Designing for Depolarisation with AI
In my Master’s Thesis (2021), I explored how AI drives political polarisation in social media – and how we can use the same technology to reverse it through UI interventions.
5 min read
Political division over information is a paradox of the modern hyperconnected world. AI-driven social media platforms created a unique unprecentend situation where despite being brought so close to each other people increasingly get polarised in the
way they see the reality.
It has to do with the way information is presented in social media today – personalised newsfeeds provide each of us with the picture of the world that we like and engage with the most. On a macro level, different truths and separate worlds have emerged
as a result of this process. This has diminished the common grounds of shared reality and developed hostility in interactions between these different groups of people who hold contradicting beliefs.
In my thesis, I explored the cognitive-behaviuoral aspect of political polarisation in the U.S. between far-left and far-right social media users. Using the digital ethnography as the main research method,I embedded into circles of social media users from both ends of the American political spectrum and compared their online experiences, information processing mechanisms and behaviours.
The goal of this project was to attempt to depolarise social media users with means of design. Results of the digital ethnography study identified several points in the experiences of social media users, where the design intervention could make a change.
As an outcome of this project, I proposed a concept that was designed to motivate respectful discussions and unite people around common understadning of the reality, which removes certain aspects of polarisation from social media environments.
A Design Intervention: UI Concept
Most platform responses are punitive: remove content, suspend accounts. Necessary, but limited. I explored a complementary approach: make constructive behaviour visible and rewarding.
Speaker Badge (the concept)
While you type: An AI layer runs semantic checks (is the claim corroborated by reliable sources?) and sentiment/toxicity checks (is the tone abusive). You see private guidance before you post.
Earnable signal: If your posts are consistently credible and respectful (plus some community support), you earn a Speaker Badge – a lightweight cue that helps your posts stand out. It’s revocable if you start pushing misinformation or harassment. Loss aversion does the nudging.
Self-reflection: A private dashboard lets you review past posts and see why the score changed (reliability, tone, support), so the rules aren’t a black box.
This isn’t a “social credit score.” It’s scoped to conversation quality, with transparent criteria and no off-platform consequences. The point is to create a second path to visibility: be accurate, be civil, be heard.
Feedback
I prototyped the concept as a landing page and shared it on Gab to collect reactions. Some users rejected it as “speech control.” One target-profile user, however, opted in and gave positive feedback, which suggested value as a prevention tool for people drifting toward extremes, not a cure for those already entrenched.
Limits (+ Why I Still Think It’s Worth Considering)
This is not a silver bullet; it’s a small intervention inside a vast socio-technical problem. But it shifts incentives at exactly the moment they’re most malleable, before publishing, while keeping enforcement tools for severe cases. In practice: lower toxicity per thread, equal or higher conversation length, more link-outs to sources, and high adoption of pre-publish suggestions are success signals to track.
Reflections
If engagement-first design can amplify polarisation, design can also nudge toward accountability and respect. Speaker Badge is a modest attempt to do that: coach privately, recognise publicly, and let healthier discourse become visible again.
Like What You See? See More in Work & Blog.
Contacts
v.dzhekanovich@gmail.com