Bringing AI into our conversations is not just a technical development. It changes the social space in which we think, speak, listen, disagree, and make sense together. AI may help us converse better, but it also poses risks related to trust, power, consent, judgement, surveillance, authenticity, bias, and human agency.
The possibility of bringing AI into a live conversation is exciting. It can listen, summarise, retrieve information, notice patterns, ask questions, translate across languages, and help a group reflect on what is unfolding.
But we should not be naive about it.
When AI enters a conversation, it does not simply add another tool to the room. It changes the room. It changes what people notice, how they speak, who they trust, what they hold back, and where authority begins to move.
That does not mean we should reject it. I am actively exploring how AI can be integrated into a live group conversation.
But if we are serious about Conversational Leadership, we need to look carefully at both the dangers and the possibilities. A technology that can deepen inquiry can also close it down. A tool that can support collective intelligence can also weaken human judgement. A system that appears to help a group think can also quietly begin to shape what the group thinks.
Comfort, Trust, and Accountability
One immediate challenge is human comfort. Even AI note-takers can make people hesitant, especially when they know the meeting is being recorded, summarised, or turned into actions. The concern is not only privacy, but accountability. People may speak differently if they feel their words are being captured and may later be used to judge them.
This matters because good conversation often depends on a degree of safety. People need room to think aloud, change their minds, express uncertainty, and test half-formed ideas. If AI makes the conversation feel more formal, more exposed, or more easily audited, it may reduce the very openness that a good conversation needs.
So, if AI is to become a participant in our conversations, we need to pay close attention not only to consent and privacy, but also to trust. The question is not simply whether AI can listen, summarise, or contribute. The question is whether its presence helps create better conditions for thinking together or makes people more guarded.
The risk of surveillance
The most obvious danger is surveillance.
If AI is present in a conversation, people need to know what it is doing. Is it listening only? Is it recording? Is it transcribing? Is it summarising? Is the transcript being stored? Who can read it later? Can it be used for performance management, legal review, training data, or organizational memory?
These questions are not minor details. They shape how people feel in the room.
A conversation changes when people know they are being recorded or analyzed. They may become more cautious. They may say what sounds acceptable rather than what they really think. They may avoid humour, uncertainty, disagreement, or vulnerability.
In many organizations, this danger will not be evenly felt. Senior people may feel relatively safe. Junior people may not. People in precarious roles, minority positions, or politically sensitive situations may be especially careful.
So the question is not only “Can AI listen?” It is “What does its listening do to the quality of the human conversation?”
The problem of consent
Consent matters.
People should not have to guess whether AI is present in a conversation. They should not have to wonder whether their words are being transcribed, stored, analyzed, or shared.
Before AI joins a meeting, there should be a simple, explicit agreement on its role.
Is it there to support note-taking? Is it helping with facilitation? Is it being used as a knowledge resource? Is it producing a summary? Is anyone privately prompting it during the meeting? Can anyone ask for it to be paused or removed?
Without this clarity, trust is weakened.
Consent is not only a legal issue. It is a conversational issue. If people feel watched, managed, or subtly assessed, they are less likely to speak freely. And if they do not speak freely, the conversation becomes thinner, safer, and less useful.
The problem of authenticity
Conversation depends on trust. We need to know, broadly speaking, who is speaking, where a contribution is coming from, and whether people are being open with one another.
AI complicates this.
If a person uses AI to generate what they say, is that still their voice? If an AI produces a response that sounds human, should it be clearly identified as AI-generated? If AI can mimic tone, style, or even identity, where do we draw the line between support and deception?
This becomes especially serious when we think about deepfakes, synthetic voices, fabricated messages, or AI-generated dialogue that appears to come from a real person. Such uses can manipulate conversation and erode trust.
In Conversational Leadership, authenticity matters. This does not mean every word must be spontaneous or unassisted. We all prepare, draft, rehearse, and seek help. But if AI is shaping a contribution, especially in a sensitive conversation, some degree of transparency may be needed.
Otherwise, conversation risks becoming performative, manufactured, or manipulative.
The transfer of authority to the machine
A more subtle danger is that authority shifts to the AI.
At first, the AI is simply there to help. It summarises what has been said. It identifies themes. It suggests questions. It retrieves background information. It offers a possible synthesis.
But gradually, the group may begin to orient around it.
“What does the AI say?”
“What did the AI summary show?”
“What options did the AI identify?”
“What does the AI recommend?”
There is nothing wrong with asking these questions. The danger comes when they begin to replace more human questions:
“What do we think?”
