1. Devcon in Mumbai, India, is done and dusted for 2026. It was a great platform, bringing together like-minded developers from around the globe. Interestingly, the U.S. midterms hardly featured in our discussions. We focused on technology and how it can help heal divided societies.

    From Taiwan to the U.S. and everywhere else in between, many are worrying about AI interfering in elections: fake accounts and mass-produced social media pages that seal us ever more tightly inside our echo chambers. But, I see another possibility — inviting AI into an argument as an even-handed referee.

    In May, Meta’s social media platform Threads began trialling a feature in Taiwan that lets users ‘call in’ an AI. When a post provokes polarisation, many choose to tag @meta.ai rather than trade insults. Drawing on real-time searches and citing sources, the AI examines both sides’ claims and replies in public, directly beneath the post.

    But surely AI is all too good at telling us what we want to hear? How can it build consensus among opposing camps?

    In one-on-one conversations, that concern is justified. In March, a Stanford team published a study in Science testing 11 leading models. When users sought advice or assistance with interpersonal conflicts, the models went along with their views and affirmed stated behaviour nearly 50 percent more often, on average, than humans did. The experiments also found that a single instance of an AI agreeing with participants made them more convinced they were right and less willing to repair the relationship.

    If we simply consult AI in private, then share a screenshot saying, ‘Look, even the AI agrees with me’, we may only be adding another brick in the wall of our echo chambers.

    In a public thread, though, everyone can check the answer together. Even if the person asking tries to steer the discussion, others from across the divide would ask follow-up questions on the spot, press the AI system to verify figures and address errors. People who come across the post later can then join a public conversation about ‘What is actually going on here?’ rather than another polarised argument over ‘Whose side are you on?’

    Bringing the discussion back to specific facts can allow people across party lines to address everyday concerns. In the U.S. midterms, for example, voters on both sides worried that data centres might drive up household electricity bills. The more voters get into the habit of calling in AI referees to assess the figures and cross-check claims, the more candidates will have to offer proposals and evidence that stand up to scrutiny. Slogans alone will no longer suffice.

    On the Taiwan front, city and county elections are an ideal place to practise, since they are closest to the issues impacting daily life. If a post incites enragement and the source cannot be found, it is important to resist the urge to share. Take a deep breath, pause and tag AI for support in checking the claims and establishing facts worth discussing. Once emotions no longer dominate the thread, those who encounter the post later need not be provoked all over again. They can put their energy into discussing policy and finding common ground.

    We may vote for different candidates and parties, but after the election we can check the facts and hold those elected to account — together. From pavement works and childcare budgets to progress on social housing, we can combine AI with open government data to check official progress reports against what is actually happening and call on elected representatives to explain the shortfalls.

    Used well, AI can be our even-handed referee, helping us score the societal goal of making democracy a daily practice of dialogue and scrutiny, not a single day at the polls.

  2. (Interview and Compilation by Yu-Tang You. License: CC BY 4.0)