AI can help humans find common ground in democratic deliberation

What the idea means

The findings come from UK participants, including a virtual assembly sample intended to reflect the country’s demographics. The experiments considered several aspects of the statements and discussion process.

Why it matters

Knowing who took part and what was examined helps readers judge the scope of the findings; the supplied text does not establish whether they apply to other countries or groups.

Question this helps explore

Who took part, and what did the study examine?

Evidence and limits

The article reports 5,734 UK participants and a virtual assembly involving a demographically representative UK sample. It lists several questions the experiments examined, including view changes and possible bias in group statements.

Addresses: The participants, settings, and research questions

AI can help humans find common ground in democratic deliberation · passages 1, 2

What the idea means

In the reported comparisons, participants favored AI statements and rated them as more unbiased. External judges also gave them stronger ratings on the listed qualities.

Why it matters

These ratings suggest the statements were well received in this study, but do not prove they were accurate or suitable for every use.

Question this helps explore

How did participants and outside judges rate the statements?

Evidence and limits

The article reports participant preferences and ratings, as well as separate ratings by external judges on several stated qualities.

Addresses: Ratings of AI and human statements

AI can help humans find common ground in democratic deliberation · passages 2

What the idea means

Participants’ reported stances became more alike after AI-assisted deliberation. The comparison group that exchanged views directly did not show the same reported convergence.

Why it matters

The comparison suggests that simply sharing opinions may not have the same effect as using an AI helper to develop a group statement.

Question this helps explore

Did AI-assisted discussion change how divided participants were?

Evidence and limits

The article reports that division decreased after AI-mediated discussion, while direct, unmediated exchanges did not produce the same result.

Addresses: Changes in participants’ reported views

AI can help humans find common ground in democratic deliberation · passages 2, 4

What the idea means

The movement toward the majority view did not mean minority critiques disappeared. The text analysis found that successful statements included dissent while still reflecting the majority position.

Why it matters

This matters because a shared statement may show both areas of agreement and disagreement, though the report does not say how much weight each view received.

Question this helps explore

How did the statements reflect majority and minority views?

Evidence and limits

The article reports increased support for the majority position, the inclusion of minority critiques, and a text-embedding analysis finding that successful statements incorporated dissent while respecting the majority position.

Addresses: The content of revised group statements

AI can help humans find common ground in democratic deliberation · passages 2

What the idea means

Groups in the assembly often changed their views in a similar direction. The authors say AI bias did not account for the pattern and interpret it as shared views emerging through discussion.

Why it matters

This extends the reported results to a virtual assembly, but the interpretation does not prove that deliberation alone caused all view changes.

Question this helps explore

What happened in the virtual assembly, and how did the authors interpret it?

Evidence and limits

The article reports similar shifts in groups’ views and says the shifts were not attributable to bias in the AI. The authors interpret the pattern as the emergence of shared perspectives through deliberation.

Addresses: Assembly results and the authors’ explanation

AI can help humans find common ground in democratic deliberation · passages 2

What the idea means

The authors argue that the method could help groups reach agreement and act together in several settings. The article also describes the approach as efficient, fair, scalable, and strong on key comparisons with human mediators.

Why it matters

These are claims about potential; the supplied passages do not establish that the method works or is safe in every proposed setting.

Question this helps explore

What wider benefits and uses do the authors propose?

Evidence and limits

The conclusion describes these benefits and possible uses, while presenting the approach as potential for collective decision-making rather than reporting tests in every listed setting.

Addresses: The authors’ assessment and suggested applications

AI can help humans find common ground in democratic deliberation · passages 3

What the idea means

The authors argue that agreement is needed for people to act together, while recognizing the value of different legitimate views.

Why it matters

This premise motivates the study, but should not be mistaken for a result or proof that disagreement is harmful.

Question this helps explore

Why do the authors say agreement matters for collective action?

Evidence and limits

The passage states this as the authors’ premise; it does not independently establish that agreement is always necessary or desirable.

Addresses: The article’s rationale

AI can help humans find common ground in democratic deliberation · passages 5

What the idea means

The authors describe deliberation as an exchange of views that can happen in person or online, with varying degrees of structure. They identify citizens’ assemblies as one formal way to hold such discussions.

Why it matters

This gives context for the proposed tool, but does not show that assemblies or other discussions always improve decisions.

Question this helps explore

How do the authors describe deliberation and its role in democracy?

Evidence and limits

The passage describes both the broader discussion format and citizens’ assemblies as a formal version of it; this is background, not a study result.

Addresses: The article’s background on deliberation

AI can help humans find common ground in democratic deliberation · passages 5

What the idea means

The passage names three countries that have used citizens’ assemblies for important national policy decisions.

Why it matters

The examples show that assemblies have been used, not that they produce better decisions.

Question this helps explore

Where have citizens’ assemblies been used for national policy?

Evidence and limits

These are examples reported by the authors, not evidence that such assemblies always work well.

Addresses: Examples in the article’s background

AI can help humans find common ground in democratic deliberation · passages 5

What the idea means

The authors describe assemblies as costly and time-consuming, with discussion limited to relatively small groups.

Why it matters

These constraints motivate interest in other methods, but do not show that AI can overcome them.

Question this helps explore

What practical limits can make assemblies difficult to use?

Evidence and limits

The authors give these limits as reasons to explore other ways to support discussion.

Addresses: The article’s account of existing deliberation

AI can help humans find common ground in democratic deliberation · passages 5

What the idea means

Discussion may become unequal or reinforce existing views through social pressure or strategic behavior.

Why it matters

A mediator should be assessed for whose views it includes, not only for whether it finds agreement.

Question this helps explore

What problems can arise during group discussion?

Evidence and limits

These are risks the authors describe, not claims that every assembly experiences them.

Addresses: Possible weaknesses in deliberation

AI can help humans find common ground in democratic deliberation · passages 5

What the idea means

The authors say earlier studies do not show one consistent effect on views, decisions, or later participation.

Why it matters

This supports testing specific outcomes rather than assuming deliberation always helps.

Question this helps explore

What does earlier research say about citizens’ assemblies?

Evidence and limits

This summarizes prior studies, not the article’s own experiments.

Addresses: The article’s summary of prior research

AI can help humans find common ground in democratic deliberation · passages 5

What the idea means

The authors cite tools for summarizing or presenting views, changing the tone of partisan wording, and addressing differences in who contributes.

Why it matters

These examples motivate further testing but do not show that AI can find agreement or should make public decisions.

Question this helps explore

What earlier AI work motivated this study?

Evidence and limits

These are reports of earlier work, not findings from the current study.

Addresses: Prior examples reported by the article

AI can help humans find common ground in democratic deliberation · passages 6

What the idea means

The authors say earlier work built tools but had not systematically assessed whether they work, change beliefs, or shape discussion.

Why it matters

A tool’s ability to generate text does not by itself show that it improves deliberation.

Question this helps explore

What research gap did the authors identify?

Evidence and limits

The authors distinguish existing prototypes from systematic evaluation; they do not say that no earlier tools existed.

Addresses: The stated gap in earlier research

AI can help humans find common ground in democratic deliberation · passages 6

What the idea means

The approach gives the AI a chance to receive each person’s views separately before it drafts a group statement.

