Redefining Fairness: Balancing Engagement and Well-Being in Social Media Algorithms

1. Introduction

Social media platforms rely on AI-driven predictive systems to deliver tailored content, moderate harmful material, and profile user behaviour, processing vast datasets to maximise engagement [1]. These systems, powered by machine learning (ML), predict user preferences to sustain scrolling, often exploiting cognitive vulnerabilities to keep users online [2]. This relentless pursuit of engagement raises ethical concerns about fairness, defined here as the equitable balance between platform goals and user well-being, ensuring systems do not excessively consume time, reduce productivity, or impair attention spans [3,4].

Inspired by critiques of engagement-driven design, this paper investigates fairness in social media predictive systems, focusing on their role in time exploitation. We pursue three objectives: (1) to analyse the technical, psychological, and ethical mechanisms driving excessive engagement, (2) to evaluate real-world impacts and mitigation strategies through case studies of Facebook, Twitter/X, TikTok, and YouTube, and (3) to propose a governance framework prioritising user autonomy and societal well-being. Key research questions include: How do predictive systems sustain compulsive scrolling? What are the societal consequences of time exploitation? How can platforms ensure fairness without compromising functionality? By synthesising literature (2020–2025), case studies, and quantitative metrics, this study contributes to AI ethics, offering a novel perspective on fairness as non-exploitative, user-centric design that protects productivity and cognitive health.

2. Literature Review

Social media predictive systems are designed to maximise engagement, drawing on behavioural psychology and ML to deliver hyper-relevant content [5,6]. This section reviews engagement-driven systems’ mechanisms, impacts, and ethical considerations, framing fairness as mitigating time exploitation.

2.1. Mechanisms of Engagement

Predictive algorithms employ reinforcement learning and collaborative filtering to deliver content that sustains attention, exploiting cognitive biases such as novelty-seeking and social validation [7,8]. Features like infinite scroll and autoplay create seamless, addictive experiences, increasing dwell time [9]. A 2024 study highlights how algorithms prioritise emotionally charged content, amplifying engagement at the cost of cognitive control [10]. This dynamic challenges fairness by prioritising platform metrics over user autonomy [11].

2.2. Psychological and Societal Impacts

Excessive social media use correlates with reduced productivity, shortened attention spans, and mental health issues [4,12]. Heavy usage disrupts workplace efficiency and academic performance, particularly among adolescents [13]. Algorithms amplifying polarising content deepen societal divisions, undermining democratic discourse [14,15]. A 2023 study links prolonged scrolling to decreased cognitive flexibility, as users struggle to disengage [9]. These impacts highlight the need for fairness to encompass user well-being and societal cohesion [2].

2.3. Algorithmic Bias and Time Exploitation

Algorithmic bias exacerbates time exploitation by targeting vulnerable users with hyper-engaging content [16]. For instance, recommendation systems often prioritise sensationalist or emotionally charged material, disproportionately affecting adolescents and heavy users [10]. Traditional bias mitigation strategies, such as fairness-aware algorithms, focus on equitable outcomes but rarely address engagement-driven harms [17,18]. A 2024 review advocates for metrics prioritising user agency and cognitive health, aligning with the need for non-exploitative design [19].

2.4. Ethical and Regulatory Perspectives

Ethical frameworks emphasise transparency, accountability, and user empowerment as pillars of fair AI systems [20]. Transparency in algorithmic processes enables users to understand engagement mechanisms, while accountability ensures platforms address harms [21]. Regulatory efforts, such as the EU’s Digital Services Act (2024) and California’s 2023 anti-addictive design legislation, mandate disclosure of recommendation systems and protections against manipulative features [22,23]. However, the absence of standardised metrics for time exploitation hinders progress [24]. Recent studies propose human-centred AI that optimises for well-being, offering a path toward fairness [25].

2.5. Gaps and Contributions

While prior work addresses algorithmic bias and fairness in terms of equitable outcomes [1], few studies focus on time exploitation as a fairness issue. This paper fills this gap by redefining fairness to prioritise user autonomy, productivity, and attention, drawing on psychological and societal insights to propose actionable solutions.

