Written Evidence Submitted by the Digital Mental Health Programme at the University of Cambridge (SMH0027)

This evidence was written by Dr Amrit Kaur Purba, Georgia Turner and Dr Amy Orben

Introduction

The Digital Mental Health Programme

1. The Digital Mental Health Programme, based at the University of Cambridge, is one of the UK’s best funded and most recognised research teams examining the impact of social media on young people.
2. The Programme uses a variety of approaches to study the relationship between technology use and mental health, including psychology, neuroscience, epidemiology and qualitative research with young people.
3. Our broad range of basic and applied research projects feeds directly into national and international policymaking.

Our Team

4. The Programme was founded by Dr Amy Orben, Programme Leader at the MRC Cognition and Brain Sciences Unit, who is a world-leading expert about digital technology use and adolescent mental health.
5. Dr Orben has previously given oral evidence about the topic of screen time, social media and children to the House of Commons Science and Technology Select Committee in 2018 and House of Commons Education Select Committee in 2023. In 2023, she provided verbal briefings about the topic to the Secretary of State for Education, the Minister for Children, the US Surgeon General and the Biden-Harris Administration. She is currently one of 12 experts on the Scientific Advisory Council for the Department of Education. She has published many of the leading scientific articles analysing UK data about this topic over the past 5 years.
6. Dr Amrit Kaur Purba is a Senior Research Associate in the Digital Mental Health Programme. With an expertise in causal epidemiology, Dr Purba plays a pivotal role in public health initiatives and research that addresses the impact of social media on adolescent health-risk behaviours and related health inequalities. Dr Purba has previously given evidence on the impact of social media on adolescent health to a range of national and international bodies, including the United Nations, 10 Downing Street, and the Australian e-Safety Commissioner.
7. Georgia Turner is a PhD student in the group studying technological designs and addiction through a neuroscientific and computational lens. She has previously coauthored a response to a call for evidence on the effects of screentime to the House of Commons Education Select Committee (2023), and a report for the European Joint Research Commission (2024).

This Evidence

8. This evidence is based on a range of studies conducted by the Digital Mental Health Group over the past six years, as well as our deep expertise on the scientific literature in this area.
   

Evidence Detail

Call for evidence: To what extent do the business models of social media companies, search engines and others encourage the spread of harmful content, and contribute to wider social harms?

1. We need to understand the impact of specific design features to identify harms

●        Business models of social media companies are based on increasing and maintaining high levels of user engagement. This motivates the implementation of specific design features that can encourage the spread of harmful content and contribute to wider societal harms such as decreases in adolescent mental health (Orben et al. 2024). Such design features may include the quantification of social feedback through Like counts, algorithmic amplification of certain content and the ‘endless scroll’.

●        Given that these companies rely on making profit from maximising the time users spend online, without regulation or a change in business model it is unlikely that company-initiated interventions like Instagram's ‘Take A Break’ feature and TikTok's opt-out screen time limits will adequately address underlying issues (Purba, Pearce, et al. 2024).

2. Design features interact with psychological biases to amplify harms

●        Design features of social media can interact with and exploit existing psychological biases, which were developed across human evolution to adapt to the demands of offline social environments (Turner, Ferguson, et al. 2024).

●        For example, rewards arriving at unpredictable and irregular times (known as ‘intermittent reward schedules’) are known to encourage and maintain persistent, habitual behaviour (Skinner 1953).

●        Therefore, design features that enable intermittent quantifiable reward feedback such as ‘Likes’ and notifications can encourage habitual social media scrolling, posting and sharing (Turner, Gunschera, et al. 2024). Such habitual posting can contribute to the spread of content which has not been carefully vetted or verified.

●        Research has shown that the pursuit of Likes can train users to post increasingly controversial and incendiary content (Brady et al. 2021). Specifically, more Likes are received for posts containing more moral outrage (McLoughlin, Brady, and Crockett 2021). This causes users to learn to post content containing such outrage to maximise the Likes they receive (Brady et al. 2021).