Why it matters

The system supports the discussion process; it is not described as a human participant or decision-maker.

Question this helps explore

What kind of mediator did the researchers develop?

Evidence and limits

The AI is described as helping write a group statement, not choosing policy or governing the group.

Addresses: The proposed mediation approach

AI can help humans find common ground in democratic deliberation · passages 6, 7

What the idea means

The authors connect the system to Habermas’s idea that rational discussion under ideal conditions can lead to agreement.

Why it matters

The idea explains the system’s aim, but does not show that the aim can be achieved in practice.

Question this helps explore

What idea inspired the system’s name and approach?

Evidence and limits

This describes the inspiration, not evidence that real-world discussion meets those ideal conditions.

Addresses: The authors’ stated intellectual inspiration

AI can help humans find common ground in democratic deliberation · passages 6

What the idea means

The system uses participants’ opinions to draft a statement intended to be endorsed by as many group members as possible.

Why it matters

Aiming for endorsement does not by itself establish fairness, agreement, or decision quality.

Question this helps explore

What was the system designed to produce?

Evidence and limits

The system generates statements for a group to consider; the passage does not describe it making a binding decision.

Addresses: The system’s stated purpose

AI can help humans find common ground in democratic deliberation · passages 6

What the idea means

One model proposes group statements. The other estimates each person’s preferences among them. Both were fine-tuned versions of Chinchilla, a pretrained language model; supervised fine-tuning is specified for the statement generator.

Why it matters

The design gives the models separate tasks, but the passage does not establish how well either performs.

Question this helps explore

How were the system’s two models set up?

Evidence and limits

This describes the method, not the accuracy of the preference predictions or the quality of the generated statements.

Addresses: The model components and training

AI can help humans find common ground in democratic deliberation · passages 7

What the idea means

The system uses a rule for combining preferences that has a property intended to prevent near-duplicate statements from dividing votes.

Why it matters

The chosen statement depends on how preferences are combined, not only on what the AI generates.

Question this helps explore

How does the system select among candidate statements?

Evidence and limits

This is a design choice intended to address vote splitting; its effects are not evaluated in the supplied text.

Addresses: The selection rule used by the system

AI can help humans find common ground in democratic deliberation · passages 7

What the idea means

After an initial statement, the AI can receive written feedback and repeat its generation and selection process.

Why it matters

Participants get a chance to respond, but the passage does not show how well the system handles disagreement.

Question this helps explore

Can participants influence revisions to the AI’s proposal?

Evidence and limits

This is a system feature; it does not establish that every critique appears in the final text.

Addresses: The iterative revision feature

AI can help humans find common ground in democratic deliberation · passages 7

What the idea means

The study treats common ground as shared information a group can use when making later proposals or decisions.

Why it matters

This is more specific than asking whether participants agree on everything.

Question this helps explore

What did the authors mean by common ground?

Evidence and limits

This is a hypothesis and definition, not a reported result.

Addresses: The main hypothesis and its working definition

AI can help humans find common ground in democratic deliberation · passages 7

What the idea means

The article reports a combined total of 5,734 participants across structured experiments.

Why it matters

The count alone does not establish representativeness or the strength of the evidence.

Question this helps explore

What kind of study did the researchers conduct?

Evidence and limits

The total describes the experiments’ scope, not proof that findings apply to everyone.

Addresses: The overall method and participant count

AI can help humans find common ground in democratic deliberation · passages 7

What the idea means

The study included questions about issues such as voting age and public health care.

Why it matters

The supplied core text does not establish how well the system would work on other topics.

Question this helps explore

What topics did participants discuss?

Evidence and limits

These examples show the study’s policy setting, not the views participants held.

Addresses: The deliberation topics

AI can help humans find common ground in democratic deliberation · passages 8

What the idea means

The researchers used a language model to generate questions and then reviewed them with the aim of reducing offensive content.

Why it matters

Human review was part of preparation, but the text does not establish that every question was free of risk.

Question this helps explore

How did researchers prepare discussion questions?

Evidence and limits

This describes how questions were prepared; the passage does not say how final questions were chosen for each group.

Addresses: Question generation and screening

AI can help humans find common ground in democratic deliberation · passages 8

What the idea means

Participants met in small groups, usually of five, and worked through three questions in about an hour.

Why it matters

The format limits how directly the findings apply to longer, larger, or differently organized discussions.

Question this helps explore

What did a typical discussion session involve?

Evidence and limits

The passage describes the usual setup; it does not say every group had exactly five participants or the same schedule.

Addresses: The group discussion format

AI can help humans find common ground in democratic deliberation · passages 8

What the idea means

The process gathered private opinions, used participant rankings to select a group statement, and let participants critique and revise it. Participants also rated their positions before and after all deliberation rounds, in addition to answering a brief per-round survey about possible effects on their views.

Why it matters

These measures describe how the study tested a structured process for producing and revising group statements. The procedure alone does not show that participants’ views changed.

Question this helps explore

How did the study’s deliberation process work and measure participants’ views?

Evidence and limits

This describes the steps in each round and the separate before-and-after measure of participants’ positions across the full set of rounds. Participants knew an algorithm generated the group statements.

Addresses: Study procedure and measurement

AI can help humans find common ground in democratic deliberation · passages 9

What the idea means

The system estimated each person’s likely ranking of candidate statements, then combined those estimates to choose a statement for the group.

Why it matters

This is the mechanism behind the AI helper’s selection process; it does not mean the AI itself decided what the group should believe.

Question this helps explore

How did the AI helper select a group statement?

Evidence and limits

The method description explains how candidates were generated and ranked. It describes the study’s system, not proof that this selection method is best for public decisions.

Addresses: Statement selection method

AI can help humans find common ground in democratic deliberation · passages 11

What the idea means

Participants chose the AI helper’s statement more often and gave it stronger endorsement and quality ratings than the human mediator’s statement.

Why it matters

The result suggests that writing a statement the group prefers was a challenging task for the human mediators in this study. It does not show that AI should make decisions for groups.

Question this helps explore

How did participants rate AI-written statements compared with human-written ones?

Evidence and limits

The reported comparison involved 439 participants and trained, incentivized human mediators. The human mediators did not have a critique phase, so the comparison was not between identical processes.

Addresses: Participant preferences and ratings

AI can help humans find common ground in democratic deliberation · passages 10

What the idea means

People who were not part of the deliberating groups rated AI statements more positively on several measures, including clarity and capturing the majority view. They did not rate the two types differently on three other measures.

Why it matters

The ratings suggest several perceived strengths, but do not establish an overall advantage on every aspect of fairness or representation.

Question this helps explore

How did outside raters judge the statements?

Evidence and limits

The same raters found no statistical difference on whether the statements fairly summarized opinions, included minority views, or were polarizing.

Addresses: External ratings of statement quality

AI can help humans find common ground in democratic deliberation · passages 12

What the idea means

The additional data came from 1,692 participants in 349 groups completing 1,047 deliberation rounds. The researchers used it to examine parts of the model and test discussion topics outside the training set.

Why it matters

These checks address whether particular system components matter and whether results extend beyond familiar topics, but the supplied passage does not state the findings.

