Skip to content
Citizen Infrastructure

Conversation networks

The media theorist Marshall McLuhan argued that each era’s dominant medium quietly shapes its politics. Radio — one voice reaching millions, with no way to answer back — leaned authoritarian. Television, more ambient and shared, suited the loose consensus of post-war democracies. Audrey Tang uses this lens to ask a sharper question: what shape of politics does our media push us toward, and can we choose a better one?

Broadcasting vs. conversation networks

Section titled “Broadcasting vs. conversation networks”

Today’s feeds are broadcasting turbo-charged. One voice can still reach millions, but now algorithmic targeting gives every person a different, personalised stream. Since around 2015, when platforms switched to the “for you” feed, the shared experience that held a town or a living room together got stripped away — and the attention sold to the highest bidder. The result is what researchers call engagement through enragement: the most extreme “dunk” travels furthest, because outrage is sticky.

A conversation network is the alternative Tang points to — and it’s older than it sounds. It works more like a telephone network: small groups talking, where attention is roughly symmetric and you can feel in real time whether your point landed (“oh, that’s a good point”) through ordinary human cues, not like-counts. Crucially, in this setting a single toxic dunk can’t go viral. What spreads instead is the rare remark that uncovers an uncommon ground — the thing several people didn’t expect to agree on.

AI in the background, not the foreground

Section titled “AI in the background, not the foreground”

Most people meet AI as a dyadic experience: one human chatting to one chatbot that sits between them and the world. Conversation networks use AI the opposite way — ambiently, as an interpreter and summariser that helps people focus on each other better, the way noise-cancellation works on both ends of a call. It can translate across languages, surface a quiet voice, and produce a real-time summary that reveals blind spots, while the humans stay the point of the conversation. This is the same “assistive, not automating” principle behind civic AI and civic listening.

Because the conversations are recorded and summarised, a thirty-second clip that resonated in one group can be carried into another as a prompt — so insight scales the way a rumour does, group to group, without a central megaphone. This is the subject of a 2025 paper, Conversation Networks, by Deb Roy, Lawrence Lessig, and Audrey Tang.

Sources

Section titled “Sources”