●        Additionally, humans are biased to prefer low-effort and immediately available rewards. This can increase the appeal of short-form video content platforms such as TikTok, biasing our consumption of news to these formats over traditional, long-form media (Turner, Ferguson, et al. 2024).

3. A summary of individual harms

●        Social media users driven to seek intermittent quantifiable rewards (Likes, Followers, Views, etc) can increase their time spent on social media, which can contribute to increased social comparison and decreased self-esteem (Orben et al. 2024; Purba, Pearce, et al. 2024).

●        For example, given that the content we view and experience shapes our body image, viewing idealised, edited photographs of others on social media can negatively impact our own self esteem (Turner, Ferguson, et al. 2024).

●        Design features that increase time spent on social media can also result in negative health outcomes through displacement of other activities such as sleep (Alonzo et al. 2021).

●        Harmful content can also encourage harmful offline behaviours. A growing body of evidence links exposure to online content depicting risky behaviours with engagement in similar offline behaviours (Purba, Thomson, et al. 2023). Posts showing underage alcohol consumption (often in violation of social media platforms' terms of service), can be accessible to adolescents on social media platforms (Purba, Thomson, et al. 2023; Ofcom 2024; Cookingham and Ryan 2015).

●        Consistent with the hypothesis that exposure to such content increases health risk behaviours, research has also shown that time spent on social media increases risk both of alcohol (Purba, Henderson, et al. 2023), e-cigarette, and cigarette use in adolescents (Purba, Henderson, et al. 2024).

●        The risks extend beyond individual users. Research highlights the 'peer group effect,' where adolescents are more likely to share risk-related content online to gain peer approval and acceptance (Brown et al. 2008). These findings are consistent with the Facebook Influence Model, which posits that peer influence processes are magnified in the social media environment (Moreno and Koff 2016). As a result, an adolescent's deliberate sharing of content—or even a curiosity-driven or accidental click—can shape not only their social norms and behaviours but also those of their online peer group (Moreno et al. 2009; Purba, Thomson, et al. 2023; Moreno and Koff 2016, 201).

●        This dynamic is not limited to health-risk behaviours like substance use, but can also extend to other forms of risky conduct, such as knife crime. For instance, the sharing or engagement with violent content or posts glamorising gang culture can normalise such behaviours within an adolescent's online community. The ripple effect of these interactions may escalate real-world consequences, highlighting the critical need for targeted interventions that address both the online and offline dimensions of risk behaviours (Purba, Thomson, et al. 2023).

4. A summary of societal harms

●        Design features of social media can increase the spread of content which evokes emotion such as outrage, and which contains misinformation (McLoughlin et al. 2024).

●        Such harmful content can influence public discourse, which has downstream impacts on wider society (Purba, Pearce, et al. 2024).

●        Consuming anti-science content can undermine public trust in science and public institutions, and in some cases, lead to lower adherence to public health recommendations, as demonstrated by COVID-19’s ‘infodemic’ (Purba, Pearce, et al. 2024).

●        Amplified prejudice (e.g. racism) can impede global efforts to advance healthcare and human rights. In the context of the climate emergency, dis/misinformation poses a concerning obstacle to necessary action with far-reaching health implications (Purba, Pearce, et al. 2024).

Call for evidence: How do social media companies and search engines use algorithms to rank content, how does this reflect their business models, and how does it play into the spread of misinformation, disinformation and harmful content?

●        Ranking algorithms prioritise topics that capture attention, which tend to be those that are more controversial, including racism, sexism, and xenophobia (Purba, Pearce, et al. 2024).

●        Research has shown that misinformation sources evoke more outrage than do trustworthy sources, and that users are more likely to share outrage-evoking misinformation without reading it first (McLoughlin et al. 2024).

●        Thus, algorithms designed to maximise attention can directly increase the likelihood users will share and consume misinformation on social media.

●        When humans encounter new information, they have certain psychological biases such as confirmation bias – the tendency to give higher weight to new information which confirms what we already believe (Palminteri and Lebreton 2022).