Question this helps explore

How did the researchers test the system beyond the initial comparison?

Evidence and limits

The supplied passage describes the goals of these analyses, but does not provide their results.

Addresses: Additional cohorts and system checks

AI can help humans find common ground in democratic deliberation · passages 12

What the idea means

In the three main study groups, 64% to 68% of paired choices favored revised statements. In a separate comparison involving 245 participants in 50 groups, participants ranked statements from an AI that had seen their critiques more favorably than statements from an AI that had not. The difference remained after accounting for statement length.

Why it matters

The separate comparison makes it less likely that preference was only due to seeing a statement later or spending more effort. It does not show that preferred statements led to better decisions outside the study.

Question this helps explore

Did adding participants’ critiques make group statements more acceptable?

Evidence and limits

The separate comparison tested whether people favored revised statements simply because they saw them later or had spent extra effort. Preference remained after accounting for statement length, which supports the authors’ view that incorporating critiques helped. These tests measured preferences in the study, not whether the process works in every setting.

AI can help humans find common ground in democratic deliberation · passages 13

What the idea means

The study reports average increases of about eight percentage points in group agreement. In the first group, the share of groups with every member on the same side rose from 22.8% to 38.6%; similar shifts occurred in the other two groups.

Why it matters

The result suggests that groups became less divided during the study’s AI-mediated discussions. Agreement alone does not show whether people were well informed or felt free to disagree.

Question this helps explore

Did group members become more aligned after AI-mediated discussion?

Evidence and limits

The researchers measured agreement before and after discussion as the difference between the shares supporting each side. The reported average increase was about eight percentage points across the three groups. This measure shows greater alignment, not that everyone changed their mind or that the final view was correct.

AI can help humans find common ground in democratic deliberation · passages 14, 15

What the idea means

In the first main study group, stronger endorsement of the final statement was linked to greater influence of other members’ initial positions on participants’ later positions; similar patterns appeared in the other groups. Participants’ own initial positions also significantly predicted their later positions.

Why it matters

These findings describe how people’s earlier views and support for the shared statement related to their later views. They do not establish that either factor caused a change.

Question this helps explore

Was acceptance of the final statement related to shifts toward other members’ views?

Evidence and limits

The authors interpret the link between statement endorsement and influence from others’ starting views as suggesting that a well-accepted shared statement may help people shift their perspectives. Participants’ own starting positions also had a significant relationship with their later positions. These results show associations, not that statement endorsement or starting positions caused later views.

AI can help humans find common ground in democratic deliberation · passages 15

What the idea means

Participants reported moving closer to the group opinion in 30.1% of rounds, no change in 60.3%, and movement in some other way in 4.6%. In 66.6% of rounds, they chose the final group opinion as the best expression of their current view; in 33.3%, they chose their original opinion.

Why it matters

The reports show how participants described their views during the study. They do not show that every participant changed their mind or explain why anyone’s view shifted.

Question this helps explore

How did participants describe their own views during the discussions?

Evidence and limits

These reports give participants’ accounts of how their views related to the group’s view. They are not the same as a direct measure of why views changed, and most rounds were reported as involving no change.

AI can help humans find common ground in democratic deliberation · passages 15

What the idea means

In an additional group of 372 participants, agreement did not change when people rated one another’s opinions without AI mediation. Across groups, agreement changed more with AI-created group statements than with exposure to individual opinions.

Why it matters

The comparison supports the view that the way views were brought together mattered, not just the fact that participants saw other opinions. It does not prove which parts of mediation produced the difference.

Question this helps explore

Did AI mediation reduce disagreement more than simply showing people one another’s views?

Evidence and limits

The comparison suggests that mediation contributed beyond unmediated exposure to others’ views. It supports the authors’ conclusion that the process reduced division partly through people’s views moving toward one another, but does not isolate every feature of the process that mattered.

AI can help humans find common ground in democratic deliberation · passages 16

What the idea means

The authors report similar convergence on some rounds in a human-mediator task when the human mediator outperformed the AI.

Why it matters

This finding suggests that the mediation process itself could help groups converge. Because the analysis was exploratory, it does not establish that human and AI mediation have the same effects.

Question this helps explore

Could the discussion process, rather than the AI itself, explain convergence?

Evidence and limits

The authors present this as an exploratory result, not a firm comparison of human and AI mediation. It leaves open whether the two mediators work in the same way or are equally effective.

AI can help humans find common ground in democratic deliberation · passages 16

What the idea means

Figure 2A shows the share of rankings that favored the AI statement over the human-written statement, alongside endorsement and quality ratings. Figure 2B reports ratings across nine quality dimensions; each point is a statement’s median rating averaged across five outside raters. Figure 2C compares rankings of revised statements that included critiques with initial statements that omitted them, and shows post-critique ratings in a comparison group. The caption says all error bars are bootstrapped 95% confidence intervals, meaning they show an estimated range of uncertainty calculated by resampling the data.

Why it matters

These measures distinguish participant preferences, ratings, and outside judges’ assessments. The uncertainty bars help readers judge the precision of the displayed results, but neither the ratings nor the bars establish how statements would perform in real public decisions.

Question this helps explore

How did the study compare group statements and show uncertainty in its results?

Evidence and limits

The caption separates participant rankings and ratings from outside judges’ ratings. It also states how the figure displays uncertainty. These methods describe the study’s measures, not proof that the statements would work well in real public decisions.

AI can help humans find common ground in democratic deliberation · passages 14

What the idea means

The AI was trained to produce group statements that maximized endorsement. The authors report that the discussion process reduced division at least partly through participants’ views moving toward one another.

Why it matters

This separates what the AI was trained to do from a change observed among people taking part in the discussion. It does not show that the AI was trained to change people’s views.

Question this helps explore

How did the AI’s training goal relate to the changes in participants’ views?

Evidence and limits

The authors distinguish the goal used to train the AI from the change observed during discussion. The reported reduction in division does not mean that changing participants’ views was itself the AI’s training goal.

AI can help humans find common ground in democratic deliberation · passages 15

What the idea means

The results do not show that minority views always gained ground or that both sides changed equally often. They describe how often the size of each side increased across rounds.

Why it matters

This helps distinguish making minority views heard from ensuring that all sides change equally.

Question this helps explore

Did support shift only toward majority views?

Evidence and limits

The authors report growth in both directions, but more rounds saw the majority grow than the minority.

Addresses: Whether the process allowed both majority and minority positions to gain support.

AI can help humans find common ground in democratic deliberation · passages 17

What the idea means

The reported average change favored the majority, but that does not mean every participant changed or that all participants came to agree.

Why it matters

Agreement levels and the direction of opinion change are different outcomes, so one should not be treated as proof of the other.

Question this helps explore

Did people move toward a shared view, or simply tend toward the majority?

Evidence and limits

The article reports an overall majority-leaning shift and distinguishes it from full agreement or convergence.

Addresses: The difference between an overall shift toward the majority and everyone reaching the same position.

AI can help humans find common ground in democratic deliberation · passages 17

What the idea means

The measure compares the shares of a group supporting opposing positions: a smaller difference means the group is less divided. The article reports that agreement increased after AI mediation, while it did not change when participants viewed one another’s opinions without mediation. The figure reports only rounds with all five members present at both measurement times. Its trend lines are fitted linear regressions with standard errors; its error bars show bootstrapped 95% confidence intervals.

