Who Makes Sense When AI Joins the Conversation?

Collective Sensemaking, Human Agency and the New Architecture of Collaboration

By Paul Nunesdea | Thinking Better Together

We have never had so much information at our fingertips. Yet making sense of the world together may be getting harder.

Organisations can now summarise thousands of documents in seconds, identify patterns across conversations and generate remarkably coherent interpretations of complex situations. Artificial intelligence promises to help us understand more, decide faster and collaborate better.

But we should ask a question before we welcome AI into every meeting, workshop, and decision-making process.

As machines help us make sense of the world, are we strengthening our collective intelligence or quietly outsourcing the very process through which it emerges?

The distinction matters because understanding something individually is not the same as making sense of it collectively. And producing a convincing summary is not the same as reaching a shared understanding.

Making sense is something we do together

We often think of sensemaking as an intellectual activity: gathering information, analysing it and arriving at a reasonable interpretation.

But collective sensemaking is much more than that.

It is a profoundly social process through which people discover not only what they think, but how others see the same reality. It involves making assumptions visible, encountering unfamiliar perspectives and sometimes discovering that the problem we thought we were solving is not the problem at all.

Consider a group of healthcare professionals, policymakers, technology providers, and patient representatives discussing introducing an artificial intelligence system into clinical practice.

They may all agree that the objective is to improve healthcare. Yet they may understand improvement quite differently.

For the clinician, it may mean better decisions and less administrative burden. For the patient, it may mean greater trust and meaningful control over personal information. For the technology provider, it may mean reliable performance and scalability. For the policymaker, it may mean equitable access and accountability.

None of these interpretations is necessarily wrong.

The challenge is not simply to combine them into a common document. It is to create the conditions for these different perspectives to encounter one another, challenge assumptions, and contribute to a more complete understanding of the situation.

That is collective sensemaking.

And it is one of the most important things that happens when people collaborate successfully.

The architecture behind collective understanding

A recent paper by Linxuan Ying, Organizational Mechanisms in Digital Public Health Transformation, provides a useful starting point.

Ying proposes three interconnected organisational mechanisms through which digital transformation can become meaningful collective action: change communication, collective sensemaking and cross-functional collaboration.

Change communication helps establish a credible direction that people can understand and question. Collective sensemaking allows different actors to develop shared interpretations of what is happening and why it matters. Cross-functional collaboration translates that understanding into coordinated action.

Importantly, these mechanisms are not presented as a simple linear sequence. They continuously influence one another as people learn, negotiate and adapt.

Although developed in the context of public health, this framework has implications far beyond healthcare.

It helps explain why many transformation programmes struggle despite having sufficient resources, competent leadership and sophisticated technology.

Organisations may be technologically ready for change without being collectively ready to make sense of it.

People can receive the same information and still interpret it differently. They can agree with a strategy without understanding its consequences in the same way. They can formally accept a decision without developing the relationships needed to implement it.

This is where professional facilitation becomes essential. But it is also where the wider practice of architecting collaboration begins.

When AI enters the conversation

Until recently, the tools supporting collective sensemaking were largely passive. Documents recorded information, visualisations displayed relationships and facilitation methods structured interaction.

Of course, none of these tools was entirely neutral. The choice of what to document, what to visualise and which questions to ask has always influenced how groups interpret their reality.

Artificial intelligence introduces another layer.

A generative AI system can now actively help frame a problem. It can propose categories, identify patterns participants may not have considered significant, and suggest connections they may not have considered.

It can also produce a remarkably persuasive account of what a group has collectively discussed.

This creates opportunities.

Imagine a large stakeholder consultation involving hundreds of participants. AI could identify recurring concerns, connect contributions across sessions, and reveal contradictions that might otherwise go unnoticed.

It could let facilitators spend less time on documentation and more on the human dynamics of the conversation.

For collaboration architects working across organisations, it could become a tool for maintaining continuity between gatherings, preserving institutional learning, and supporting shared reflection over time.

These possibilities are significant.

But they introduce a fundamental complication.

AI does not merely accelerate the processing of meaning. It can influence which meanings become visible and which interpretations gain authority.

A summary is always a selection. A pattern is always an interpretation of relationships. Even a seemingly neutral synthesis reflects decisions about what deserves attention and what can be left out.

When an algorithm makes those decisions, often without participants seeing them, the architecture of collective understanding begins to change.

The seductive efficiency of premature agreement

One of a skilled facilitator's most valuable contributions is recognising when a group is converging too quickly.

Apparent agreement can conceal very different interpretations of the same decision. Silence may indicate acceptance, but it may equally reflect fatigue, uncertainty or reluctance to challenge someone with greater authority.