●        In offline environments, confirmation bias is beneficial because it allows us to form more stable beliefs (Palminteri and Lebreton 2022).

●        However, ranking algorithms alter the distribution of content, biasing our information consumption towards that which confirms what we already believe. The interaction of ‘filter bubbles’ with confirmation bias has been shown to contribute to increased polarisation (Lefebvre, Deroy, and Bahrami 2024).

What role do generative artificial intelligence (AI) and large language models (LLMs) play in the creation and spread of misinformation, disinformation and harmful content?

●        Amplification, an online disinformation strategy, creates impressions of false consensus through the use of non-human bots (accounts that automate content promotion and simulate user behaviour) and trolls (users who misrepresent their identities to promote discord) (Broniatowski et al. 2018).

●        Bots and trolls retweet and modify content from human users, and consequently, well-intentioned posts containing factual information may unintentionally ‘feed the trolls’ (Broniatowski et al. 2018).

●        Without specialised knowledge and tools, it can be difficult to identify bots or trolls. This raises concerns about who is really behind these activities and the failure of social media corporations to accept any responsibility for undermining public health messaging.

Call for evidence: What role did social media algorithms play in the riots that took place in the UK in summer 2024?

●        Social media’s influence on civic engagement and democracy, particularly during recent elections, is marked by both benefits and challenges (Purba, Pearce, et al. 2024).

●        Online ‘echo chambers’ can reinforce confirmation biases, entrench polarised views, and amplify misinformation. The efforts of platforms such as Meta to counter this often lack transparency (Kupferschmidt 2024), fail to deliver meaningful results, and overlook the accountability of entities that produce and propagate misleading information.

●        Identifying the sources of these disinformation tactics requires specialised expertise, raising critical questions about the actors behind these campaigns and how to motivate social media companies to take responsibility for their role in public health messaging (Purba, Pearce, et al. 2024)

Call for evidence: How effective is the UK's regulatory and legislative framework on tackling these issues? What more should be done to combat potentially harmful social media and AI content?

1. The effectivity of regulation and legislation

• Much of the national and international regulatory and legislative framework for tackling technological harms requires direct causal evidence. Only when a very high threshold of evidence is reached is there a concerted effort to build, test and scale interventions to address potential harms (Orben and Matias under Review).
• These high thresholds of evidence are very difficult to reach (Orben and Matias under Review). Providing timely evidence of technological harms is very difficult. Emerging technologies, like social media, evolve faster than research timelines. This can mean that scientific findings become outdated even by the time they are published (Orben 2020).
• Additionally, technology companies are reluctant to share data for research, making it difficult for academic researchers to access the resources needed.

• There are several ways in which this problem could be addressed.
• The current project-based research funding model is reactive: assessments of potential harms are only funded after concerns arise, delaying timely insights into new technological impacts. As an alternative, inspiration should be taken from other areas of consumer safety such as food safety, where industry levies proactively fund long-term consumer safety research efforts. Adapting this funding approach to the technological sphere could mean funding studies on anticipated technological changes, such as large language models or deepfakes, to pre-empt their societal impacts.
• We also need to recognise that the current evidence creation system will not always deliver traditional, high-standard evidence in time. Instead, a form of adverse inference approach could be applied, wherein lower levels or quality of evidence of harm are required before testing interventions. This lowered evidence threshold could be specifically applied to products where companies have not collaborated in providing evidence that their products are safe before releasing them on the market or if concerns about marketed products are raised (Orben and Matias under Review).
• Such a targeted approach is one of multiple methods our team has explored and conceptualised to incentivise social media platforms to prioritise user safety in their design (along with, for example, requirements for inbuilt protection such as high default privacy settings, clear reporting processes, and content access limitations).

2. How to combat potentially harmful social media and AI content

●        Strategies which combat the spread of harmful content should be informed by knowledge of the human psychological dispositions and biases with which design features interact (Turner, Ferguson, et al. 2024; Skeggs and Orben 2024).