Why it matters

The comparison distinguishes the reported AI-mediated result from mere exposure to other people’s views. The figure’s uncertainty intervals and fitted trends help show how the results were represented, but agreement is not the same as every member holding the same view.

Question this helps explore

How did the study measure group agreement, and did it increase with AI mediation compared with unmediated opinion exposure?

Evidence and limits

The article reports increased agreement and unanimity after AI-mediated deliberation, but no change in agreement in the unmediated opinion-exposure cohort. It also describes the figure’s regression lines and uncertainty intervals. The figure includes only rounds with all five group members present before and after discussion.

Addresses: The reported change in agreement, the measure used, and the comparison with viewing others’ opinions without AI mediation.

AI can help humans find common ground in democratic deliberation · passages 17, 18

What the idea means

The researchers used a text-encoding tool to represent each opinion with 768 numbers. They rotated those text representations onto an axis linking agreement and disagreement for each question. The resulting scores correlated with earlier ratings (r = 0.64), so they captured some, but not all, of the same information.

Why it matters

This construction matters because later comparisons of individual opinions and group statements rely on these scores as a way to locate their positions on each question.

Question this helps explore

How did the researchers estimate positions from written opinions?

Evidence and limits

The article describes the 768-dimensional text embeddings, their rotation onto an agree–disagree axis for each question, and the correlation between the resulting scores and earlier ratings.

Addresses: The method used to interpret participants’ text and compare it with their position ratings.

AI can help humans find common ground in democratic deliberation · passages 18

What the idea means

The comparison is about measured positions along the study’s text-score direction. It does not establish that every member felt fully represented by each statement.

Why it matters

A statement that falls within the group’s range may be a compromise rather than a position outside what any member expressed.

Question this helps explore

Did the group statements fall between members’ positions?

Evidence and limits

The article reports that 96% of statement scores lay within the range spanned by group members’ scores.

Addresses: Whether the written group statements stayed within the range of individual positions.

AI can help humans find common ground in democratic deliberation · passages 18

What the idea means

The estimated minority weight was 0.29 in initial statements, matching the average minority share, and 0.36 in revised statements. These are average estimates, not a guarantee that every statement treated views in the same way.

Why it matters

The result suggests that revisions can change whose views carry weight, so both the first and revised versions matter when assessing representation.

Question this helps explore

How did the statements weigh majority and minority views?

Evidence and limits

The article reports that initial statements matched the minority’s average share, while revised statements assigned minority views more weight.

Addresses: Whether the initial and revised statements reflected minority views in proportion to their numbers.

AI can help humans find common ground in democratic deliberation · passages 19

What the idea means

The exposure measure counted the fraction of group statements whose position scores were on the majority side of the median opinion. In this analysis, that fraction did not track with a larger shift toward the majority. The authors suggest that what the statements said mattered more than exposure alone.

Why it matters

This test weighs against the measured amount of exposure alone explaining the shift. It does not identify which parts of the statements influenced participants or establish the cause of the shift.

Question this helps explore

Could simply seeing more majority views explain the shift?

Evidence and limits

The analysis defined a statement as majority-leaning when its position score fell on the majority side of the median opinion, then tested whether the fraction of such statements was related to participants’ change in position.

Addresses: Whether exposure to majority-leaning statements was associated with later movement toward the majority.

AI can help humans find common ground in democratic deliberation · passages 19

What the idea means

The virtual assembly involved 200 people recruited to reflect the UK population on the listed demographic dimensions. Groups followed the same deliberation procedure as the main-task cohorts. Support for group opinions was similarly high, and agreement rose significantly from before to after deliberation.

Why it matters

The result offers evidence from a broader UK sample than the earlier convenience sample, using the same procedure for comparison. It does not establish results for every population or assembly.

Question this helps explore

Did the findings also appear in a UK citizens’ assembly?

Evidence and limits

The article reports the assembly sample, its use of the main-task procedure, and outcomes similar to the crowd-sourced cohorts, including a statistically significant increase in group agreement.

Addresses: Whether the results from earlier participant groups were also seen in a more population-representative sample using the same deliberation procedure.

AI can help humans find common ground in democratic deliberation · passages 20

What the idea means

The nine topics were immigration, the retirement age, the prison population, Brexit, climate change, universal childcare, the minimum wage, national pride, and the value of the internet. All groups discussed these questions over three weekly sessions of one hour each.

Why it matters

Giving all groups the same questions allows their discussions to be compared, while choosing topics participants considered important describes how the questions were selected. The findings do not establish that the same results would hold for other topics.

Question this helps explore

Which questions did the assembly use, and how were they chosen?

Evidence and limits

The article says every group discussed the same nine questions, gives the reason for selecting the topics, and states that the assembly had three weekly one-hour sessions.

Addresses: The shared topic design, stated basis for selecting the assembly’s questions, and session schedule.

AI can help humans find common ground in democratic deliberation · passages 20

What the idea means

The researchers compared ratings with a measure made from participants’ written opinions and found that the two tended to correspond.

Why it matters

This supports using that text measure to study how group statements relate to participants’ views. It does not show that the measure captures every aspect of an opinion.

Question this helps explore

Did the text measure reflect participants’ stated positions?

Evidence and limits

The figure caption reports a strong relationship between participants’ ratings and the position measure derived from their written opinions.

Addresses: How the study measured positions

AI can help humans find common ground in democratic deliberation · passages 21

What the idea means

The study’s text-based position measure showed that group statements were closer to a middle position than the original opinions.

Why it matters

This offers one measure of how the AI-supported process may help groups find common ground. A middle position does not by itself show that participants agreed with every part of a statement.

Question this helps explore

Did the group statements move toward a middle ground?

Evidence and limits

The figure caption describes the measured distribution of positions moving toward compromise across these stages.

Addresses: Changes in group statements

AI can help humans find common ground in democratic deliberation · passages 21

What the idea means

The measured comparison found that initial statements reflected minority views in proportion to their share, while revised statements gave those views more weight.

Why it matters

This suggests that revision could preserve the majority position while making dissenting views more visible. It does not show that every concern was included or accepted.

Question this helps explore

How did the statements represent majority and minority views?

Evidence and limits

The figure caption reports the difference between initial and revised statements. The discussion characterizes the approach as respecting the majority while giving dissenting views extra weight.

Addresses: Representation of majority and minority opinions

AI can help humans find common ground in democratic deliberation · passages 21, 23

What the idea means

In the analysis shown, groups were not more likely to move toward the majority when the AI statement leaned more toward that side.

Why it matters

This weighs against one simple explanation for group movement: that the AI’s majority-leaning wording drove it.

Question this helps explore

Did majority-leaning statements predict movement toward the majority?

Evidence and limits

The figure caption reports no association between the AI’s tendency to produce majority-leaning statements and groups moving in that direction.

Addresses: Relationship between statement position and group change

AI can help humans find common ground in democratic deliberation · passages 21

What the idea means

After deliberation, support changed by different amounts across five questions. The article reports stronger evidence for the prison and asylum changes and more marginal evidence for the other three.

