Most debates about democracy get stuck between two positions: defend the institutions we have, or declare them broken. This paper takes a third route. It treats democracy as something that was designed, and so can be redesigned.
Written by TIAL co-founder Sir Geoff Mulgan and Dr Alex Fischer of the Australian Resilient Democracy Network, and published by Demos in collaboration with TIAL and ARDN, Work in Progress argues that the problem is not the idea of democracy but how it is put into practice. Most of its methods took shape in the 19th century, and they have absorbed little of the digital change that has reshaped almost everything else in public life.
Democracy as an assembly of parts
The paper’s main move is to stop treating democracy as a single thing. It describes six connected functions instead: authorisation (elections, parties, constitutions), decision-making (parliaments, budgets), interpretation and constraint (courts, regulators, a free press), implementation (public services), knowledge institutions, and participation and feedback.
Like a car, democracy can be upgraded one part at a time, gaining new capacity to sense and respond while staying recognisable. This changes the question from “why is democracy failing?” to something more useful: which functions are working, which are breaking down, and where is adaptation already happening that no one has yet named?
Under all six functions sit two conditions that are usually taken for granted. The first is relational health, the quality of connection between citizens and institutions. The second is cognitive infrastructure, the shared capacity to hear many voices, combine them at scale, and show how citizen input shapes decisions.
More than 50 innovations, function by function
The paper maps more than 50 democratic innovations from over 25 countries onto this framework. Examples include Taiwan’s use of Polis to map consensus, Brazil’s Ulysses system for making mass public input usable in Congress, Ireland’s Citizens’ Assembly, Wales’s Future Generations Commissioner, and Japan’s Team Mirai, which used an AI interviewer to build its party manifesto. The argument is not that any one of these is the answer. Democracy’s strength comes from combining them.
AI as cognitive infrastructure
The authors reject both techno-optimism and despair. AI-driven platforms have spread misinformation and deepened polarisation. Yet the same tools can help groups of any size reason together: bringing in evidence and institutional memory, laying out trade-offs, combining large volumes of input, and showing where agreement is possible. The paper notes that only a tiny share of public AI funding goes to research and development for democracy.
From diagnosis to design
For TIAL, the paper’s conclusion matters as much as its cases. It calls for fewer autopsies and more experiments, and for roadmaps for democratic renewal as detailed as the ones countries now write for net zero. This is the same shift we argued for in Mobilising the academic study of organisations: the knowledge exists, but it is rarely turned into design. And like Stack thinking, the paper treats reform as reworking an architecture layer by layer, not tearing it down.
This guest paper was published by Demos on 8 October 2026, in collaboration with TIAL and the Australian Resilient Democracy Network. It grew out of Geoff Mulgan’s 2025–26 ARDN International Fellowship. Recommended citation: Mulgan, G. and Fischer, A. (2026). Work in Progress: How to upgrade democracy in the age of AI. Demos, with TIAL and ARDN, Discussion Paper 21.
