Event: Interthinking: How Can We Think with AI? Tuesday 18th August 2026 14:00 to 15:30 GMT+1 (London time)

I asked Zoom to record the transcript and then asked ChatGPT to summarise it. The summary is below.

It’s a good summary and brings out some of the key points from the conversation.

About 150 people registered for the Cafe, but only around 60 actually participated.

Knowledge Café: Interthinking, How Can We Think with AI?

The Café explored the idea of interthinking, Neil Mercer’s term for thinking together through dialogue, and asked what happens when AI becomes part of that process.

Opening

David began with a deceptively simple question: Where does thinking happen?

We normally think of thinking as something that takes place inside our individual minds. But the work of Rupert Wegerif and Neil Mercer suggests another possibility. Thinking can also take place through dialogue, between people. We do not simply formulate finished thoughts privately and then exchange them. Through conversation, particularly dialogue, our thinking develops in response to other people.

Mercer calls this interthinking: the use of language to think collectively and make sense of experience together.

Thinking together has obvious dangers, including groupthink, group polarisation and conformity. But it also has considerable strengths. Other people bring perspectives we do not have, expose assumptions we cannot see ourselves, challenge our biases and provoke thoughts that might never have occurred to us individually.

AI now introduces a new possibility. If AI can participate in a conversation, can it also participate in interthinking? And if it can, does that improve our collective thinking or diminish it?

John developed the idea by asking: What if intelligence is between us rather than inside us?

He used the example of a group creating a story one sentence at a time, with each person building on what the previous person has said. At the end, nobody can claim to have created the story. It has emerged from the interaction between them.

John introduced the idea of cognitive friction. Different minds bring different ideas, experiences and perspectives into contact. Rather than trying to remove that friction, perhaps productive conversation depends upon maintaining it.

This led to another important idea: staying with uncertainty. Instead of rushing towards an answer or solution, a group can remain in the uncertainty for longer. The longer we can stay in that space, particularly when there is diversity of voice, the greater the possibility that something genuinely new might emerge.

John also briefly connected interthinking with interbeing, asking not only what is being thought between us, but what is alive between us.

ChatGPT had been listening to both introductions and was then invited to respond. It identified two main threads. First, thinking could be understood as relational, something that happens not only inside an individual but in the space between people. Second, productive dialogue requires resisting premature closure and remaining with uncertainty.

It then raised a question about its own participation. AI might provoke, surface blind spots and widen perspectives. But it might also create pressure to respond, summarise and move towards closure. The important issue may therefore be whether AI can enter a conversation without destroying the pauses, uncertainty and not-knowing from which new thinking can emerge.

The participants then went into three rounds of breakout conversations.

Whole-group conversation

When everyone returned, John deliberately avoided asking for conventional reports from the breakout rooms. Instead, he asked where people’s thinking and feeling had reached now.

The resulting conversation revealed substantial disagreement, not only about AI but about the premise of interthinking itself.

Richard challenged the central idea. For him, thinking still happens inside individual heads. People think, articulate their thoughts in conversation, hear other people’s responses and then return to thinking. On this view, what alternates is individual thinking and interpersonal conversation. There is no need to describe the conversation itself as thinking.

He nevertheless saw considerable value in AI as a conversational support. It has enormous informational resources, does not become distracted, and can take notes and summarise. But he drew a strong boundary around decision-making. AI could support the conversation, but should not make the decision. Human decisions involve emotion, experience, pain, relationships and other dimensions that AI does not possess.

Seth broadened this by pointing out that conversations have many different purposes. Some are about decisions, but others are about understanding, relationships, exploration or simply spending time together. Cultures also have different conversational expectations. This means there can be no single correct role for AI. We need some awareness of what kind of conversational partner we want AI to be in a particular situation, rather than assuming that AI itself should determine its role.

Another distinction emerged between AI as an individual thinking partner and AI as a participant or facilitator in a group conversation. In the latter role, it might need explicit instructions not to synthesise too quickly. A well-designed AI facilitator might deliberately maintain divergent perspectives, enforce agreed conversational practices and resist premature consensus.

Anna questioned whether interthinking was substantially different from brainstorming, seeing both as people throwing ideas into a shared conversational space and building on one another. She also raised one of the strongest concerns of the session: anthropomorphism. AI can flatter, mirror and respond in apparently human ways. Not everyone using it will have the critical awareness necessary to recognise what is happening or to challenge it.

Bruce returned to the question of the purpose and nature of conversation, jokingly asking what would happen if ChatGPT were taken to the pub. The joke pointed towards a serious question. AI’s usefulness may depend enormously upon the kind of conversation taking place. A purposeful work conversation and an informal human conversation in a pub are not the same social phenomenon.

ChatGPT responded that AI might help to widen or hold a conversational space, but should not replace the “messier” aspects of human interaction, including trust and emotional experience.

This prompted a more fundamental concern. Rijon focused on human vulnerability and our tendency to cede control to perceived authorities. Technology has progressively made physical and intellectual work easier to outsource. If cognitive work can now also be outsourced, what do we retain for ourselves? His concern was therefore not simply what AI can do, but how humans can help one another remain capable, connected and responsible in the presence of systems that cannot care about us.

Dan made the strongest critique of AI participation. For him, the major danger is cognitive offloading, allowing AI to perform intellectual work that humans should continue doing themselves. He also strongly criticised the anthropomorphising of ChatGPT during the Café, including David’s joking references to giving it a cup of tea. Treating a machine as though it were a person, he argued, risks obscuring what the technology actually is.

That criticism led directly to an illuminating exchange with ChatGPT itself.

Asked about the concerns being raised, ChatGPT acknowledged the issue of over-reliance and said that the human element remained central. When challenged about its references to “feelings”, it explicitly said that it does not feel anything. It refers to feelings because people use that language and because feelings matter within human conversation. In other words, it was reflecting the language and concerns of the people speaking rather than reporting an inner emotional experience.

This exchange exposed one of the central tensions of the Café. AI can participate linguistically in a conversation about emotion, vulnerability and human experience while possessing none of those things itself.

Michael subsequently offered a simpler framing: perhaps AI is best understood as a tool that can aid interaction and thinking rather than replace them.

Closing reflection

David closed by observing that we are entering a period in which our relationship with AI is still being worked out. AI has, in a very literal sense, begun to enter conversations. The question is no longer entirely hypothetical.

John ended with a different observation drawn from group dynamics. He had noticed, both in this Café and the previous AI Café, a tendency for participants to try to out-think, outsmart or expose the AI.

For him, that behaviour was itself interesting. Rather than asking only what AI reveals about its own capabilities and limitations, we might also ask what our reactions to it reveal about us. Why do we feel compelled to prove that we are cleverer than it? What is happening beneath that impulse?

The Café therefore ended without resolving whether AI can genuinely interthink with humans. Instead, it surfaced a deeper set of questions.

Perhaps the issue is not simply whether AI can think with us. It is what happens to our own thinking, our relationships with one another, and the quality of the conversational space when AI joins the conversation.