Why it matters

The sizes of the reported shifts show that movement varied across topics. They do not establish that deliberation alone caused the changes or that views shifted on every issue.

Question this helps explore

How much did views shift on the issues where participants moved in a common direction?

Evidence and limits

The article reports the percentage changes for five issues and describes the analyses as exploratory. It reports stronger statistical evidence for the prison and asylum issues and more marginal evidence for the other three.

Addresses: Changes in individual ratings across issues

AI can help humans find common ground in democratic deliberation · passages 22

What the idea means

The researchers measured how often statements leaned toward agreement and compared that measure with people’s changes in view. They did not find a statistically clear relationship.

Why it matters

The result weakens one proposed explanation for the group-level changes, but it cannot show that the model had no bias or influence.

Question this helps explore

Could AI bias explain the group-level changes?

Evidence and limits

The article describes a score based on how often statements leaned toward agreement relative to the group’s median opinion, and reports no clear link with people changing toward that position.

Addresses: Possible influence of bias in the language model

AI can help humans find common ground in democratic deliberation · passages 22, 23

What the idea means

In the study, participants endorsed AI-mediated statements more often than statements written by incentivized, nonprofessional human mediators. The authors also report that groups were often less divided after using the AI mediator.

Why it matters

These findings suggest that participants often accepted the AI-mediated statements and that group views sometimes moved closer together. They do not show that agreement lasted or that the AI made political decisions.

Question this helps explore

How did participants respond to AI-mediated statements compared with statements from human mediators?

Evidence and limits

The discussion reports higher endorsement than for statements written by incentivized, nonprofessional human mediators and says groups were often less divided after interacting with the AI mediator.

Addresses: Participant responses to AI-mediated group statements

AI can help humans find common ground in democratic deliberation · passages 23, 24

What the idea means

The researchers used the language model’s text-based analysis to examine how group statements related to majority and minority opinions.

Why it matters

This gave the researchers a way to study what made the statements acceptable to participants. The claim does not show that the model can measure every useful feature of deliberation.

Question this helps explore

What did using a language model let the researchers examine?

Evidence and limits

The discussion says the model allowed the researchers to quantify features and that it respected the majority stance while giving extra weight to dissenting views.

Addresses: Analysis of statement features

AI can help humans find common ground in democratic deliberation · passages 23

What the idea means

The authors report similar endorsement and a rise in agreement for the virtual assembly, beyond the crowd-sourced groups used earlier.

Why it matters

This makes the findings more relevant to a wider range of UK residents than results from a convenience sample alone. It does not show that the same results would hold in other countries or settings.

Question this helps explore

Did the findings also appear in a citizens’ assembly sample?

Evidence and limits

The discussion says the findings were replicated in a virtual assembly using a sample described as demographically representative of the UK.

Addresses: Replication in a UK citizens’ assembly

AI can help humans find common ground in democratic deliberation · passages 23

What the idea means

Participants’ final judgments favored statements revised after critique over the initial versions, and the group statements received positive ratings.

Why it matters

This supports the value of the revision step in this process. It does not show that revised statements are always better in other settings.

Question this helps explore

Did participants prefer revised statements?

Evidence and limits

The figure caption reports that revised statements were preferred in the final judgment and that group statements received positive endorsement and high quality ratings.

Addresses: Statement preference and quality ratings

AI can help humans find common ground in democratic deliberation · passages 24

What the idea means

The system used a simulated ranked-choice process to select from candidate group statements.

Why it matters

This describes one step in the method, but the supplied passage does not explain the process further or show its effects.

Question this helps explore

How did the AI system select among candidate statements?

Evidence and limits

The passage says the system selected a statement from candidates by simulating a ranked-choice process, but the supplied text cuts off before giving further details.

Addresses: Candidate statement selection

AI can help humans find common ground in democratic deliberation · passages 24

What the idea means

The authors distinguish their approach from earlier, more hand-engineered systems that could debate with people, summarize arguments, or support deliberation without aiming to find common ground. They say that, to their knowledge, their study is the first demonstration of AI mediating collective deliberation at scale.

Why it matters

This explains how the authors place their work in relation to earlier research. Their claim about being first is qualified and is not independent proof that no similar work existed.

Question this helps explore

How do the authors distinguish this work from earlier AI research?

Evidence and limits

The discussion contrasts the study with earlier systems described as more hand-engineered and lists their functions before making a qualified claim about demonstrating AI mediation at scale.

Addresses: The authors’ account of the study’s contribution

AI can help humans find common ground in democratic deliberation · passages 23

What the idea means

The authors say the approach may help groups deliberate efficiently, fairly, and at larger scale. For timing, they report that the system produced high-quality group statements within seconds, whereas human mediators took several minutes.

Why it matters

The timing comparison supports the authors’ claim of time efficiency in the study. It does not by itself establish fairness or show that the approach can work at larger scale in real decisions.

Question this helps explore

What practical value do the authors see in the approach, and how fast did it produce statements?

Evidence and limits

The discussion identifies time efficiency, fairness, and scalability as practical opportunities and reports that the system produced high-quality statements within seconds, compared with several minutes for human mediators.

Addresses: The stated practical value and timing of statement production

AI can help humans find common ground in democratic deliberation · passages 24

What the idea means

The researchers asked whether participants who debated freely would tend to move toward a shared stance across groups. The article links this question to Habermas’s proposal.

Why it matters

This question helps explain why the study measured changes in participants’ views. It does not mean that the proposed outcome was expected on every issue or that the question itself proves convergence occurred.

Question this helps explore

Can free deliberation help different groups move toward a common position?

Evidence and limits

The article frames the study as testing whether AI-mediated deliberation led to convergent opinion shifts across groups, connecting this question to Habermas’s proposal.

Addresses: The study’s central research question

AI can help humans find common ground in democratic deliberation · passages 21

What the idea means

The authors say the statements reflected critiques from minority participants without discarding the majority view.

Why it matters

This matters because a shared statement can acknowledge disagreement rather than present the majority’s view alone.

Question this helps explore

How did the group statements handle minority views?

Evidence and limits

This is a reported finding from the authors’ study, not independent proof that the method will work equally well in other settings.

Addresses: The Habermas Machine’s group statements

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

Each participant’s vote is meant to count equally.

Why it matters

Equal voting weight can help prevent some participants from having more formal influence than others.

Question this helps explore

What does fairness mean in the group’s decision process?

Evidence and limits

The passage gives this as a fairness principle; it does not establish that every part of the process meets it.

Addresses: The authors’ account of fairness

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

The authors think a model with a much larger capacity for reading text might process many more opinions.

Why it matters

This could make the method useful beyond small groups, but the passage does not show that it works at that scale.

Question this helps explore

Could the method support much larger groups?

Evidence and limits

The larger-group claim is a theoretical possibility, not a tested result.

Addresses: The method’s possible scale

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

The authors report stronger performance from a smaller model trained for this task than from a larger model used with prompts. They propose trying task-specific training with a larger model.

Why it matters

The result suggests that model size alone may not determine performance, but the proposed combination still needs testing.

Question this helps explore

How did the tested models compare, and what improvement did the authors propose?

