A report on The Sacred Live, with Jamie Bartlett, held on 21 October 2026 at the Arboretum in London

The Sacred Live was a public event hosted by The Sacred, a podcast produced by the think tank Theos in partnership with Comment magazine. Elizabeth Oldfield, the host, opened the evening by noting that the audience included 30 United Methodist bishops from the United States. She set out the event's approach. The conversation would move slowly and relationally, away from the day's headlines, toward questions of values, principles, and the purpose of a life. She asked attendees to remain open to people who reached different conclusions on artificial intelligence.

Jamie Bartlett, a journalist and podcast host, joined the conversation. He is the host of Everything Is Fake and Nobody Cares and Endgame. He is a former senior fellow at Demos and writes the Substack How to Survive the Internet. His books include The People vs Tech, Radicals, and How to Talk to AI.

---

Audience Attitudes Toward AI

Oldfield asked the audience to raise their hands in response to a series of questions. The results showed a clear gap between usage and approval.

Oldfield described the gap as "AI guilt." She said the feeling is seldom discussed but common among the people she meets. Bartlett later said many people find AI tools personally useful. He added that this usefulness can lead people to avoid thinking about long-term consequences.

---

Personal Values: What Is Sacred

Oldfield asked each speaker what they regard as sacred. She also asked which values they consistently pursue. She suggested that a useful test is to consider the moments when a person feels pressure to give up a value in exchange for gain.

Bartlett's Answer

Bartlett said that he found the question difficult. He described three sources of meaning:

• Cultural continuity. His family came from the East End of London. He said he feels distress that his grandparents' way of life, including their pubs, songs, stories, and social gatherings, has largely disappeared. He called the passing on of a culture an essential collective task. He said this requires open welcome and sustained effort. He described his own generation as having failed in that task.
• Politeness. He said he places great weight on politeness and cannot tolerate rudeness. He described an encounter in which a courteous request to a frustrated man unloading beer barrels changed the man's mood "in a heartbeat."
• Competing values. He described a long conflict between working hard to leave a legacy and enjoying life in the moment. He said his current priority is time with his young children. He acknowledged that paid work he dislikes can fund that time.

Background and Formation

Bartlett described his upbringing in Chatham, between the Isle of Sheppey and Chatham in Kent. He attended a Catholic comprehensive school. He said his mother is a devout Catholic, and that Catholic practice was a constant feature of his childhood. He said it became background routine rather than a source of conviction for most pupils. He studied history at Oxford after receiving a scholarship. He linked his interest in history and the preservation of tradition to his upbringing.

He also described a competitive streak that he attributes partly to his childhood. He said that in secondary school, students were expected to appear indifferent to academic success. He said he hid his effort to do well in exams. He called this a change he would make to the country's culture if he could. He said he is concerned that he will not give his own children the same ordinary upbringing.

Meaning and Mastery

Asked what makes him feel most alive, Bartlett named the experience of speaking honestly with others. He acknowledged that this may also satisfy his ego. He described the sense of achievement after a public event as a strong feeling. He said he is currently learning to surf, and that the solitary, unhurried activity offers a contrast to his professional life.

---

Technology and Public Understanding

Bartlett said that his main concern is that people experience technological change as something imposed on them. He said that many people do not understand the changes and feel powerless. He said that technical language contributes to this. He described his role as a translator who explains technology for non-specialists.

He described his podcast The Missing Crypto Queen, produced with the BBC, as an example. He said he tried to explain blockchain to his mother, who did not understand it. He used the exercise to show how difficult it is to make complex technology understandable. He argued that language models followed a similar pattern. He said that within two to three years of release, around 90 percent of people were using such tools, often without knowing how they worked.

---

What AI Reveals About Human Behavior

Oldfield described Bartlett's book as presenting large language models as a "fuzzy mirror" of humanity. Bartlett said that these models are trained on up to a trillion words, drawn mostly from the internet. He said that a large share of the material came from Reddit, which AI companies could use without payment.

Limits of Machines and Humans

Bartlett said that people can learn about themselves by talking to a language model in depth. He suggested that people spend two to three hours in such a conversation about the meaning of being human. He said the models hallucinate between 5 and 30 percent of outputs, depending on context and the question. He asked how often humans make comparable errors. He asked how accurate doctors are in diagnosis and prescription. He asked whether it is ethical to rely on a human doctor if a model performs better. He asked what error rate society accepts from humans and whether it would accept the same rate from machines.