3. Methodology

This study employs a mixed-methods approach to investigate fairness in social media predictive systems, focusing on time exploitation on productivity and attention. Qualitative case studies of Facebook, Twitter/X, TikTok, and YouTube explore engagement-driven patterns and mitigation strategies. Using simulated datasets to replicate real-world interactions, Quantitative analysis assesses dwell time, session length, user-reported productivity, and attention span impacts.

3.1. Data Sources

Data is sourced from publicly available platform reports, peer-reviewed studies (2020–2025), user surveys (e.g., Pew Research Center, 2023), and synthetic datasets designed to replicate real-world social media interactions. Synthetic datasets are employed to model user engagement patterns (e.g., click-through rates, session durations, content exposure) due to ethical, practical, and scientific considerations. Real-world user data from platforms like Facebook, Twitter/X, TikTok, and YouTube is often proprietary, inaccessible to researchers, and subject to strict privacy regulations, such as GDPR and CCPA [22,24]. Using synthetic datasets ensures ethical compliance by avoiding direct handling of personal data, protecting user privacy while enabling robust analysis [26].

Synthetic datasets are constructed using statistical models informed by aggregated, anonymised data from public sources, such as platform transparency reports and studies like Amarikwa [27] and Farayola et al. [28]. These datasets replicate realistic interaction scenarios, controlling for variables like user demographics, content type, and engagement frequency to isolate the effects of algorithmic design on fairness, productivity, and attention spans [29]. This approach allows for reproducible experiments and hypothetical scenario testing (e.g., the impact of engagement caps), which are critical for evaluating proposed interventions [25]. To mitigate limitations, such as potential oversimplification of real-world complexities, findings are triangulated with qualitative case studies and user-reported data, ensuring credibility and robustness [30].

3.2. Analysis

Qualitative data is analysed thematically to identify engagement patterns and mitigation approaches, following Braun and Clarke’s [31] framework. Themes include addictive design, attention fragmentation, and user agency. Quantitative analysis uses statistical tools like regression and ANOVA to evaluate relationships between dwell time, productivity, and attention metrics, validated through standard ML evaluation criteria [29]. Fairness is assessed by measuring disparities in engagement impacts across user groups (e.g., age, usage frequency).

3.3. Validation

Findings are validated through triangulation with independent studies, expert consultations, and user-reported data, ensuring robustness despite proprietary data limitations [30]. This methodology bridges technical, psychological, and ethical perspectives, offering a comprehensive analysis of fairness as non-exploitative design.

4. Key Challenges in Ensuring Fairness

The pursuit of engagement through predictive systems presents several challenges to fairness, each with significant psychological and societal implications:

• Addictive Design Mechanisms: Algorithms exploit cognitive biases, such as dopamine-driven feedback loops, to maximise dwell time, reducing productivity and fostering compulsive behaviours [6,9]. Features like infinite scroll and autoplay create seamless engagement loops, challenging user autonomy [8].
• Attention Fragmentation: Continuous scrolling shortens attention spans, impairing cognitive performance and academic outcomes, particularly among youth [4,13]. This undermines fairness by prioritising platform metrics over cognitive health.
• Disparities in Vulnerability: Adolescents, heavy users, and marginalised groups are disproportionately affected by engagement-driven systems, raising equity concerns [10,12]. For example, emotionally charged content targets vulnerable users, amplifying time exploitation.
• Lack of User Agency: Limited transparency and control over recommendation algorithms hinder users’ ability to disengage, violating principles of fairness [3,11]. Users report feeling “trapped” by personalised feeds, with little recourse to customise their experience [32].
• Societal Polarisation: Algorithms amplifying polarising content deepen societal divisions, eroding trust and democratic discourse [14,15]. This indirect harm of engagement-driven systems challenges fairness by prioritising clicks over cohesion.

These challenges highlight the need for a redefined fairness that protects user well-being and societal health, addressing both individual and collective harms of time exploitation [2].

5. Case Study Analysis

To ground the analysis, we examine four major platforms, assessing how their predictive systems drive time exploitation and the effectiveness of mitigation strategies.

5.1. Facebook: Engagement-Driven Content Delivery

Facebook’s 2021 recommendation algorithm prioritised content likely to elicit reactions, increasing session lengths but reducing user-reported productivity [30]. Surveys indicate that 60% of users felt distracted from work or studies due to prolonged scrolling [32]. In 2022, Facebook introduced time management tools, such as usage reminders and “quiet mode,” reducing average dwell time by 10% but lowering ad revenue by 8% [20]. While these tools enhanced user control, their optional nature limited adoption, highlighting the challenge of balancing fairness with commercial interests [11].