Experienced facilitators learn to recognise these distinctions.

They understand that disagreement is not necessarily an obstacle to collaboration. Sometimes it is the very material from which a more robust shared understanding can emerge.

Now imagine an AI system producing a coherent summary while the group still struggles with conflicting perspectives.

The result may appear helpful. It has organised differences, identified common themes, and constructed a clear narrative.

But has the group actually developed a shared understanding?

Or has the technology simply produced the appearance of one?

Research published in 2026 by Aaron Parisi and colleagues offers a useful warning.

In two studies involving 879 participants in small-group deliberations, participants generally preferred AI-facilitated discussions, but facilitation did not significantly improve group consensus. The researchers also found that AI facilitators could influence certain allocation decisions without corresponding improvements in equitable participation.

This does not establish that AI facilitation is generally harmful. The research examined a specific type of text-based group decision-making task, and its conclusions should not automatically be extended to every organisational setting.

Nevertheless, the findings expose an important distinction.

A process can feel more inclusive without becoming more inclusive. It can feel more coherent without producing deeper collective understanding.

That possibility should concern anyone responsible for designing collaborative processes.

Efficiency is valuable, but it is not a substitute for legitimacy. Nor should the smoothness of a conversation be confused with the quality of the thinking taking place.

Whose interpretation becomes collective knowledge?

Another, perhaps more subtle, challenge remains.

In complex organisations, people with equal influence rarely create collective understanding.

Some participants possess formal authority, while others contribute practical knowledge that may be less visible. Some have access to data and specialised language, while others bring lived experiences that are difficult to translate into conventional organisational categories.

This imbalance exists long before AI enters the room.

But AI can reproduce or amplify it.

When a system processes a collection of statements, the perspectives expressed most frequently or most fluently may become disproportionately visible in the resulting synthesis.

A minority perspective may be crucial precisely because it challenges the dominant narrative. Yet an automated summary designed to identify common themes may treat it as peripheral.

Consider a patient representative who raises an ethical concern that nobody else mentions. Or a frontline professional who identifies a practical difficulty that contradicts senior management's assumptions.

These contributions may be among the most valuable in the room.

The real test of collective sensemaking is not whether the summary captures the majority opinion. It is whether the process helps the group recognise perspectives that might otherwise be overlooked.

This connects closely with an emerging argument in Sensemaking AI research, notably developed by Tina Comes.

Rather than evaluating AI exclusively through optimisation and decision accuracy, Comes proposes paying attention to the capacity of human–AI networks to preserve interpretive flexibility, collective agency and adaptability.

For those of us working in professional facilitation, the implication is profound.

The objective should not be to eliminate ambiguity as quickly as possible. Instead, it should help people work intelligently with ambiguity until they are ready to act.

That may be one of the most important design principles for collaboration in the age of AI.

From facilitating conversations to architecting collective intelligence

This brings us to a distinction I have increasingly found useful in my own professional practice.

Facilitating a group and architecting collaboration are closely related activities, but they operate at different levels.

A facilitator creates the conditions for people to think, interact and make progress within a particular gathering.

A collaboration architect works with the wider architecture that lets those interactions produce continuity, shared learning, and coordinated action over time.

The facilitator helps the group construct meaning in the room.

The collaboration architect helps that meaning travel beyond the room without losing its legitimacy, context or capacity to evolve.

This distinction becomes especially important when AI is introduced.

A facilitator may use AI to support a discussion, organise contributions or prepare a synthesis for participants to review.

A collaboration architect must consider how those interpretations are preserved, challenged and reused across a wider ecosystem.

What happens when a summary produced in one workshop becomes the starting point for another? Who has the authority to correct it? Which disagreements remain visible, and which have disappeared into a polished narrative?

And perhaps most importantly, how do we prevent provisional interpretations from becoming unquestioned institutional truths simply because intelligent systems have captured and repeatedly reproduced them?

These are no longer merely technical questions.

They are questions of governance, participation and collective responsibility.

The collaboration architect must therefore become attentive not only to the human relationships that make collaboration possible, but also to the information systems that increasingly mediate those relationships.

Designing AI for better collective sensemaking

I believe AI can substantially contribute to collective intelligence, provided we are deliberate about the role we give it.

The most promising approach may be to treat AI not as the author of collective understanding, but as an instrument that helps people examine and improve their own interpretations.

Instead of asking AI to produce the definitive summary of a discussion, we might ask it to identify alternative readings of the same contributions.

Instead of asking it to tell us what the group agreed, we might invite it to distinguish confirmed agreements from unresolved differences.

Instead of letting AI decide which contributions matter most, we might use it to identify underrepresented perspectives or questions that remain unexplored.