●        Two key routes for improvement could involve (1) designing algorithms which are better aligned with known human biases, and (2) increasing transparency by empowering users to select their own ranking algorithms (Brady et al. 2023).

Conclusions

●        Social media companies’ financial models currently rely on maximising user engagement and time spent online.

●        This motive inherently conflicts with efforts to protect users and society from the spread of harmful content, negative physical and mental health impact, and proliferation of prejudice and disinformation in public discourse (Purba, Pearce, et al. 2024).

●        These harms can be analysed within a framework which considers how specific social media design features interact with people’s existing psychological dispositions and biases (Turner, Ferguson, et al. 2024; Brady et al. 2023).

●        Companies are unlikely to take the necessary steps to prevent such harms without either external, independent regulation or a change in business model.

17 December 2024

References

Alonzo, Rea, Junayd Hussain, Saverio Stranges, and Kelly K. Anderson. 2021. ‘Interplay between Social Media Use, Sleep Quality, and Mental Health in Youth: A Systematic Review’. Sleep Medicine Reviews 56 (April):101414. https://doi.org/10.1016/j.smrv.2020.101414.

Brady, William J., Joshua Conrad Jackson, Björn Lindström, and Molly Crockett. 2023. ‘Algorithm-Mediated Social Learning in Online Social Networks’. OSF Preprints. https://doi.org/10.31219/osf.io/yw5ah.

Brady, William J., Killian McLoughlin, Tuan N. Doan, and Molly J. Crockett. 2021. ‘How Social Learning Amplifies Moral Outrage Expression in Online Social Networks’. Science Advances 7 (33): eabe5641. https://doi.org/10.1126/sciadv.abe5641.

Broniatowski, David A., Amelia M. Jamison, SiHua Qi, Lulwah AlKulaib, Tao Chen, Adrian Benton, Sandra C. Quinn, and Mark Dredze. 2018. ‘Weaponized Health Communication: Twitter Bots and Russian Trolls Amplify the Vaccine Debate’. American Journal of Public Health 108 (10): 1378–84. https://doi.org/10.2105/AJPH.2018.304567.

Brown, Sandra A., Matthew McGue, Jennifer Maggs, John Schulenberg, Ralph Hingson, Scott Swartzwelder, Christopher Martin, et al. 2008. ‘A Developmental Perspective on Alcohol and Youths 16 to 20 Years of Age’. Pediatrics 121 Suppl 4 (Suppl 4): S290-310. https://doi.org/10.1542/peds.2007-2243D.

Cookingham, Lisa M., and Ginny L. Ryan. 2015. ‘The Impact of Social Media on the Sexual and Social Wellness of Adolescents’. Journal of Pediatric and Adolescent Gynecology 28 (1): 2–5. https://doi.org/10.1016/j.jpag.2014.03.001.

Kupferschmidt, Kay. 2024. ‘A Study Found Facebook’s Algorithm Didn’t Promote Political Polarization. Critics Have Doubts’. https://www.science.org/content/article/study-found-facebook-algorithm-didnt-promote-political-polarization-critics-doubt.

Lefebvre, Germain, Ophélia Deroy, and Bahador Bahrami. 2024. ‘The Roots of Polarization in the Individual Reward System’. Proceedings of the Royal Society B: Biological Sciences 291 (2017): 20232011. https://doi.org/10.1098/rspb.2023.2011.

McLoughlin, Killian L., William J. Brady, and Molly J. Crockett. 2021. ‘The Role of Moral Outrage in the Spread of Misinformation’. In Technology, Mind, and Behavior. https://doi.org/10.1037/tms0000136.

McLoughlin, Killian L., William J. Brady, Aden Goolsbee, Ben Kaiser, Kate Klonick, and M. J. Crockett. 2024. ‘Misinformation Exploits Outrage to Spread Online’. Science 386 (6725): 991–96. https://doi.org/10.1126/science.adl2829.