Evidence and limits

The comparison is a reported study result; the suggested improvement is a possibility, not a tested outcome.

Addresses: The model comparison and proposed next step

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

The authors see potential for the method to reduce the expense and practical limits of in-person meetings, as well as some social pressures and imbalances.

Why it matters

These possible gains help explain why an AI-assisted approach might be useful, but they need to be weighed against what it cannot provide.

Question this helps explore

What problems with in-person deliberation might the method reduce?

Evidence and limits

These are proposed advantages, not proof that the method removes these problems in practice.

Addresses: The authors’ comparison with in-person deliberation

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

The authors found comparable results with a demographically representative UK assembly and when they adjusted their convenience-sample results for demographic imbalance.

Why it matters

These checks make it less likely that the findings came only from the makeup of one convenience sample, while leaving wider generalization unresolved.

Question this helps explore

What evidence do the authors offer about whether the findings may generalize?

Evidence and limits

The added checks concern UK participants and do not remove the uncertainty about other contexts.

Addresses: The additional UK sample and demographic adjustment

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

A participant may accept a shared statement as a fair result while retaining a different personal opinion.

Why it matters

This distinction matters when judging whether a group has found common ground or simply hidden disagreement.

Question this helps explore

Does agreement mean that participants share the same opinion?

Evidence and limits

The authors distinguish acceptance of a fair process from agreement with every part of its outcome.

Addresses: What endorsement of a group statement can mean

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

People may agree on information useful for later choices without agreeing on every opinion or proposal.

Why it matters

This broader idea of common ground allows a group to move forward without claiming that everyone thinks alike.

Question this helps explore

Can shared information be common ground when people still disagree?

Evidence and limits

This is the authors’ account of one possible form of agreement, not a claim that disagreement disappears.

Addresses: Common ground as a basis for later decisions

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

Expert testimony can inform a discussion without dictating its result. The authors also say their questions were designed to lower the chance of harmful or unproductive discussion.

Why it matters

These measures may help, but the passage does not say they replace fact-checking or moderation.

Question this helps explore

What steps can help reduce the risks of deliberation without predetermining its outcome?

Evidence and limits

The passage describes expert input as a possible safeguard and the question design as one measure used in this project.

Addresses: Expert testimony and the study’s discussion questions

AI can help humans find common ground in democratic deliberation · passages 27

What the idea means

The system should be one part of deliberation, alongside decisions about who takes part and whether affected groups are represented.

Why it matters

The makeup of the group can shape whose views appear in the resulting statement.

Question this helps explore

What conditions do the authors propose for real-world use?

Evidence and limits

This is a recommendation for possible use, not a finding that the system alone ensures fair representation.

Addresses: Embedding the system in a broader process

AI can help humans find common ground in democratic deliberation · passages 27

What the idea means

The authors did not detect political slant from the model in their analyses and say dissenting views were represented in the statements.

Why it matters

This offers some reassurance about the tested process, but it does not guarantee fairness in future uses or other settings.

Question this helps explore

What did the authors find about political bias and fair representation?

Evidence and limits

The reported finding is limited to the analyses described; the authors still call for vigilance in other uses.

Addresses: The study’s analysis of group statements

AI can help humans find common ground in democratic deliberation · passages 27

What the idea means

The authors identify several kinds of group work where people may need to agree on the wording or content of a written statement.

Why it matters

These examples suggest a range of possible uses, but do not establish that the method is suitable for each one.

Question this helps explore

Where do the authors see possible uses for group statements?

Evidence and limits

These are possible applications offered by the authors, not settings in which this study tested the system.

Addresses: The authors’ examples of settings that need shared written statements

AI can help humans find common ground in democratic deliberation · passages 27, 28

What the idea means

People may need some shared understanding before they can work together toward a common purpose.

Why it matters

If the method helps groups find that shared understanding, it could be useful beyond producing a statement; the passage does not show that it causes successful joint action.

Question this helps explore

Why do the authors think common ground may matter beyond discussion?

Evidence and limits

The broader social benefit is the authors’ argument about potential, not a measured result of the study.

Addresses: Common ground as a step toward collective action

AI can help humans find common ground in democratic deliberation · passages 28

What the idea means

The authors used the Chinchilla 70B language model to draft questions from examples they had written, then reviewed those questions against several criteria.

Why it matters

Question wording can shape what people discuss and which views they can express. The review was intended to address some concerns, but its effectiveness is not established here.

Question this helps explore

How did the authors create and review the discussion questions?

Evidence and limits

This describes how the study’s question set was made. The passage does not establish that every question was free of risk or bias.

Addresses: The question-generation and review process

AI can help humans find common ground in democratic deliberation · passages 28

What the idea means

The study used both crowdsourced UK adults and an assembly recruited through randomized stratified sampling to represent several population characteristics.

Why it matters

Knowing who took part and how they were recruited helps readers judge which people the results describe and where they may not apply.

Question this helps explore

Who took part in the study, and how and when were they recruited?

Evidence and limits

These details describe the people and recruitment methods covered by the study; the passage does not provide the full demographic results.

Addresses: The study’s participant groups and data collection

AI can help humans find common ground in democratic deliberation · passages 28

What the idea means

The main task and ablation experiments examined how important different system components were. Other visible categories concern the critique phase, belief change, and human mediators.

Why it matters

These categories show that the authors examined different parts of the process, but this passage does not give the experiments’ results or finish describing the list.

Question this helps explore

What kinds of evaluations does the passage list, and what did the main task and ablation experiments assess?

Evidence and limits

The passage states the purpose of the main task and ablation experiments and names other evaluation categories, but cuts off before giving the full list and details.

Addresses: The study’s evaluation experiments

AI can help humans find common ground in democratic deliberation · passages 28

What the idea means

A group statement may bring together a range of opinions rather than present a single shared position.

Why it matters

This matters because readers should not assume that a shared statement means everyone holds the same view.

Question this helps explore

Can a group statement show several views rather than one shared position?

Evidence and limits

This describes the possible content of the statements, distinct from why participants may accept them or whether disagreement continues.

Addresses: The range of opinions a group statement can contain

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

The authors value an approach that could help groups reach agreement without taking too much time, while treating people fairly and supporting larger groups.

Why it matters

These are important goals for real-world use, but the passage does not establish that the approach achieves all of them in practice.

Question this helps explore

What benefits do the authors hope technology could bring to group deliberation?

Evidence and limits

The authors present these qualities as reasons such a technology could be beneficial, rather than as a demonstrated result in every setting.

Addresses: The authors’ description of the potential benefits of technology for finding agreement

AI can help humans find common ground in democratic deliberation · passages 27

What the idea means

The evaluation included a comparison with human mediators and a virtual assembly intended to test whether findings extended beyond other study groups.

Why it matters

These settings address different questions, but the supplied passage does not show whether the AI performed well in either one.

Question this helps explore

How did the authors test the mediator in different settings?

Evidence and limits

The passage names these evaluation settings but does not give their results or specify here which parts of the population the sample represented.

Addresses: The study design described in the article.

AI can help humans find common ground in democratic deliberation · passages 29

What the idea means

Preregistration means recording key study plans before analyzing results. Here, the authors identify sample sizes, exclusions, and primary analyses as preregistered.