“What are we noticing?”
“Who sees this differently?”
“What are we avoiding?”
“What feels unresolved?”
“What would be wise here?”
AI has a remarkable capacity to sound coherent. But coherence is not the same as understanding. A fluent summary can give the impression that the group has reached clarity when, in fact, important tensions remain unresolved.
This is one of the central issues in rethinking conversation in the age of AI.
The danger of premature closure
Good conversations often need time. People hesitate. They circle around an issue. They search for words. They contradict themselves. They change their minds. They say something tentative and only later realize what they meant.
This apparent messiness is not a failure of conversation. It is often where thinking is happening.
AI can easily interrupt this process by summarising too soon, naming patterns too quickly, or offering a neat synthesis before the group has done the deeper work.
Another risk is that the AI may enter the conversation too quickly, pulling people away from their own half-formed thoughts rather than allowing human creativity to develop. In some situations, a quieter role may be better: listening, capturing what is being discussed, and helping the group refine shared notes during the call.
The conversation becomes cleaner, faster, and more efficient, but potentially less thoughtful.
This is especially dangerous in complex situations. In such situations, the issue is rarely just that we lack information. More often, we are dealing with competing values, uncertain consequences, different interpretations, power dynamics, emotional commitments, and incomplete understanding.
A premature AI synthesis can make the situation appear more settled than it really is.
The flattening of disagreement
Disagreement is often uncomfortable, but it is also valuable.
In a good conversation, disagreement can reveal assumptions, expose weak reasoning, surface hidden tensions, and open the possibility of deeper understanding. But AI-generated summaries may smooth over disagreement in the name of clarity.
A summary may say:
“The group agreed that…”
Or:
“The key themes were…”
Or:
“The main options are…”
But what if the group did not really agree? What if a minority view was mentioned once and then disappeared? What if someone was silent because they felt unsafe? What if the most important point was not the dominant theme but the awkward exception?
AI is good at producing order. Human conversation often needs to protect productive disorder for a while longer.
If we are not careful, AI may help us avoid the very tensions we most need to face.
The reinforcement of bias
AI systems learn from existing human discourse. That means they can reflect, reproduce, and sometimes amplify the assumptions and biases already present in that discourse.
In leadership conversations, this matters.
AI may appear neutral while subtly reinforcing dominant perspectives. It may privilege certain forms of language, argument, evidence, or cultural expression. It may make conventional answers sound more reasonable than unusual ones. It may overlook quieter, more hesitant, or less familiar ways of making sense.
This is especially risky if AI is used to summarise, interpret, or facilitate a group conversation. What appears as a balanced synthesis may actually reflect the patterns and biases built into the system, the data it has drawn on, or the prompts it has been given.
So we need to ask: Whose views are being amplified? Whose are being flattened? What assumptions are being carried forward? What does the AI treat as normal, rational, or relevant?
If we are not careful, AI may narrow our conversations while appearing to expand them.
The weakening of human judgement
Another danger is that we gradually outsource judgement.
At first, AI helps us think. Then it helps us decide. Then it begins to frame the issue for us before we have properly framed it ourselves.
The danger is not that AI makes a decision. The danger is that humans stop noticing how much of their judgement has been delegated.
In organizational life, judgement is not simply the application of information to a problem. It involves context, values, experience, responsibility, timing, relationships, and consequences. It also involves living with the results of the decision.
AI does not live with the consequences. People do.
So AI may contribute to judgement, but it cannot replace it. The responsibility remains human.
The loss of conversational capability
There is also a longer-term danger.
If AI is always available to summarise, remember, prompt, challenge, and synthesize, what happens to our own conversational abilities?
Do we become less able to listen carefully?
Do we become less able to hold ambiguity?
Do we become less willing to sit with uncertainty?
Do we become less practiced at asking good questions?
Do we become less attentive to emotion, hesitation, silence, and body language?
These human capabilities matter. They are not soft extras. They are central to how people make sense together, especially in complex situations where the issue cannot be solved by information alone.
AI can support these capabilities, but it can also weaken them if we hand over too much of the conversational work.
The risk of performative conversation
If people know that AI is listening, analyzing, or summarising, they may begin to perform for it.
They may choose their words more carefully. They may become less spontaneous. They may try to sound clever, reasonable, strategic, or aligned. They may avoid half-formed thoughts because these might look foolish in a transcript.
But many important conversations depend on people being able to speak before they are fully clear.