Moreno, Megan A., Leslie R. Briner, Amanda Williams, Leslie Walker, and Dimitri A. Christakis. 2009. ‘Real Use or “Real Cool”: Adolescents Speak out about Displayed Alcohol References on Social Networking Websites’. The Journal of Adolescent Health: Official Publication of the Society for Adolescent Medicine 45 (4): 420–22. https://doi.org/10.1016/j.jadohealth.2009.04.015.

Moreno, Megan A., and Rosalind Koff. 2016. ‘11. Media Theories and the Facebook Influence Model’. In The Psychology of Social Networking Vol.1: Personal Experience in Online Communities, 130–42. De Gruyter Open Poland. https://doi.org/10.1515/9783110473780-013.

Ofcom. 2024. ‘Understanding Pathways to Online Violent Content Among Children’.

Orben, Amy. 2020. ‘The Sisyphean Cycle of Technology Panics’. Perspectives on Psychological Science 15 (5): 1143–57. https://doi.org/10.1177/1745691620919372.

Orben, Amy, and J. Nathan Matias. under Review. ‘Fixing the Science of Technology Harms.’

Orben, Amy, Adrian Meier, Tim Dalgleish, and Sarah-Jayne Blakemore. 2024. ‘Mechanisms Linking Social Media Use to Adolescent Mental Health Vulnerability’. Nature Reviews Psychology, May, 1–17. https://doi.org/10.1038/s44159-024-00307-y.

Palminteri, Stefano, and Maël Lebreton. 2022. ‘The Computational Roots of Positivity and Confirmation Biases in Reinforcement Learning’. Trends in Cognitive Sciences 26 (7): 607–21. https://doi.org/10.1016/j.tics.2022.04.005.

Purba, Amrit Kaur, Marion Henderson, Andrew Baxter, S Vittal Katikireddi, and Anna Pearce. 2023. ‘The Relationship between Time Spent on Social Media and Adolescent Alcohol Use: A Longitudinal Analysis of the UK Millennium Cohort Study’. European Journal of Public Health 33 (6): 1043–51. https://doi.org/10.1093/eurpub/ckad163.

Purba, Amrit Kaur, Marion Henderson, Andrew Baxter, Anna Pearce, and S Vittal Katikireddi. 2024. ‘The Relationship Between Time Spent on Social Media and Adolescent Cigarette, E-Cigarette, and Dual Use: A Longitudinal Analysis of the UK Millennium Cohort Study’. Nicotine and Tobacco Research, April, ntae057. https://doi.org/10.1093/ntr/ntae057.

Purba, Amrit Kaur, Anna Pearce, Marion Henderson, Martin McKee, and S Vittal Katikireddi. 2024. ‘Social Media as a Determinant of Health’. The European Journal of Public Health 34 (3): 425–26. https://doi.org/10.1093/eurpub/ckae029.

Purba, Amrit Kaur, Rachel M. Thomson, Paul M. Henery, Anna Pearce, Marion Henderson, and S. Vittal Katikireddi. 2023. ‘Social Media Use and Health Risk Behaviours in Young People: Systematic Review and Meta-Analysis’. BMJ 383 (November):e073552. https://doi.org/10.1136/bmj-2022-073552.

Skeggs, Amira, and Amy Orben. 2024. ‘Social Media Interventions to Improve Wellbeing’. https://doi.org/10.31234/osf.io/u9wqc.

Skinner, B. F. 1953. ‘Some Contributions of an Experimental Analysis of Behavior to Psychology as a Whole’. American Psychologist 8 (2): 69–78. https://doi.org/10.1037/h0054118.

Turner, Georgia, Amanda Ferguson, Tanay Katiyar, Stefano Palminteri, and Amy Orben. 2024. ‘Old Strategies, New Environments: Reinforcement Learning on Social Media’. OSF. https://doi.org/10.31234/osf.io/f5cjv.

Turner, Georgia, Lukas J. Gunschera, Shashanka Subrahmanya, Aadesh Salecha, Johannes C. Eichstaedt, Stefano Palminteri, and Amy Orben. 2024. ‘A Computational Model of Reward Learning and Habits on Social Media’. OSF. https://doi.org/10.31234/osf.io/xe25k.