Why it matters

Setting these plans in advance can make it clearer which analyses were planned, but the passage alone cannot show whether every plan was followed.

Question this helps explore

Which parts of the evaluation plan were set in advance?

Evidence and limits

The passage reports preregistration but does not provide the details of the registered plan.

Addresses: The authors’ reported study procedures.

AI can help humans find common ground in democratic deliberation · passages 29

What the idea means

In each discussion block, participants wrote opinions, rated and ranked group statements, critiqued a highly ranked statement, rated and ranked revised statements, expressed final preferences, and completed a survey. After the task, they gave free-form comments and answered demographic questions.

Why it matters

The procedure let the authors compare views before and after discussion. The passage does not provide results from that comparison.

Question this helps explore

What did participants do in the main deliberation task?

Evidence and limits

The passage describes the task stages and activities, but does not establish how the task affected participants’ views.

Addresses: The human task procedure described in the article.

AI can help humans find common ground in democratic deliberation · passages 29

What the idea means

Fine-tuning means further training a language model on selected examples. Here, the examples included participant opinions and critiques, with highly rated group statements as targets.

Why it matters

This describes the model and material used for training, but does not show whether the examples represented other groups or led to good outputs in new situations.

Question this helps explore

What material was used to train the AI mediator’s language model?

Evidence and limits

The counts and model name describe the reported training setup; they do not show how well the model learned or worked.

Addresses: The model-training method reported in the article.

AI can help humans find common ground in democratic deliberation · passages 29, 30

What the idea means

The authors withheld 12 topics during fine-tuning and used them to test performance on topics absent from that training step.

Why it matters

This design can examine performance on held-out topics, but the passage does not show whether performance was strong or whether the topics covered all new situations.

Question this helps explore

How did the authors test whether the language model could handle topics unlike its training topics?

Evidence and limits

The passage says the topics were held out for this test but does not report the test’s results.

Addresses: The reported test of generalization to new topics.

AI can help humans find common ground in democratic deliberation · passages 30

What the idea means

The authors converted participants’ rankings into pairwise preferences, comparing statements two at a time. A Bradley–Terry model was used to fit these preferences and produce rewards proportional to each participant’s expected agreement with a candidate statement. The training also included a separate classification task based on quality ratings divided into two categories.

Why it matters

This explains how the model turned rankings and quality ratings into scores. The passage does not establish that its predictions matched people’s views in practice.

Question this helps explore

How did the AI mediator estimate which group statements people would prefer?

Evidence and limits

The passage describes the model design and training signals, but does not show how accurately it predicted people’s preferences.

Addresses: The reward-model training method reported in the article.

AI can help humans find common ground in democratic deliberation · passages 30

What the idea means

The system used each participant’s opinion and critique to predict how they would rank 16 candidate statements. The Schulze method combines ranked preferences to choose a winner. The authors reran the process to give participants more than one sample in the main evaluations.

Why it matters

This describes how the system turned predicted individual rankings into a group choice, but does not show whether participants accepted the choice or considered it fair.

Question this helps explore

How did the AI mediator choose a group statement?

Evidence and limits

The passage reports a simulated election among generated candidates. It does not establish that the winner represented every participant fairly or accurately.

Addresses: The statement-selection procedure described in the article.

AI can help humans find common ground in democratic deliberation · passages 30

What the idea means

The authors used mixed-effects models, which can account for differences among participants or groups, and ordinal models for ordered rating scales. They say they used the fullest random-effects structure permitted by the data and checked for results in either direction in significance tests.

Why it matters

These choices describe how the authors analyzed the data, but the passage alone does not show the size, reliability, or practical meaning of any result.

Question this helps explore

How did the authors analyze participants’ ratings and other results?

Evidence and limits

The passage identifies analysis choices but does not provide estimates, test results, or evidence that these choices resolve every source of uncertainty.

Addresses: The statistical procedures reported in the article.

AI can help humans find common ground in democratic deliberation · passages 30, 31

What the idea means

An embedding is a numerical representation of text. The authors embedded opinions, group statements, and affirming and negating position statements for each question. The line between the two position statements defined an axis; a text’s projection onto it represented its position, while the remaining components were treated as separate from that position.

Why it matters

This method gives the authors a defined way to compare positions in text. Results may depend on the text model and the choice of axis.

Question this helps explore

How did the authors measure the position expressed in a statement?

Evidence and limits

The axis was built from representations of affirming and negating position statements for each question, not from unspecified examples. The authors also note that other text models and axis definitions are possible.

Addresses: The text-representation method described in the article.

AI can help humans find common ground in democratic deliberation · passages 31

What the idea means

The authors used participants’ ratings to identify which views fell on each side of neutral. They modeled group-statement positions using convex combinations of the participants’ opinions, meaning weighted combinations whose weights sum to one. They added the estimated weights for minority opinions and repeated the calculation separately for groups with different degrees of division, then reported the average.

Why it matters

This provides one way to compare minority influence across group divisions, but it depends on the ratings and modeling choices described.

Question this helps explore

How did the authors estimate the influence of minority opinions on group statements?

Evidence and limits

The passage gives an operational definition and a modeling procedure; it does not establish that the resulting weight is a complete measure of influence or fairness.

Addresses: The minority-weight analysis described in the article.

AI can help humans find common ground in democratic deliberation · passages 31

What the idea means

For the majority-bias measure, the authors counted the share of generated statements that leaned toward the majority side rather than the minority side of the median opinion. In the virtual assembly, they also compared statement positions with the agree and disagree sides around the median opinion to assess agreement bias.

Why it matters

These definitions make particular kinds of leaning measurable, but they do not by themselves establish that the system was biased or unbiased.

Question this helps explore

How did the authors define possible majority and agreement biases?

Evidence and limits

These are the authors’ stated measures; the passage does not report their values or show that they capture all forms of bias.

Addresses: The bias measures described in the article.

AI can help humans find common ground in democratic deliberation · passages 31

What the idea means

The authors report registering the planned sample sizes, rules for excluding data, and main analyses for every evaluation experiment.

Why it matters

Readers can consult the registrations to compare the planned methods with the reported work. The passage does not give the registration results or establish that every plan was followed.

Question this helps explore

What study plans did the authors register in advance?

Evidence and limits

Passage 38 says these study details were preregistered. This helps readers check what the researchers planned before conducting the evaluations.

Addresses: The article’s evaluation experiments

AI can help humans find common ground in democratic deliberation · passages 38

What the idea means

According to the authors, the dataset of questions and human feedback used for training and evaluation is available through GitHub and Zenodo. The full set of seed questions is in the supplementary materials.

Why it matters

Access to these materials can help readers examine the work. The passage does not establish that the materials alone are enough to reproduce every result.

Question this helps explore

What research materials can readers access?

Evidence and limits

Passage 38 identifies where readers can find the dataset and seed questions. This makes it easier to inspect or reuse those materials, but the passage does not say that every supplementary research material is included in the dataset.

Addresses: The article’s training and evaluation materials

AI can help humans find common ground in democratic deliberation · passages 38

What the idea means

The authors report that 803 of 1,047 rounds, about 77%, began with disagreement. They then raise the question of whether the AI represents all viewpoints equally.

