these remain poorly understood. Modern research into the relationship between prolonged social isolation and social reward learning in humans has, thus far, been translational in nature: aiming to leverage the findings from acute isolation in human models and prolonged isolation in animal models to extrapolate potential relationships. Animal research into prolonged social isolation has consistently demonstrated a relationship between social isolation and reward learning (see Lee et al., 2021 for a review). For example, early-life isolation from cagemates resulted in an increase in socially conditioned place preference (Nardou et al., 2019). Novel research has indicated that acute social isolation in both rodents and humans can result in a craving effect akin to that of malnutrition and hunger signals for subsequent social rewards – suggesting a ‘social homeostasis’, whereby there is an adaptive function to regulate behaviours that govern social connection to an optimal level (Lee et al., 2021; Tomova et al., 2020). Computational modelling approaches of general reward learning following acute social isolation have demonstrated that acute isolation confers higher reward sensitivity (Gigli et al., 2024). Notably, the influence of ‘acute’ social isolation on learning rate has showed contradictory findings, with evidence emerging for both an increase (i.e., Tomova et al., 2022, in which learning rates from negative prediction errors for social feedback were higher in isolation groups compared to baseline) and decrease (i.e., Gigli et al., 2024, in which lower daily social interactions predicted lower reward learning rate). Further, there is evidence in adulthood that chronic social isolation results in anhedonic behaviours, suggesting an impact of chronic isolation on reward processing (Wallace et al., 2009). Thus, until recently, the hypothesis that prolonged social isolation would significantly impact social reward learning in humans remained reliant primarily on lab-based data of acute isolation, rather than more ecologically valid real-world social media data. Objective social isolation became a widespread phenomenon following the lockdown procedures implemented in attempt to mitigate the spread of COVID-19. Global governments were tasked with containing the spread of the disease, which resulted in unprecedented 'social distancing' strategies to minimise face-to-face social interaction. As such, people were required to maintain their 'social homeostasis' through non-traditional means, given the unprecedented level of social isolation and disconnection (Tull et al., 2020). One such avenue of fulfilling this social need was the increased uptake and usage of social media: where physical interaction was at a global minimum, people spent more time on social media than ever before (Drouin et al., 2020). Social media and digital technologies became the primary means for accessing information about the pandemic, and accessing social support (Saud et al., 2020). However, no study has made use of naturalistic social media data in order to empirically evaluate how the pandemic context altered the very way individuals utilised social media and their appraisal of the social reward this provided. This study aims to synthesise these convergent studies by fitting computational models of reward learning to Twitter data from users in New York and California, following the models designed by Turner et al. (2024). We leverage COVID-19 as a natural experiment whereby we can explore how prolonged regional-level social isolation impacted how individuals interact with social media – in a context where opportunities for real-life social interactions were significantly reduced. Through this, we aim to evaluate how different aspects of reinforcement learning and/or habit formation were impacted by chronic social isolation. Thus, we examine the following research question at both a between- and within-user level: What is the relationship between levels of social isolation and social media RL parameters?