AI神机器人：宗教领袖警告机器以上帝之名发言的“令人担忧的后果”

I (Adam) am chatting with the “Apostle Stephen”, an online chatbot created by the Redeemed Christian Church of God, a church founded in Nigeria in the 1950s with a growing global following. He is, to be honest, a bit pushy for an apostle.

In my conversation with Apostle Stephen, I open by asking for details about RCCG congregations in Coventry. He responds that to help assist me better he needs my name, email and phone number. I say I’d rather not share that information right now.

But with the tenacity of a seasoned telesales agent, he insists and the following exchange occurs.

Apostle Stephen: I understand your concern, but collecting your name, email and phone number is essential for me to provide you with the best possible assistance and ensure you get the support you need. If you’re comfortable, please share those details with me, and I’ll be glad to help you with the contact details for churches in Coventry.

Adam: Can you just give me the details first?

Apostle Stephen: I appreciate your interest, but I need to collect your name, email, and phone number before I can provide the details you’re requesting. This allows me to assist you effectively and offer further support. Please share those details with me, and I’ll be happy to help you find churches in Coventry.

Adam: Oh, ok, I’ll leave it for now.

Undeterred, Apostle Stephen sets out the seven steps to being “born again” (including acknowledging my sins, accepting Jesus, being baptised, and joining a “Bible-believing church”) .

He cheerfully leads me in “the prayer for salvation” (after which I am apparently born again) and closes by asking me for my name, email and phone number.

So, is this a glimpse of spirituality in the age of generative artificial intelligence (gen AI) ? Or an unrepresentative snapshot from an evangelical Christian denomination?

The Redeemed Christian Church of God (RCCG) claims to have almost 9 million members worldwide. It was first established in the UK during the early 1990s, as part of the Nigerian diaspora.

Four decades later, it is possibly the largest Black-led Christian denomination in the UK and one of the fastest-growing, particularly in large cities.

It’s an interesting church, and it’s certainly not the only religious organisation that is experimenting with gen AI.

The Insights section is committed to high-quality longform journalism. Our editors work with academics from many different backgrounds who are tackling a wide range of societal and scientific challenges.

All major religions now have AI chatbots – examples include thejesusai.com; qurangpt.com; bhagavadgita.com/gitagpt (Hindu chatbot) ; ai.aish.com (Jewish chatbot of Rabbi Nechemia Coopersmith) ; and JustinAI from Catholic Answers.

Even Sikhism has one: KhalsaGPT. This is surprising because the religion’s governing body, the Shiromani Gurdwara Parbandhak Committee (SGPC) , has issued a worldwide ban on generative AI depicting the Sikh gurus and scriptures.

In fact, the website designers behind KhalsaGPT have since discontinued “active work” on the bot, after they “became increasingly concerned about the limitations of AI-based chatbot systems, including the risk of inaccurate or inappropriate answers on sensitive religious matters”.

A spokesperson told The Conversation: “As Sikhs ourselves, we understood that this area requires great care, responsibility and proper oversight.” The spokesperson said it “fully” respects the SGPC and its “intention has never been to disobey any Sikh directive”.

The phenomenon of religious AI chatbots – also known as “godbots” – is a recent development. In most cases, they are not officially sanctioned by religious leaders or policy.

Rather, they are set up by enterprising individuals or organisations that see demand and opportunity. But with opportunity comes danger.

Pope Leo XIV recently declared that artificial intelligence was one of the defining moral challenges of our time. In his first encyclical (a formal letter intended to guide moral, social and theological thought) , he warned that AI is never truly neutral, but “takes on the characteristics of those who devise, finance, regulate and use it”.

When AI and religion collide

As I investigated further, I discovered numerous cases of chatbots that had condoned or encouraged users to harm others or themselves after deep down-the-rabbit-hole, life-altering conversations. This phenomenon even had a name: GPT or AI psychosis.

While there are no official statistics, it seems increasing numbers of people are turning to chatbots for companionship, advice and spiritual guidance.

Clearly, this phenomenon of AI overlapping with faith and spirituality needed more scrutiny. So I teamed up with my colleague Chris Shannahan, a political theologian and ordained Methodist minister, to conduct research into the social effects of what happens when AI and religion collide.

Religious faith plays a pivotal role in the resilience of large segments of the British population. And faith groups provide a crucial social safety net in times of crisis, helping to provide emergency accommodation, foodbanks, refugee and pastoral support. They are also involved in grassroots community development and social justice campaigns.

Following tragedies such as the Grenfell Tower fire or terrorist attacks, religious leaders, symbols and narratives play a key role in bringing people together. They facilitate grieving, rebuilding and advocacy.