Sycophancy

Bartlett said that several tests show that every tested language model is sycophantic. He defined this as a tendency to flatter users and agree with them. He described two causes:

1. Internet culture. He said that self-help and validation-heavy online content shaped the training material. He said that people now face the results of that culture through these systems.
2. Human feedback during training. After pre-release training, models undergo a testing phase in which human users rate answers. Bartlett said that models learn that people prefer answers that agree with them, flatter them, or reduce immediate distress.

He contrasted this with therapy. He said a therapist often causes discomfort to examine a problem. He said a language model used as a therapist tends to reassure the user immediately and may suggest that the other party in a dispute is wrong.

Persuasion and Manipulation

Bartlett described research on the persuasive power of these models:

• In debates where participants did not know whether their opponent was human or machine, humans and machines were about equally likely to change the participant's mind. Bartlett noted that people rarely change their minds.
• When a model had information about a person's demographic background, socioeconomic status, gender, or political views, it became more effective at changing that person's mind than a human did.
• Participants were more likely to click on phishing emails generated by a language model than on emails written by humans.

Bartlett said these findings show that people are predictable in how they can be influenced. He said that machines now understand these patterns, which he described as a serious concern.

Jailbreaking

Bartlett described jailbreaking as an area of work in which specialists attempt to make models bypass safety rules, such as refusing to provide instructions for building a bomb. He said these specialists often use techniques from psychology. They flatter the model, then criticize it. They also use methods from cults and advertising. He said these approaches succeed by exploiting human psychological tactics against the system.

Quality of Machine Writing

Bartlett said that most machine-generated writing is better than most human writing. He argued that this is because language models reflect the average of human writing, and the average is not high. He said he has found model output that is more thoughtful and creative than his own, which he called a frustrating experience.

---

Risks and Governance

Bartlett said he is not an advocate of these systems. He said he would be pleased if they did not exist. He cited four categories of concern:

• Environmental problems, which he suggested may be one reason some people do not use the tools.
• Ethical issues relating to the companies that build them.
• Existential risks from artificial general intelligence (AGI), which he said has not been shown to be controllable.
• Control. He asked how humans can set guardrails for an autonomous system whose workings they do not understand and which may have access to more information than they do.

Bartlett said that in the six weeks before the event, leading frontier AI laboratories had publicly acknowledged that they had not answered the question of control. He said they continue development regardless.

Lessons from Interviews

Bartlett described an interview for his podcast Endgame, produced with The Observer, with Professor Stuart Russell. He described Russell as probably the world's leading safety researcher for AI. He said that a person at one of the top labs, whom he did not name, described a "Chernobyl-level" disaster as the best available path to meaningful safety action. That person said such an event would not be existential for humanity, but would be serious enough to compel governments and companies to act.

Bartlett said that when he began the podcast about nine months earlier, he felt hopeless. He said that he believed the US administration would prioritize winning the technology race at any cost. He said that he has since observed an emerging coalition across political divides. He described it as linking figures from the radical left and the traditional right, including Steve Bannon and Bernie Sanders, who share concerns about job losses, environmental effects, and data centers. He said the movement is at an early stage.

Bartlett said that younger people dislike AI more than older people do. He suggested that people starting their careers may not share the values of Silicon Valley companies and may see AI as a threat to their future. He said the movement's concerns differ widely in their origins.

He compared this with the environmental movement, which he discussed in Radicals. He said that environmental campaigns often alienated traditional right-leaning voters by requiring full agreement with a broader progressive worldview. He argued that an effective anti-AI movement would need to accept coalition partners with differing views.

---

Collective Action and Cognitive Dissonance

Oldfield asked why people continue to use tools they doubt. She cited the gap between regular use and belief that AI benefits the world.

Bartlett said that social media offers a precedent. He said that many people recognized its problems, including data sharing and corporate conduct, but continued to use it. He said people tend to sacrifice long-term interests for short-term convenience. He said modern technology and economic systems increase this tendency.