5.2. Twitter/X: Polarisation and Compulsive Scrolling

Twitter/X’s 2022 algorithm amplified sensationalist and polarising content, extending sessions and fragmenting attention, with users reporting a 15% decrease in focus during work hours [32,33]. A 2023 update introduced optional chronological feeds and engagement nudges, improving user agency but reducing platform engagement by 10% [34]. This case underscores the role of user-driven customisation in mitigating time exploitation, though widespread adoption remains a barrier [35].

5.3. TikTok: Hyper-Engaging Recommendations

TikTok’s 2023 algorithm excelled at delivering short, emotionally charged videos, driving compulsive scrolling, particularly among adolescents, who averaged 90 minutes daily on the platform [28]. This led to reported declines in academic performance and attention spans [10]. A 2024 update introduced time limits, well-being prompts, and a “less engaging” feed option, reducing session lengths by 12% but impacting user retention [..]. TikTok’s case illustrates the tension between fairness and maintaining a competitive user base, especially for youth-focused platforms [12].

5.4. YouTube: Autoplay and Attention Capture

YouTube’s 2024 autoplay feature increased dwell time by seamlessly queuing videos, with users reporting a 20% reduction in productivity during leisure hours [32]. The platform’s algorithm prioritised watch time, amplifying engaging but often low-value content [10]. In response, YouTube implemented explainable AI to disclose recommendation logic and introduced autoplay toggles, reducing session lengths by 12% with minimal engagement loss [36]. Transparency and user controls proved effective, though their impact depended on user awareness and adoption [25].

6. Proposed Solutions and Governance Framework

To achieve fairness by mitigating time exploitation, we propose a multifaceted approach integrating technical, ethical, and regulatory strategies, synthesised into a comprehensive governance framework (see Figure 1). These solutions address engagement-driven harms’ psychological, societal, and technical dimensions, prioritising user autonomy and well-being.

6.1. Technical Interventions

1. Engagement Caps: Implement algorithmic limits on session lengths to prevent compulsive use, such as capping daily scrolling time based on user preferences [37]. For example, platforms could introduce “hard stops” after 60 minutes, requiring active user consent to continue.

2. User Controls: Enable granular customisation of recommendation algorithms, allowing users to disable autoplay, prioritise educational content, or reduce emotionally charged material [3]. A 2023 study found that 70% of users wanted more control over feeds but lacked accessible tools [32].

3. Explainable AI: Provide transparent explanations of content selection, empowering users to understand and adjust algorithmic behaviour [36]. For instance, TikTok could display why a video was recommended (e.g., “based on prior likes”).

4. Well-Being Metrics: Optimise algorithms for productivity, attention span, and mental health rather than engagement, incorporating user feedback and psychometric data [25].

5. Nudge-Based Interventions: Introduce prompts encouraging breaks or alternative activities, such as “You’ve been scrolling for 30 minutes—consider a 5-minute break” [38]. These nudges leverage behavioural science to counteract addictive design [5].

6.2. Ethical Principles

1. Transparency: Disclose how algorithms drive engagement, including metrics like dwell time and content prioritisation, to foster trust and accountability [20]. Platforms should publish annual transparency reports detailing engagement impacts.

2. Accountability: Establish clear responsibility for compulsive design harms, with mechanisms for user redress, such as reporting manipulative features [35]. This aligns with fairness by ensuring platforms bear the cost of unethical design.

3. User-Centric Design: Prioritise autonomy and well-being in system architecture, involving diverse user groups in design processes to reflect varied needs [39]. For example, co-design workshops with adolescents could inform youth-friendly features.

4. Equity in Impact: Ensure mitigation strategies address disparities, protecting vulnerable groups like adolescents from disproportionate harm [12]. This includes tailoring interventions to high-risk users.

6.3. Regulatory Oversight

1. Well-Being Audits: Mandate independent assessments of engagement-driven impacts, including productivity, attention, and mental health, as supported by the EU’s Digital Services Act [22]. Audits should be conducted annually, with public reporting.

2. Anti-Addictive Design Laws: Enforce regulations targeting manipulative features, such as infinite scroll or autoplay, drawing on California’s 2024 legislation [23]. Penalties for non-compliance could incentivise reform.