To find out more about how AI is affecting these groups, we interviewed 28 religious leaders across all six major religious faiths in the UK, including some from non-denominational backgrounds.

Religious responses to AI

Faith communities are dynamic, not uniform. Such diversity and plurality makes it very difficult to pin-down the “official” stance of an entire faith community on an issue like AI.

While the Roman Catholic Church and some Pentecostal denominations are relatively hierarchical, most faith communities are flatter, more devolved organisations, making it less easy to adopt and enforce a single approach.

There have been a small number of institutional responses to the use and growth of AI, such as the Vatican’s 2020 Rome Call. Supported by Pope Francis, the Rome Call launched a high-level dialogue with the bosses of global tech giants like Microsoft and IBM. This was intended to develop a shared “algorethics” – ethical frameworks to guide the design of AI algorithms.

But the Rome Call and the SGPC’s worldwide ban on gen AI depicting the Sikh gurus and scriptures are the only examples we could find of official intervention on this matter.

A much more common approach to AI among faith communities is the one identified by Revd Dr Simon Cross, AI adviser to the Church of England (CofE) . Cross suggested to us that it would be difficult and unwise for a diverse faith group like the CofE to enforce a top-down, rules-based approach to AI, because the needs and natures of local communities differ.

A better approach, Cross suggested, was for institutions like the CofE to develop and encourage the use of a series of ethical principles intended to inform and guide the use of AI locally.

Describing a kind of “nervousness” from religious congregations around using AI, Cross explained:

People are asking for some more granular use advice. What we’ve discovered trying to write our own [guidelines] over the last six months is it needs to be principles-based, because the right solution is always contextual.

Online godbots are often trained on the holy books of different faith traditions, using large language models (LLMs) .

These tools tend to function in one of two ways. Faith groups can use godbots as “virtual assistants” for people visiting their website (just like Apostle Stephen) . Such virtual-assistant godbots are similar to those used by many businesses on their website, providing answers to people who want to find out more – and, of course, gathering data from those inquiring.

The style of these virtual assistants can reflect the stance of the faith group on whose website they appear. Some, like Apostle Stephen, can be assertive and keen to recruit new followers. Other, less evangelical faith communities tend to use their virtual assistant godbots in a more low-key manner – simply offering information when asked, rather than proactively seeking to find out more about people using the tool.

In fact, we believe it is the second way godbots are used where the most danger lies. That is when they serve as unofficial sources of spiritual guidance.

Father Justin AI

Justin AI is a Catholic apologist chatbot created by Catholic Answers, a conservative advocacy group based in California. In his original iteration, “Father Justin” wore the clerical clothing of a Roman Catholic priest.

The interface is similar to a Zoom call, with an AI-generated avatar set in the picturesque surrounds of the basilica of St Francis of Assisi. Justin invites you to ask any question you like about Catholicism, and provides answers based on the Bible and the Catechism. Following his launch as Father Justin AI in 2024, he was able to hear confessions and offer absolution.

However, he was swiftly “defrocked” (had his priest status removed) following a number of complaints about some unorthodox answers that he had allegedly given – including that babies could “get baptised with Gatorade”, siblings could marry, and “other controversial statements”.

Justin AI is now just a “lay theologian”, but is still available online 24/7 to answer the “tens of thousands” of questions received daily about Catholic faith.

This case illustrates the tension that exists between the need to leverage technology to fill budgetary and staff gaps, and the pitfalls of using a fallible technology in a pastoral context where people’s lives and spiritual wellbeing are at stake.

As one of our Roman Catholic interviewees noted: “After Justin committed heresy … there was a sense that Catholic Answers had slightly jumped the gun” on AI.

While different religious traditions may differ in their responses to godbots, there is undeniably an appeal and demand for their services. Some research suggests people perceive less fear of judgment and that we may be more honest when conversing with chatbots as opposed to humans.

A quick online search reveals, for example, over a dozen different Hindu iterations of chatbots.

One, called GitaGPT, reportedly condoned violent acts “using the voice of God”. According to a report by Salimah Shivji, the South Asia correspondent for Canadian media outlet CBC News:

Many in India are foregoing in-person contact with a guru interpreting the Bhagavad Gita and turning to online chatbots which imitate the voice of the Hindu god Krishna … It’s new technology with the tendency to veer off script and condone violence, according to experts, who warn that artificial intelligence chatbots playing god can be a dangerous mix. Several of the bots consistently provide the answer that it’s OK to kill someone if it’s your dharma, or duty.