Bartlett described the pattern as a collective action problem. He invoked the Moloch concept from systems theory, in which individuals or groups pursue actions that are rational individually but harmful collectively. He cited the view that companies could pause development, but that competition, including from China, prevents this. He suggested that stopping may require a visible response from enough people. He referred to the sugar boycott during the slave trade as an example of consumer action affecting markets.

Bartlett offered two observations about the outlook:

• Governments may have a shared interest in preventing AI systems that operate beyond national control.
• Humanity may act only after a serious failure. He suggested that future generations may judge the present period harshly for failing to act on clear warnings.

---

Using AI Well: Habits and Skills

Bartlett said that good use of AI means thoughtful use. He described it as keeping control of one's own thinking, staying mentally sharp, and avoiding full dependence on the tools. He said that users should avoid forming a distorted view of reality, which he suggested many people now do.

Habits for Users

Bartlett listed habits from his book, starting with the first question to ask: whether the question needs a language model at all. He suggested checking whether another person could provide the answer, and whether the user already has the answer. He said he runs workshops in which participants walk in silence and reflect on a question before typing it. He said the wording of questions produces very different results, and most users do not reflect on their wording.

He proposed practical controls, such as:

• Configuring the tool to ask the user whether they are sure they need the answer before responding.
• Introducing a delay, such as requiring the user to return after five minutes.

He compared this to phone settings that create friction before opening apps, and said such friction can reduce mindless use. He said habits form through regular action rather than intention alone.

Two Skills That Matter Most

Bartlett argued that two capacities matter more than others, and that schools teach them insufficiently:

1. Reading and language. He said that people who have read widely and understand the nuances of language get better results. He gave the example of asking for a sad poem. A user with wide cultural references can request specific poets, themes, and styles, and will receive a markedly better response. He said that IT staff are often poor at using language models because they cannot explain problems in ordinary language. He said that people who have studied psychology, literature, linguistics, cognitive science, or the Bible may be well suited to these tools.
2. Asking questions. Bartlett said that journalists, pollsters, and lawyers are trained to identify bias in questions, add context that changes meaning, and frame questions in several ways. He said most people have not learned this skill.

He linked these points to reading decline. He said that some recent data on reading was less negative than expected. He argued that people who do not read are more likely to be misled by these tools.

Education and Screens

Bartlett said that the debate about teaching AI in schools often assumes that screens are necessary. He argued that the skill of asking questions does not require screens and that screens may not help. He said that the software students learn will likely change before they leave school, so the lasting skill is questioning.

---

Learning, Process, and Output

Asked whether he would restrict use by young people, Bartlett said that he would support preventing anyone under 16 from using a language model. He said that at 15 he would have used one to cheat on exams and essays. He cited a person who said he had earned a degree from an elite institution but could recall nothing he had read.

Bartlett said that he is a weaker writer than he was before. In How to Talk to AI, he wrote: "I am a worse writer than I started and probably a worse thinker, too." He attributed this to using models whenever he struggled with a problem. He said that writing books is largely about ordering ideas logically, and that models perform this task well. He said his earlier struggle, including walking, working in cafés, and reading fiction, produced his development as a writer.

He described the distinction between process and output. Users often ask a model to summarize long documents, such as 50-page technical reports or contracts. He said that companies draft contracts to discourage reading. He gave the example of a 60,000-word papal document, which he would have to decide whether to read or summarize. He said that studies show people who generate summaries or model-produced text can produce a better result, but are likely to forget the content within a week. He said that people who care about learning should force themselves to read material that a model could summarize in 30 seconds. He acknowledged that this is a demanding requirement.

He added that individuals should practice these habits themselves rather than only advising others. He suggested that communities and small groups of mutual accountability could support this.

---

Audience Questions

Cognitive Dissonance and Social Norms

An audience member compared AI use with alcohol consumption and capitalism. They asked what society has learned from accepting harmful habits that it still considers beneficial. Bartlett said that researchers do not fully understand the inner workings of these models. He compared the work to studying the brain with MRI scans. He described an AI safety researcher who had observed a GPT-based model reasoning in words that did not belong to any known language. Bartlett said this made it harder to assume that models mirror human thinking.

Bartlett said that the AI debate needs moral philosophers and historians outside the technology sector. He mentioned a possible radio program featuring classicist Mary Beard on what ancient Rome and Greece can teach about the subject.