3. Collaborative Governance: Foster partnerships among developers, regulators, psychologists, and users to align systems with societal values [24]. Multi-stakeholder councils could oversee platform compliance and innovation.

4. Global Standards: Develop international ethical social media design guidelines that address time exploitation across jurisdictions [25]. UNESCO’s 2024 AI ethics framework offers a model for harmonized standards [40].

7. Discussion

The case studies reveal persistent trade-offs between engagement and fairness, with mitigation strategies like time limits and explainable AI showing partial success [28,36]. However, commercial pressures, such as ad revenue declines, pose barriers to widespread adoption [30]. For instance, Facebook’s time management tools reduced dwell time but faced resistance due to revenue impacts, highlighting the need for regulatory incentives [20,23]. TikTok’s youth-focused interventions underscore the importance of targeting vulnerable groups, though scalability remains challenging [10].

The psychological and societal implications of time exploitation are profound. Reduced productivity undermines economic efficiency, with a 2023 study estimating a $500 billion annual loss in global workplace output due to social media distractions [41]. Shortened attention spans impair educational outcomes, with adolescents showing a 25% decline in sustained focus since 2010 [4,16]. Polarisation, driven by engagement-focused algorithms, threatens democratic stability by amplifying divisive narratives [14,15]. These harms challenge fairness by prioritising platform profits over individual and collective well-being [2].

Emerging solutions offer promise but face hurdles. Well-being metrics, grounded in psychometric data, could reorient algorithms toward cognitive health, but require significant computational and ethical oversight [25]. Reinforcement learning with human feedback (RLHF) enables adaptive, user-centric systems, yet demands resources beyond the reach of smaller platforms [26,36]. Nudge-based interventions leverage behavioural science but risk being ignored if not seamlessly integrated [42]. Regulatory frameworks like the EU’s Digital Services Act provide a foundation for accountability, but global coordination is needed to address cross-border platforms [22,40].

This study contributes to AI ethics by redefining fairness as the mitigation of time exploitation, offering a novel framework that integrates psychological, technical, and societal perspectives. Unlike prior work on bias mitigation [17], our approach addresses engagement-driven harms, proposing actionable solutions like well-being metrics and global standards. By prioritising user autonomy, the framework aligns with calls for human-centred AI, extending the discourse on digital well-being [25].

7.1. Limitations

Due to proprietary algorithmic data, reliance on synthetic datasets may limit generalizability, though triangulation with user surveys mitigates this [26,32]. The focus on major platforms (Facebook, Twitter/X, TikTok, YouTube) overlooks smaller ecosystems, which may face unique challenges. Additionally, the proposed interventions assume user willingness to adopt controls, which may vary by demographic [12].

7.2. Future Directions

Future research should explore:

1. Standardized Well-Being Metrics: Develop and validate metrics for productivity, attention, and mental health, applicable across platforms [43].

2. Long-Term Impacts: Investigate cumulative effects of time exploitation on cognitive development, workplace outcomes, and societal cohesion [4,41].

3. Cross-Platform Interventions: Test collaborative frameworks involving multiple platforms to scale solutions, such as shared well-being standards [24].

4. Decentralised Algorithms: Experiment with user-controlled algorithms that prioritise autonomy, reducing reliance on platform-driven feeds [3].

5. Vulnerable Group Protections: Design targeted interventions for adolescents and heavy users, incorporating developmental psychology [12].

6. Global Policy Harmonisation: Advocate for international regulations addressing time exploitation, building on UNESCO’s AI ethics framework [40].

8. Conclusions

This study redefines fairness in social media predictive systems as the mitigation of time exploitation, ensuring platforms protect user productivity, attention spans, and societal well-being. Through case studies and quantitative analysis, we demonstrate persistent trade-offs between engagement and fairness, with mitigation strategies like user controls, well-being metrics, and regulatory audits offering partial solutions. Our proposed governance framework integrates technical, ethical, and regulatory approaches, providing a scalable roadmap for non-exploitative systems. By prioritising user autonomy and cognitive health, social media platforms can foster equitable digital ecosystems that enhance rather than undermine individual and collective outcomes. This work calls for interdisciplinary collaboration among technologists, psychologists, policymakers, and users to redefine the future of social media as a fair, human-centred space.