In May 2018, DSP Executive Director Susan Benesch spoke on a panel at RightsCon, entitled Data Driven Decency: New, Collaborative Experiments to Diminish Online Hate and Harassment. Panelists and participants discussed several such experiments, including best practices, lessons learned, and lingering challenges. Below are some highlights from the discussion, including links to some of the projects discussed at Data Driven Decency (including Susan Benesch’s) and ideas for future research into decreasing online abuse.
Ongoing projects to diminish online abuse
- Susan Benesch and J. Nathan Matias’ experiment with Twitter, to explore whether descriptive and/or injunctive norm messages improve user behavior
- CivilServant – working directly with online communities to conduct research, all approved by university ethics boards
- The Redirect Method – using Google Adwords to send anti-extremist content to users whose search terms relate to extremism
- Kevin Munger’s work – experimenting with bot accounts identified with different racial and political groups, to reduce the use of racist slurs on Twitter
Considerations for future research
Though there are plenty of ongoing efforts to study hateful and abusive behavior online (and how to diminish it), Data Driven Decency participants noted lingering questions and challenges for researchers:
- When companies themselves conduct studies, encourage them to share at least some data with at least some researchers – instead of simply reporting in the aggregate.
- Researchers should encourage more internet platforms to participate in studies and interventions. Twitter and Reddit are popular among researchers because they make data available; but if researchers only run experiments on these platforms, any solutions they create may only be useful on Twitter and Reddit, and not elsewhere.
- Communities of researchers should explore developing ethical guidelines, especially for how to obtain participants’ consent while respecting their dignity and autonomy. Importantly, how can researchers obtain informed consent without biasing sample sets? J. Nathan Matias outlines some ideas in this Medium piece. Jonathan Zong’s debriefing system may be an example – obtaining informed consent at the tail end of a study, providing the opportunity to opt-out.
- When publishing research results, researchers and companies must consider their communications strategy. Companies are rightly concerned that when research is publicized, the headlines will focus on toxicity on their platforms, rather than the efforts to diminish it.
- When considering their involvement in studies, companies have both legal concerns and compliance concerns. Does handing over data inherently subject them to liability? Or will sharing data reveal things which could subject them to liability?
- How do researchers avoid and mitigate unintended consequences?
- Not only must researchers study hateful content – they need to study the producers of such content as well.