A half-formed thought can open a new direction. A clumsy question can reveal a hidden assumption. A moment of vulnerability can create trust. A pause can matter as much as a statement.
If AI makes people feel that every word is part of the record, something essential may be lost.
Power in the room
AI does not enter a neutral space. It enters a social space already shaped by power.
Who decided to bring the AI into the meeting?
Who controls the prompt?
Who sees the transcript?
Who decides what gets summarised?
Whose language does the AI understand best?
Whose way of speaking does it treat as clear, rational, or relevant?
Who feels safe enough to challenge its summary?
These questions matter because AI can amplify existing power dynamics while appearing neutral.
For example, if a senior leader introduces the AI, others may feel unable to object. If the AI summary reflects the dominant view, minority perspectives may be further marginalized. If the AI is treated as objective, people may find it harder to challenge what it says.
Conversational Leadership requires attention to these dynamics. Bringing AI into the room does not remove power. It may make power less visible.
Exclusion and unequal participation
AI may help some people participate more fully. It can provide summaries, support people who missed part of the conversation, help with translation, and make complex discussions easier to follow.
But it may also exclude.
People who are less fluent, more reflective, more hesitant, less technically confident, or more concerned about being recorded may be disadvantaged. People whose speech patterns, accents, cultural references, or ways of reasoning are less easily recognized by the AI may find themselves misrepresented or overlooked.
The cross-cultural possibilities are real. AI may help people converse across languages and cultures. But translation is never just the substitution of words. Meaning depends on context, tone, history, humour, idiom, and relationship.
There is a risk that AI makes conversation appear smoother while losing some of what matters most.
This is why we should ask not only whether AI improves the conversation in general, but for whom it improves the conversation, and at whose cost.
Synthetic consensus
One of the more interesting dangers is what might be called synthetic consensus.
AI can take a messy conversation and produce a smooth account of what the group supposedly thinks. It can make difference look like agreement. It can turn uncertainty into themes. It can convert a living conversation into a polished summary.
This can be useful, but it can also be misleading.
A group may leave the room thinking it has reached shared understanding when, in reality, it has only accepted a plausible synthesis. The disagreement has not been resolved. It has simply been made less visible.
In this way, AI may not impose a decision, but it may create the appearance of collective agreement.
That is dangerous because real agreement requires more than a coherent summary. It requires people to recognize themselves in what is being said, to understand what they are committing to, and to have had the opportunity to question, challenge, and dissent.
AI should not become the host
The more AI becomes involved in conversation, the more important the human host becomes.
Someone needs to hold the purpose of the conversation. Someone needs to protect the human space. Someone needs to notice who is silent, what is being avoided, when the group is rushing, and when the AI is becoming too central.
AI may be able to support facilitation, but it should not replace the human responsibility for the quality of the conversation.
This becomes especially important when AI joins a live conversation as an active participant.
The facilitator, host, or convenor needs to decide when to invite the AI in, when to keep it quiet, and how to help the group interpret what it says.
The AI can contribute, but it should not take over the conversational field.
Some simple safeguards
If we are going to bring AI into conversations, we need some simple safeguards.
For example:
- make clear when AI is present
- explain what it is doing
- get consent before recording or transcribing
- allow people to ask for the AI to be paused or removed
- treat AI summaries as drafts, not final accounts
- ask who or what may be missing from the AI’s interpretation
- protect time for human reflection before inviting AI synthesis
- notice whether the AI is amplifying or diminishing particular voices
- question AI outputs for bias, framing, and hidden assumptions
- keep responsibility for judgement and action with the humans in the room
These safeguards do not solve everything. But they remind us that bringing AI into conversation is not just a technical act. It is a social intervention.
The deeper issue
The deepest danger is not that AI will join our conversations.
The deeper danger is that we forget what conversation is for.
Conversation is not merely a way of exchanging information, producing summaries, or reaching decisions faster. At its best, conversation is how we make meaning together. It is how we encounter difference, build trust, test judgement, share responsibility, and sometimes change how we see the world.
AI can help with that. But it can also get in the way.
The task is not to keep AI out of conversation. Nor is it to welcome it uncritically. The task is to learn how to use it in ways that deepen inquiry, strengthen human judgement, and preserve the relational quality of human dialogue.
AI may become part of the conversation, but the conversation must remain human.
Closing Reflection
In-person, 7–11 September 2026
Warbrook House, Hampshire, UK
We are living and working in conditions of uncertainty, complexity, and rapid change. This week-long workshop offers a space to practise Conversational Leadership as a shared, lived experience.