Don’t 'Research Fast and Break Things': On the Ethics of Computational Social Science

No recognised ethical standards, no broad consent: navigating the quandary in computational social science research

“…Ethical dilemmas also arise when participants consent to anonymised data use, yet re-identification becomes possible through new datasets (Leslie, 2022). This situation prompts consideration of the validity of consent and anonymity in CSS research, further bringing to light the ethical implications of individuals consenting to the use of their anonymised data, even when they are aware that re-identification may be possible in the future.…”

“…Moreover, whereas CSS questions also focus on social science topics, data science research questions have broader scopes. Importantly, traditional research ethics principles from social and biomedical fields may not fully apply to data science due to distinct privacy issues and the nature of risk (Ada Lovelace Institute, 2022; Leslie, 2022; The Turing Way Community, 2022). Unlike social and biomedical research, risks in data science may emerge over time during the research process, thereby requiring continuous ethics reviews (Ferretti et al, 2021).…”

“…The lack of recognised research ethics in CSS has led to proposals for a ‘roadmap’ to develop CSS-specific ethics, drawing from Responsible Research and Innovation (RRI) principles (Leslie, 2022). This roadmap suggests procedural tools for the CSS research community, emphasising context, impact anticipation, stakeholder analysis, normative criteria for impact assessment, reflection on purposes, positionality, power, inclusive engagement, transparency and responsibility.…”

Algorithmic accountability on social media platforms in the context of alcohol‐related health behavior change

“…Although algorithms used by SM platforms to determine what content is made visible to users remain opaque, these algorithms are designed to enhance user attention and prolong platform engagement, relying on past user activity as an indicator for the type of content likely to capture users' attention [11]. Although the effects of engagement with pro-alcohol content on SM companies' algorithmsand its downstream effects on likelihood of being exposed to proalcohol posts-are unknown, research has shown that digital algorithms impact behavior [12,13]. Therefore, it is plausible the effects of engaging with alcohol content on subsequent drinking behavior may be mediated, in part, by algorithm-driven exposure to pro-alcohol content.…”