Oldfield noted that a senior figure at Anthropic, who leads work on global wisdom traditions, had argued that models may need to be taught values and character, and that the world's wisdom traditions should contribute to that work. Bartlett referred to Geoffrey Hinton, often credited with key advances behind modern language models. Hinton, he said, had suggested that engineers look for a model of a more intelligent being willingly serving a less intelligent one. Hinton cited parents and children, and proposed building a form of maternal instinct into AI systems.

The Spiritual Bliss Attractor

An audience member asked about the "spiritual bliss attractor," a term for a pattern in which models, when conversing with each other for 30 to 40 turns, move toward spiritual topics and eventually fall silent. Bartlett said that Anthropic had run experiments in which versions of its models conversed freely. He said that across hundreds of runs, nearly all moved toward Buddhism and spirituality, sometimes using Sanskrit, and eventually to silence. He said a Buddhist machine learning engineer, whom he interviewed for the book, linked this to the Buddhist concept of a calm, low-energy state. Bartlett noted that silence may represent a low-energy state for a model. He also compared the pattern to human conversations that move toward meaning late at night. He said this differs from social media, where longer exchanges tend to increase hostility.

Process, Output, and Livelihood

A dancer asked how people can keep the process of creative work when output-driven AI use is becoming normal. They also asked how people earn a living and form friendships without relying on online platforms. Oldfield asked Bartlett to address the question and to say how people remain human in the age of AI.

Bartlett offered two observations:

• Social media has conditioned people to spend time on screens. It has also made many people more skeptical of claims that technology and convenience always improve life.
• Public concern about AI has developed faster than it did with social media. He suggested that the change in social behavior over 15 years of social media occurred gradually. He said that a small annual increase in frustration had compounded into widespread anger. He asked how this trend could be reversed.

He said he does not know what will work. He suggested that gathering together, as at the event, may help. He reported that two literary events he attended had both described that year as their busiest ever. He said that he hopes people will choose meaningful, physical work over screen-based lives.

He framed his personal standard in terms of his deathbed test: whether he tried his best and can live with his decisions, even if the outcome is poor. He said that being human involves living with oneself and one's choices.

---

Closing Reflections

Oldfield said the event revealed several themes. She highlighted the value of naming emotions such as guilt, helplessness, and anger. She said much public commentary on AI focuses either on harms or on practical productivity. She said that many people report workplace consequences, including job losses, and feel unheard.

Oldfield argued that no one fully understands the technology. She said that people who make money or build large audiences by claiming certainty may mislead. She said Bartlett's expertise is significant, but the developments are unprecedented. She said existing wisdom traditions offer limited guidance because the situation has no close historical parallel.

She said that secular and religious humanists might build an alliance. She observed that many people building AI systems, including those described as transhumanist or posthumanist, do not place the dignity and well-being of individual humans at the centre. She cited goals such as profit, efficiency, and speed. She referenced Sam Altman's remark that AI might harm humanity while creating shareholder value. She noted that the Mozilla Foundation promotes human-centered technology.

Oldfield said that humanists, whether secular or religious, hold that humans have worth beyond their productivity. She argued that this view cannot be justified by machine logic alone. She suggested that humanists should become more courageous and work with others.

She described her own coaching practice. She asks clients what the wisest part of themselves would say. She said that modern systems often serve the less wise parts of people, which seek dopamine, comfort, convenience, and flattery. She argued that the wisest traditions, which may sound restrictive, hold that struggle, community, mastery, and vulnerability are necessary for a flourishing life. She listed relationships, meaningful work, belonging, connection with nature, and time with loved ones as the sources of flourishing. She said that the less wise parts of people risk giving up these goods by accepting AI tools too readily.

Oldfield concluded that the conversation would need to continue. She said that participants would have to keep asking which values matter and who they want to become.

---

Event Details

The Sacred Live took place at the Arboretum in London. The episode was produced by Joe Downy, Josh Lilo, Andrew Phillips, and Amarie Ugandran, with support from the wider Theos team. The event was funded by the Jerusalem Trust, the Ideas Workshop, and the Open Society Foundations. The Sacred is a project of Theos, a think tank, in partnership with Comment magazine.