The Optimist's Case for Disagreement

AI once optimised public discourse for engagement; and engagement rewarded outrage, fracturing the public square. This talk shows how a different design intent, seen in tools like Pol.is and DeepMind's Habermas Machine, can help societies find genuine consensus at scale rather than manufactured agreement. Joannes Vandermeulen argues that while AI can help us listen better, deciding what that listening is for (and which disagreements must stay open) remains a human responsibility.

This short talk makes the case that AI can finally help societies listen to themselves at scale, while the harder question of what that listening is for remains for humans to answer.
For a decade, the main AI layer between people was tuned for engagement, and engagement rewarded outrage. The result is the fractured public square we now take for granted: a technology that connected everyone while setting them against each other. That outcome came from a particular design intent; and a different intent is already producing different results. Pol.is looks for the statements that opposing groups quietly agree on, and has helped move real legislation in Taiwan. DeepMind's Habermas Machine drafts group positions that participants judge fairer and clearer than the ones human mediators write.
But there is a warning here too: a machine tuned to manufacture agreement can quietly wear away the disagreements a democracy depends on. Consensus that is engineered is not the same as consensus that is discovered, and a system built to produce the first can erode the second without anyone noticing. Joannes Vandermeulen argues that AI can finally let us listen at scale while the harder questions of what that listening is for, and which divisions must stay open, remain ours to answer.