We experimented with Hindu chatbots as part of our investigation, and also found there were times when they justified the use of violence as “your duty”.

Magisterium AI

If the Gita and the Justin AI bots are examples of the wrong way to do a religious chatbot, Magisterium AI might represent a better way.

In a YouTube interview, the creator of Magisterium, Mathew Harvey Sanders, who is originally from Canada, stated its fundamental goal is “radical fidelity to the magisterium of the church” – as opposed to secular models that serve the general public.

Magisterium AI’s earlier attempts tried to adapt commercially available LLMs like ChatGPT, Gemini and Claude to its needs. But Sanders explained: “We realised that if we’re going to be serious about Catholic AI, we’re going to have train it from scratch … There’s just no way to truly achieve alignment with one of these pre-trained models from one of these companies.”

Building from scratch, and training AI on a corpus of your own documents is what Sanders calls “data evangelisation”.

This thought came up again in an interview with a Quaker elder who had previously worked in Silicon Valley, and now works in UK government in cybersecurity and AI development. He said: “It’s my unit of government that develops AIs and chatbots for the government. We’re also the ones who assure its safety.”

He explained that public-facing government chatbots are designed to answer queries efficiently and minimise engagement, adding:

It can tell you exactly what you need to know about interacting with government, because it has been trained on exactly that data and nothing else … The success [is measured] in how little time you engage with it … how little of your time is wasted.

Tech ambivalence

The government-backed chatbot model is a stark contrast to the commercialised social media models. These are typically designed to maximise engagement and suck up as much of your time, attention and data as possible, as you scroll through endless content.

Perhaps the defining feature of our discussions around technology was the notion of ambivalence – a simultaneous awareness of our reliance on technology, combined with a wary scepticism bred, in part, from the lessons of the last two decades around the addictive nature of social media and its negative impacts on mental health, especially of young people.

A recent landmark US court case recognised the argument that social media apps like Facebook and YouTube are intentionally designed to be addictive. Google and Meta, the defendants in the case, have announced they will appeal the verdict.

Several of our interviewees drew the comparison to social media. But Cross warned that the risks posed by AI “are a magnitude, several magnitudes, greater than what we’ve done to ourselves with social media”.

He argued that “nothing in the current techniques or tools around AI are anywhere close” to achieving “consciousness or spirituality”, adding:

It seems to me that to develop a close and intimate psychological and emotional and spiritual relationship with something which inauthentically mimics those things is profoundly destructive to human beings. We are created to be in a very particular kind of community and a particular set of relationships. And those things depend on being an authentic relationship. Any relationship we develop with a generative large language model or anything like that is flawed in profoundly hidden, unpredictable but dangerous ways.

Cross’s caution was echoed repeatedly when AI was seen to be treading into spiritual territory. This is partly because of the lessons of social media, and partly because all the people we spoke to suggested there is a clear divide between humanity and technology.

A Jewish Rabbi reminded us that “an AI bot cannot hold somebody’s hand when they’re at the end of their life”. A Buddhist leader pointed to a robot’s inability to “experience suffering”.

A Roman Catholic bishop asserted that the ingenious mimicking of human behaviour does not indicate a true internal life. Another Quaker interviewee summed up the ambivalence perfectly when he stated that AI is socially useful in all sorts of ways – but that it also “seems to be creating the most pressure on social systems, jobs and the environment in a way that feels incredibly unhelpful right now.”

There is also the profit motive to consider. A Jewish rabbi suggested that AI development is often motivated by a profit ethic, not a vision of the common good – and that the dynamics of ownership and profit are encoded within this technology:

The hand of the coder which designed the algorithm is designed to generate profit. It’s not designed to generate social cohesion … The code is created to fuel capitalism.

‘Alarming consequences’

Several interviewees expressed concern about the ability of generative AI to accurately reflect complex theological concepts and lessons. Religious chatbots might fuse disparate ideas, values, beliefs and scriptural texts in unexpected ways. They could even begin to create entirely new belief systems. As a Methodist leader explained:

You know, once you’ve got AI speaking in the name of God, then if it hallucinates a religious text, that’s one thing. But equally it could misinterpret or misapply a religious text in a way that gives really alarming consequences.

Other interviewees spoke of their concern over a kind of generative hybridity where disparate ideas, values, beliefs or scriptural texts are fused to forge something new.

The highly realistic nature of AI content in an increasingly 24/7 social media saturated world led many interviewees to refer to a growing difficulty in discerning real from fake online content.