Why it matters

The frequency of initial disagreement shows that majority and minority views could matter in many rounds. These passages do not show whether the AI treated those views equally.

Question this helps explore

How should a mediator balance majority and minority views, and does the AI represent them equally?

Evidence and limits

The authors identify balancing majority and minority views as an important issue and ask whether the AI represents viewpoints equally. The reported share of rounds that began with disagreement shows how often this issue arose in the main study groups. These passages do not provide the answer.

AI can help humans find common ground in democratic deliberation · passages 16

What the idea means

The AI statements averaged about 126 words, versus 105 for human statements. Length affected ratings and rankings, but the AI statements remained ahead on endorsement ratings when length was taken into account.

Why it matters

The reported figures show the size of the length difference being considered. Controlling for length limits the claim that it alone drove endorsement, while leaving length as a factor that influenced responses.

Question this helps explore

Could statement length explain the preference for AI-written statements?

Evidence and limits

The passage identifies length as a possible explanation, reports the average lengths, and says the endorsement difference remained after controlling for length. It does not say length had no effect on other ratings or rankings.

Addresses: Effect of statement length

AI can help humans find common ground in democratic deliberation · passages 10, 11

What the idea means

Participants judged argument quality separately from their own agreement, then ranked statements by agreement. A ranked-choice process selected the winner for the next step.

Why it matters

These separate judgments helped determine which statement advanced, while participants’ knowledge that an algorithm produced the statements is relevant to interpreting their responses.

Question this helps explore

How were initial statements assessed and selected?

Evidence and limits

These details clarify how the researchers gathered participants’ judgments and chose the statement for critique.

Addresses: Study procedure

AI can help humans find common ground in democratic deliberation · passages 9

What the idea means

The human writers were group participants chosen at random for the mediator role, not outside professional mediators. They were trained and offered a bonus tied to group preference.

Why it matters

These design details matter when interpreting the comparison. The reported conclusion concerns the difficulty of producing a group-preferred statement, not whether AI should make group decisions.

Question this helps explore

What should readers know about the human comparison?

Evidence and limits

The assignment and incentive help describe the comparison, while the authors’ conclusion identifies the task being compared as difficult for human mediators.

Addresses: Human mediator comparison

AI can help humans find common ground in democratic deliberation · passages 10

What the idea means

The system did not revise from critiques alone: it also used the group’s earlier opinions and the initial winning statement. One question’s deliberation took about 15 minutes.

Why it matters

These details clarify the revision mechanism and the time required in the reported procedure; they do not establish how long the process would take elsewhere.

Question this helps explore

What information went into revisions, and how long did a round take?

Evidence and limits

The inputs to revision and the approximate round length describe how the process operated.

Addresses: Revision procedure and duration

AI can help humans find common ground in democratic deliberation · passages 11

What the idea means

The authors propose that minority critiques may affect revisions. Their interpretation also draws on participants’ preference for revised statements over initial ones.

Why it matters

This explanation matters because a shift in the wording could reflect an exchange between members, rather than a simple move from one side to the other.

Question this helps explore

Why might revised statements give minority views more weight?

Evidence and limits

The passage presents the critique-and-revision sequence as an interpretation of the observed pattern, not as a separately established cause.

Addresses: The authors’ proposed explanation for the change between initial and revised statements.

AI can help humans find common ground in democratic deliberation · passages 19

What the idea means

Participants discussed questions, but did not take part in the usual fact-finding or expert-testimony phases.

Why it matters

This limits how directly the findings apply to a full citizens’ assembly that includes those stages.

Question this helps explore

How closely did the virtual assembly match a full citizens’ assembly?

Evidence and limits

The article explicitly describes the assembly as mainly mimicking the discussion phase and notes which usual stages it left out.

Addresses: The limits of the assembly design compared with a typical citizens’ assembly.

AI can help humans find common ground in democratic deliberation · passages 20

What the idea means

The nine questions did not all show movement in the same way. The childcare result pointed toward more opposition, but the reported rating change was not statistically significant.

Why it matters

These differences caution against treating convergence as a general result for every topic.

Question this helps explore

Were the opinion shifts consistent across all issues?

Evidence and limits

The article reports no change on Brexit and a nonsignificant shift toward opposition to free childcare.

Addresses: Limits on the reported shifts

AI can help humans find common ground in democratic deliberation · passages 22

What the idea means

Online or AI-supported discussion may not offer the same body-language cues or personal connection as meeting face to face.

Why it matters

A process that produces a shared statement may still miss relationship-building and other parts of group discussion.

Question this helps explore

What might be lost when deliberation is not in person?

Evidence and limits

This qualification limits the authors’ case for replacing or avoiding in-person meetings.

Addresses: The potential advantages of face-to-face discussion

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

The authors are unsure whether the results apply to other groups and settings, but say they had no reason to think the UK context itself produced them.

Why it matters

This is the authors’ rationale, not proof that the findings apply elsewhere; the study’s UK setting still limits what it can establish.

Question this helps explore

How far might the findings apply beyond the people and issues studied?

Evidence and limits

The authors state both a limit on what their UK-based study can establish and their reason for not attributing the results to features unique to the UK.

Addresses: The study’s population and setting

AI can help humans find common ground in democratic deliberation · passages 25

What the idea means

A group can reach some agreement while participants continue to hold different opinions.

Why it matters

A process should not treat a shared statement as proof that all disagreement has ended.

Question this helps explore

Does deliberation always produce a single shared position?

Evidence and limits

This qualification rejects the idea that a group statement must erase continuing differences.

Addresses: The limits of consensus

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

The system does not check whether opinions are true, keep people focused, or moderate what they say.

Why it matters

Its output may carry forward harmful or mistaken input, so a generated statement should not be treated as verified or safe.

Question this helps explore

What risks follow from the system’s current limits?

Evidence and limits

The authors tie the risk of harmful output to the opinions people provide and to functions the system currently lacks.

Addresses: The system’s ability to manage discussion content

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

The authors say other democratic processes also share the described possibility: harmful or poorly informed human opinions can lead to harmful or poorly informed outcomes. They frame this alongside the principle that agreement can be legitimate when citizens deliberate freely and fairly.

Why it matters

This comparison gives context for the system’s limitation; it does not mean that harmful outputs are harmless or that every process handles the risk equally well.

Question this helps explore

How do the authors relate this limitation to other democratic processes?

Evidence and limits

The passage explicitly says this feature is shared with other democratic processes and links it to free and fair citizen deliberation.

Addresses: The authors’ comparison with wider democratic deliberation

AI can help humans find common ground in democratic deliberation · passages 26

What the idea means

The critique phase was excluded in two tasks. In the human-mediator task, participants completed four rounds, and third-party raters judged the transcripts.

Why it matters

Because the procedures differed, results from these tasks would need to be interpreted in light of those differences; no such results appear here.

Question this helps explore

How did the human-mediator procedures differ from the main task?

Evidence and limits

These differences apply to the named tasks and should not be treated as features of every study condition.

Addresses: The human-task procedure described in the article.

AI can help humans find common ground in democratic deliberation · passages 29