We Don't Need More Conversations. We Need Better Ways of Thinking Together.

For the past five years I've been building an experiment in how institutions think together. People have described it as a conference series, a collection of ThinkTalks, symposiums and summits, or a consultancy service. Those have simply been the vehicles.

At Just Gold we've put thousands of hours and a significant amount of our own money into it, not because it was commercially obvious but because Australia needs better ways of approaching what policy scholars call wicked problems. Social cohesion, disability inclusion, climate resilience, mental health, regional development: none of these yields to a single report, a single expert or a single organisation. Every intervention changes the system. Every answer produces new questions.

Our methods for addressing them have barely changed. We convene a conference, we hold a consultation, people tell their stories, someone writes a report, and six months later another consultation begins as though the first one never happened. We have become very good at asking communities what they think, and much less successful at helping institutions learn from what they hear.

Australia has never suffered from a shortage of consultation. It suffers from a shortage of institutional learning.

That is the problem THINK was built to solve. Each iteration has been another attempt at the same question. Can a conversation be designed to produce more than discussion? Can it generate collective intelligence, and can that intelligence be held in a form an institution can actually use?

The Think Social Cohesion National Summit in Parramatta was the most ambitious attempt so far. Size had little to do with it. What made it ambitious was who was in the room, and the moment they were in it: social cohesion is being argued over in this country with a heat it has not carried for years, which makes a room like that harder to convene and more consequential to get right. The summit itself was not the innovation. The methodology was.

The design starts from a different premise about what a room is for. Rather than treating discussion as something to be summarised afterwards, we treat it as evidence. Rather than smoothing over disagreement, we preserve it. Rather than recording consensus where none exists, we document where thinking converged, where it diverged, and where it remained unresolved.

That is easier to assert than to demonstrate, so here is what it looked like in practice.

The room.

The opening session asked a large room what the system owes people. The room answered, as rooms do, with principles. A fair go. Safety. Access and equity. Early intervention. A conventional record writes that up as shared commitment, and it would not be inaccurate. Nobody disputed a fair go.

But that agreement was empty. Specify a fair go as equality before the law with a remedy attached and it is contested immediately, which is precisely why this country does not have one.

A fair go is easy to endorse because it does not specify who owes what to whom. The floor the room agreed on was uncontested because nobody can stand on it.

That is not a quibble about wording. Within an hour, unprompted, a room of senior people had converted its rights into principles and lost the duty-bearer in the process. The panel then spent much of its session converting them back, and the difference between the two registers turned out to be the most useful thing that happened all day.

That distinction does not survive a notetaker. A summary would have recorded a shared commitment to fairness and safety, and the substance would have vanished into a bullet point as unspecific as the phrase it came from.

Artificial intelligence made that possible, and it is worth being precise about how, because the claims made for AI in public participation are mostly overstated. It did not replace anyone's judgement. What it did was hold the contributions of an entire room simultaneously, which one facilitator cannot do, identify the tensions the panel had not yet touched, and return them while the thinking was still live rather than after question time had flattened it. Every contribution carried the same weight regardless of who submitted it.

It also has to be able to be wrong in public. By the end of the day the record held four live contradictions, and on review one was withdrawn. It was not a contradiction at all but a category error in the analysis, and once it was separated properly the room turned out to be describing a sequencing problem the panel had already resolved. Three stood. The fourth stayed in the document as a worked example of a mistake this sector makes constantly. A record that cannot correct itself is not evidence, and any synthesis method that never reports its own errors should be treated with suspicion.

What this produces is a different model of public participation, not a better way of writing up the old one. The distinction rests on a single idea.

Consultation is not complete when people have spoken. It is complete when they can see what changed because they spoke.

That is why the day was designed to produce a roadmap rather than a report. It carries named signatories against it, and it gets revisited in six months. That is the only test that counts, and it is one we have set for ourselves. If nothing has moved by then, the method will have failed on its own terms, and the record should say so.

Every innovation reaches a point where methodology is no longer enough.It needs someone willing to trust it before it has become conventional. For five years THINK grew one conversation at a time, refined because we believed in it rather than because anyone was asking for it.

When Elfa Moraitakis began imagining a national conversation on social cohesion, she was not looking for another conference. She was looking for a different way to convene one, and she trusted us to design the architecture rather than simply run the program. What made that possible was that she chose not to outsource the thinking. The design sessions were the two of us, month after month, and most of that time was spent disagreeing. The first question was whether this should be a forum or a summit, and we went back and forth on it for a long time, because the answer determines everything that follows. A forum gathers people. A summit is built to produce something. The sector was calling for a forum. The question we kept returning to was not what the sector was calling for, but what it needed.

Elfa Moraitakis.

SSI the Settlement Council of Australia and Special Broadcasting Service (SBS) Australia backed the same experiment. So did every speaker, panellist and participant who treated the day as a problem to solve rather than an event to attend.

I'm used to building problem-solving conferences alone. This was the first time I didn't have to. That turns out to matter more than any part of the methodology, and it is the argument in miniature: a wicked problem is not something a single organisation can hold, and neither is the work of convening people to think about one.

The summit will not solve social cohesion. No summit can. But somewhere between the first ThinkTalk and this one, THINK stopped being a series of events and became an architecture: a way of holding lived experience, professional expertise, community knowledge and emerging technology in the same room without reducing any of them to consensus, and of preserving complexity long enough for insight to emerge from it. Not an architecture for running conferences, but an architecture for institutional thinking.

Kyriakos Gold, Violet Roumeliotis, Joe Ball, Maria Dimopoulos, Jesse Fleay & Hugh de Kretser. (L-R)

First published on LinkedIn, 28 July 2026.