The AI Journal

In November 2016 ninety-nine people who had never met were handed a question Irish politicians had spent decades declining to touch: what, if anything, should replace the constitutional clause banning abortion in nearly all circumstances. They had been selected by an opinion-polling company, stratified by age, sex, social class and region, with no interview, no application, and no qualification beyond being on the electoral register and willing to give up a run of Saturdays. Over the following months they heard lawyers, doctors, advocacy groups on every side, and each other, and in April 2017 they voted, by sixty-four to thirty-six, that abortion should be available with no restriction as to reason up to a gestational limit, with wider grounds beyond it. Eighty-seven per cent of them had already rejected keeping the existing near-total ban. A year later the Oireachtas put a version of their recommendation to the country, and the country repealed the Eighth Amendment by a margin of two to one.

The standard complaint about bodies like this one is that none of it actually did anything. A citizens’ assembly has no power to legislate, no power to call a referendum, no power to do anything but recommend; its members are chosen by lot rather than by election or expertise, which critics treat less as a design feature than as a confession that nobody with a mandate wanted the job. Kimmo Grönlund, André Bächtiger and Maija Setälä’s 2014 survey of these bodies, the first full-length academic stocktaking of the form, catalogues this complaint directly, cautioning that most mini-publics exert only limited influence on the policy they were convened to inform, and that their proper role is to complement elected institutions rather than to substitute for them. On that reading, a citizens’ assembly is theatre with unusually good production values: real deliberation, real strangers, and a report that a minister can read, praise, and shelve.

The Irish case is the one that breaks that pattern, and it broke it twice in succession. The 2016–18 Assembly was not Ireland’s first attempt at this; it followed the 2012–14 Convention on the Constitution, a body of sixty-six randomly selected citizens sitting alongside thirty-three politicians, which had taken up marriage equality among its topics and fed directly into the 2015 referendum that legalised it. David Farrell, Jane Suiter and Clodagh Harris’s 2019 account of the Assembly describes it explicitly as the second step in what they call a systematizing of constitutional deliberation in Irish politics, with the Convention and the Assembly each anchored to a referendum that actually happened. The Assembly went further than the Convention in one respect that mattered: no politicians sat on it at all, so the recommendation that reached the Oireachtas was unmixed with the preferences of the people who would have to act on it. What came back from government was not a simple repeal but a repeal-and-replace, handing the power to legislate on terminations to parliament rather than writing a number into the constitution, which is exactly the shape of the solution the Assembly itself had preferred to a bare removal of the clause.

That outcome answers the charge that nothing happens to these recommendations, but it does not by itself answer the harder half of the dismissal, which is about what went on on the Saturdays. A body can be followed by a policy change for reasons that have nothing to do with its own deliberation, government officials picking up whatever conclusion was politically convenient and crediting it to the citizens afterward. What rules that out, in the Irish case, is that the same research team that studied the Assembly’s predecessor had already built the instrument to check it: panel surveys of each member’s own stated opinions, taken before and after a weekend of deliberation, member by member, rather than a single snapshot of the room at the end. Jane Suiter, David Farrell and Eoin O’Malley’s 2016 analysis of the Convention’s members using this design found that opinions did move, that the movement was not simply convergence on whoever argued loudest, and that it was groups exposed to real disagreement among their own members, rather than groups that arrived already agreeing, that produced the more durable shifts. That is a specific and falsifiable claim about a mechanism, confirmed on the same population that would go on to supply the model for the Assembly two years later, and it is a different kind of evidence from a report a minister can read and shelve. James Fishkin’s decade of deliberative polling experiments, run across a dozen countries from Texas to China, give the same mechanism a wider base: controlled information plus structured disagreement, in randomly assembled groups, produces attitude change that survives the room, not merely applause inside it.

None of this requires treating the Irish panels as a scale model of the Irish public, and here the dismissal finds its real target. Paolo Spada and Tiago Peixoto’s 2025 analysis of these bodies is blunt about what random selection does and does not buy. Assemblies this size run on sampling frames built from population registers or household lists that are themselves imperfect, stratify on only a handful of variables such as age, sex and region, and recruit from among the people willing to accept an invitation to give up several Saturdays, a group that is not obviously a cross-section of anyone. Response rates to these invitations are reported in the low single digits to perhaps seven per cent in comparable designs, which means the room that eventually sits down is drawn from a self-selected slice of an already-filtered list, twice removed from the electorate it is said to stand in for. Spada and Peixoto’s conclusion is not that these bodies are worthless but that the legitimacy claim resting on statistical representativeness, the idea that ninety-nine strangers can stand in for several million citizens because a pollster drew them correctly, does not survive the scrutiny its own defenders have invited.

The claim worth defending survives that concession because it was never the claim Spada and Peixoto are attacking. What Ireland’s Assembly demonstrated was not that its hundred members were a faithful miniature of the electorate; on the evidence above, nobody should believe that of any body this size. What it demonstrated, with panel data to show it, is that a group of people given time, information and real disagreement will change their minds in a documented and lasting way, and that a parliament can take what that group concludes and act on it without first requiring that the group be a mirror of the nation. Hélène Landemore’s 2020 case for what she calls open democracy makes exactly this pivot: she grounds the authority of randomly selected bodies not in a statistical resemblance to the public but in genuine inclusion and openness to influence, which is the same alternative standard Spada and Peixoto themselves propose once they have dismantled the representativeness claim. A referendum campaign cannot produce what the panel surveys recorded, because a campaign has no mechanism for finding out whether anyone’s mind actually changed or only which side shouted longer. The hundred were not a sample of Ireland. They were, on the only evidence anyone bothered to collect, a hundred people whose opinions moved under argument and stayed moved, which a ballot box has never once been built to show.