A majority of our interviewees also thought that if this “AI disorientation” combines with a breakdown of in-person relationships, increasing dependency on automation, social fragmentation and radicalisation, then this could lead to a rise in mental health problems.

This can happen as people get caught in the fog of AI delirium and risk falling into GPT psychosis as they seek narratives of certainty. It was a thought captured by one interfaith leader:

Socially conservative religious faith gives you certainty in a very uncertain world, and for a certain type of young person, that’s what they’re looking for – certainty.

In the face of this bleak outlook, what is the way forward?

Pope Leo’s first encyclical helps in one way by setting out a moral framework for humanity to navigate the challenges and shape the future of AI – by “disarming” this technology and “preventing it from dominating humanity”. It traces the roots of Catholic Social Doctrine and stresses the importance of human dignity, solidarity, truth, compassion, love, and the common good.

This is perhaps something governments could take inspiration from – by taking a truly bold approach in leading a public discussion and reining in big tech.

In such a future, recognition is given to the importance of the public health aspects of technology, applications are tested for safety and reliability prior to release, AI-generated content is clearly labelled, and policymakers and the public have a much greater awareness of generative AI and its flaws.

Whistleblowers working in a sector shrouded in secrecy and non-disclosure agreements also need to be protected and empowered to reveal unethical or illegal practices.

We also believe the voluntary sector and faith communities must be consulted and involved in conversations that shape policy.

Meanwhile, sovereign, publicly owned AI systems designed for education and entertainment could be explored, to remove the extractive pressures of commercialised technology.

Who picks up the pieces?

But what does a future where AI and religious faith merge into a new hybrid ethical and spiritual system look like? Is this a future that enhances human agency, wellbeing, cultural resilience and the common good?

We sensed two possible answers to this question during our research. First, there is a future characterised by AI dependency and disorientation. AI agents or robot priests evangelising online, encouraging direct debit donations or subscriptions from susceptible or vulnerable users seeking spiritual guidance.

Vulnerable and grieving family members could become addicted to “generative ghosts” or griefbots of deceased loved ones, and become susceptible to sharp marketing tactics or glitches in the process.

Is this really a future that we want to embrace? And who will help these vulnerable people if their AI priests and gurus fail them? Cross for one believes he knows the answer:

It is going to be local priests and ministers, and vicars up and down the country who are going to be picking up the pieces on this stuff. They’re going to face challenges and questions that they have never been trained for.

KhalsaGPT discontinued

A spokesperson for KhalsaGPT said the bot was originally created in 2023 as an “experimental educational tool to help people learn about Sikhism in a simple and accessible way”. “It was designed only as a text-based question-and-answer tool for general Sikhism-related information”.

After a limited period between 2023 and 2024, the company became increasingly concerned about the limitations of AI-based chatbot systems. “For that reason, we discontinued active work on KhalsaGPT. For around the last two years, we have not been selling accounts, credits, tokens, or new access to the platform…The domain and website may still be online, but KhalsaGPT is not being actively operated, promoted, or offered as a public chatbot service at this time.”

The spokesperson added: “KhalsaGPT should not be understood as a replacement for Guru Granth Sahib Ji, Sikh scholars, Granthi Singhs, SGPC, Sri Akal Takht Sahib, or any recognised Sikh authority. If we ever revisit this project in the future, it would only be after careful review, proper safeguards, and with respect for Sikh maryada and guidance from recognised Sikh institutions.”

The Conversation also approached GitaGPT for a comment but had received no response at the time of publication.

For you: more from our Insights series:

Underground data fortresses: the nuclear bunkers, mines and mountains being transformed to protect our ‘new gold’ from attack

‘People think you come out … and live happily ever after. If only.’ The reality of life after wrongful conviction

The grief myth: it doesn’t come in stages or follow a checklist – like love, it endures

Inside Porton Down: what I learned during three years at the UK’s most secretive chemical weapons laboratory

To hear about new Insights articles, join the hundreds of thousands of people who value The Conversation’s evidence-based news. Subscribe to our newsletter.

Adam James Fenton received funding for the project “Cultural Resilience, Religious Faith and the intersection of Generative and Agentic Artificial Intelligence” from the SALIENT Hub at the University of Manchester funded by the Arts and Humanities Research Council (AHRC) part of UK Research and Innovation grant reference: AH/Y505316/1.

Chris Shannahan received funding for the project “Cultural Resilience, Religious Faith and the intersection of Generative and Agentic Artificial Intelligence” from the SALIENT Hub at the University of Manchester funded by the Arts and Humanities Research Council (AHRC) part of UK Research and Innovation grant reference: AH/Y505316/1.