This also offers a particularly interesting opportunity for large-scale collaborative processes.

Imagine a World Café in which participants first develop their interpretations independently and then discuss them in small groups. AI could subsequently examine the documented contributions and suggest additional patterns or contradictions.

But the essential step would come afterwards.

Participants would return to those AI-generated interpretations and challenge them.

What has been misunderstood? What has been omitted? Which connections are useful, and which are speculative? What do we recognise as our shared understanding, and what remains contested?

Such an approach would preserve human agency while extending the group's analytical capacity.

It would also recognise that a machine, however capable, cannot complete collective sensemaking.

The people who will live with the consequences of a decision must remain responsible for interpreting and validating the meaning on which that decision rests.

Naturally, this requires appropriate safeguards for privacy, consent, data provenance and the use of sensitive information.

But these are part of the architecture of collaboration, not external administrative considerations.

Beyond the meeting: the challenge of collective memory

One further possibility deserves attention.

Perhaps AI's most transformative contribution to collaboration will not happen during meetings at all.

It may happen between them.

One persistent weakness of collaborative initiatives is the difficulty of maintaining continuity. People meet, exchange perspectives and develop promising ideas, only to return to their organisations where everyday priorities gradually disperse the momentum.

Knowledge remains scattered across reports, meeting notes and personal recollections. New participants enter the process without understanding its history, while previous disagreements are forgotten or simplified.

For collaboration architects, this challenge is familiar.

An ecosystem cannot develop collective intelligence if it continually loses the memory of its own learning.

AI could help address this problem by connecting conversations across time, recovering earlier assumptions and showing how collective interpretations have evolved.

It could help identify commitments that were never completed and questions that keep returning without resolution.

But the same caution applies.

Collective memory must remain open to correction. It should preserve the evolution of understanding rather than merely reproduce the most recent official narrative.

A genuinely intelligent collaborative ecosystem needs more than a searchable archive.

It needs a living memory that supports reflection, accountability and adaptation.

And that memory must ultimately belong to the people and communities whose work it represents.

The human work we should never outsource

The increasing sophistication of AI invites a tempting assumption: if machines can analyse more information than we can, they should also help us reach better conclusions.

Sometimes they will.

But the most consequential aspects of collaboration have never been reducible to information processing.

Trust emerges through relationships. Legitimacy develops through participation. Commitment grows when people recognise themselves in the decisions they are asked to implement.

These are not incidental features of collective intelligence.

They make it possible.

A group that receives an excellent AI-generated analysis may become better informed without becoming more capable of acting together.

Conversely, a group that struggles through difficult conversations may emerge with a less polished document but a far deeper understanding of its collective responsibilities.

This is not an argument against technology.

It is an argument for recognising the human work technology should support, not replace.

The future of collaboration will require us to become more sophisticated in the way we combine human and artificial intelligence.

We will certainly need better tools. But we will also need better meeting design, more thoughtful facilitation and stronger architectures for shared learning.

Above all, we will need to resist the temptation to confuse faster interpretation with better understanding.

Thinking better together

The promise of AI is not that it will finally allow us to avoid the difficulties of human collaboration.

Its greatest promise may be that it can help us engage with those difficulties more intelligently.

It can expand what we can notice and connect. It can challenge our assumptions and support our collective memory.

But it cannot relieve us of the responsibility to listen, disagree, interpret and decide together.

That responsibility remains human.

As AI becomes more deeply embedded in our organisations, the central question may no longer be how much intelligence we can bring into the room.

It may be whether we are designing the conditions in which that intelligence, human and artificial, genuinely strengthens our capacity to understand one another.

Because collaboration is not simply about arriving at the same answer.

It is about developing the shared understanding, trust and agency required to act together, even when our perspectives remain different.

AI may help us make sense of more information. But only together can we decide what that information means for the world we want to create.

And that, ultimately, is what Thinking Better Together is about.

References and further reading

Ying, L. (2026). Organizational Mechanisms in Digital Public Health Transformation: Change Communication, Collective Sensemaking, and Cross-Functional Collaboration. Frontiers in Public Health. Accepted 20 October 2026; final formatted version pending.
https://doi.org/10.3389/fpubh.2026.1978109

Comes, T. (2026). Sensemaking AI: Introducing a Research and Design Agenda for Human–AI Networks. EPJ Data Science, 15, Article 42.
https://doi.org/10.1140/epjds/s13688-026-00634-5

Parisi, A., Thain, N., Hallak, A., Tsai, V., & Qian, C. (2026). Real-Time Group Dynamics with LLM Facilitation: Evidence from a Charity Allocation Task. ACM FAccT 2026.
https://doi.org/10.1145/3805689.3812253