AI and Quakers: eyes open, not shut

2026-07-18

AI and Quakers: eyes open, not shut

What machine learning actually is

You have probably heard the term artificial intelligence used to mean almost anything a computer does that seems clever. That is too vague to be useful. Let us start with a plainer explanation:

Machine learning is a way of writing computer programs that learn patterns from examples, rather than following fixed rules a person wrote out in advance. You show the program thousands or millions of examples. It works out the pattern for itself. Then it applies that pattern to new cases it has never seen.

A large language model, or LLM, is one particular kind of machine learning. It has read huge amounts of text and learned the statistical patterns of how words follow one another. That is what sits behind ChatGPT and similar tools. It predicts likely next words, over and over, well enough to produce readable answers.

But LLMs are only one branch of a much bigger tree. Most of the AI already shaping your life is not an LLM at all, and does not write sentences. Here are some plain examples:

• Sentiment analysis reads customer reviews or social media posts and works out whether people feel positive or negative about something, without a human reading each one.
• Robotic automation and self-driving systems use sensors and learned patterns to move machinery or vehicles safely, from factory arms to cars that keep their lane.
• Pattern matching in medical scans looks at X-rays, MRIs and other images to spot signs of disease that a doctor might miss or take longer to find.
• Entity recognition pulls names, dates, places and other useful facts out of large piles of text, such as legal documents or news archives.
• Intent recognition works out what you are actually trying to do when you type or say something to a chatbot or voice assistant, so it can route you to the right answer.
• Digital twinning builds a live computer model of a real thing, such as a jet engine or a whole city's traffic system, so engineers can test changes safely before making them in the real world.
• Trend analysis and outbreak detection spot unusual patterns in data early, such as a disease spreading faster than expected, well before it would be obvious to a person looking at raw numbers.
• Scoring systems rank or filter people, for CVs, insurance applications, credit checks and more. This is one of the oldest and most controversial uses of AI, because the pattern learned from past data can quietly repeat past unfairness.
• Image and music generation, and accessibility tools built on similar technology, can produce pictures or sound, or turn text into speech, speech into text, or one language into another, opening up communication for people who would otherwise struggle with it.

None of this needs a chatbot. It is worth knowing this distinction, because when people criticise or defend "AI" as one single thing, they are usually only thinking about LLMs. The wider picture is far bigger, and much of it has been quietly working away in hospitals, banks and factories for years.

Quakers have been here before

Quakers were shut out of university and public office for generations, because as Dissenters they would not swear the required oaths. So they went into trade instead. Banking, iron, chocolate and science all have strong Quaker roots. Barclays, Lloyds, Cadbury, Rowntree and Fry all began as Quaker firms.¹

This history matters for how we think about AI now. Quakers have never treated new tools or new industries as automatically suspect. The instinct has always been to get involved, watch closely, and correct course when the evidence demands it. That is a better starting point than either uncritical enthusiasm or blanket refusal.

There is also a theological argument. If there is that of God in everyone, then tools that extend attention and care to people currently short of both deserve serious consideration, not reflexive rejection. AI already helps doctors catch disease earlier in overstretched health systems. It gives disabled people new ways to communicate. It can lift routine admin from small charities and Quaker meetings so people can spend their time on things that actually need a human. Refusing all of that on principle is not automatically the safe choice. It has a cost too.

None of this is a case for enthusiasm. It is a case for using these tools with our eyes open, which takes more discipline than either embracing or rejecting them outright.

Where our own history warns us to look harder

The clearest warning from Quaker history is the Cadbury cocoa case, and it is not a comfortable one.

By 1901, William Cadbury had heard credible reports that cocoa from the Portuguese islands of São Tomé and Príncipe was produced using forced labour. He commissioned investigations. He wrote to the plantation owners. The company kept buying the cocoa throughout this process. It did not stop until 1909, eight years after the first warning signs.² A Tory newspaper later called this out publicly as hypocrisy, contrasting Cadbury's model village at Bournville with the conditions on the plantations that supplied its chocolate.³ Cadbury sued for libel and won, but only a token payout, and the reputational damage stuck.

The lesson is not that Quakers are hypocrites. It is that knowing about harm in your supply chain and acting fast enough on that knowledge are two different things. The gap between them is where the real damage happens, both to the people harmed and to the credibility of the people who knew and were slow to act.

This maps directly onto AI. The people who label and moderate the data behind large language models often do difficult, low paid work. A 2023 investigation by Time found that OpenAI used Kenyan workers, paid between $1.32 and $2 an hour, to read and label some of the most disturbing text on the internet, so that ChatGPT would not repeat it.⁴ Workers described lasting psychological harm. Some later organised the first African content moderators' union.⁵

If Quakers are going to use AI tools with our eyes open, this is the part we need to look at directly. The harm is upstream and invisible, in much the same way that forced labour was invisible in a bar of chocolate a century ago.

What we can actually do about it

Extend our existing ethical investment screening. Quakers in Britain already screen investments against arms and fossil fuels, and were the first UK church to divest from fossil fuels, in 2013.⁶ The same screening should ask what labour lies behind any AI firm we invest in or use, not just what the output looks like.

Ask suppliers a direct question. Before any meeting adopts an AI tool, ask how the underlying model was trained and moderated, and who did that work. If a supplier cannot or will not answer, that tells you something useful in itself.

Treat AI energy use as part of our existing climate commitment, not a new debate. The Canterbury Commitment of 2011 already binds British Quakers to becoming a low carbon community, and frames sustainability as part of testimony, not an optional extra.⁷ Running large AI models uses real energy. Any meeting adopting these tools should weigh that against the commitment already made, rather than treating it as a fresh question to argue over later.

Prefer smaller, local tools for everyday tasks. You do not need the largest available model to draft a minute or format a newsletter. Smaller tools use less energy and keep a sense of proportion, which fits comfortably with the testimony of simplicity.

Use AI to remove drudgery, not to replace discernment. Let it help with correspondence, funding applications, or formatting minutes. Keep it well away from corporate discernment itself, which depends on worship and relationship between people, not calculation.

Say when you have used it. If a piece of ministry, ministry-adjacent writing or an article draws on AI assistance, say so. This is simply the modern version of plain speech: say what you mean, and do not pass off assisted work as something it was not.

The honest point

Quakers do not need permission to use new tools. Our history is full of that already. What we have not always managed is acting on inconvenient facts about our own supply chains at the speed those facts demand, rather than at the speed that suits us. Cadbury took eight years. AI will ask us the same question that cocoa did. The test is not whether we use it. The test is how quickly we act once we know what it costs, and who is paying that cost on our behalf.

This article was published in The Friend on 17th July 2026.

Footnotes

← Back to articles