The dangers of AI systems getting rapidly smarter
I had a dramatic introduction to AI recently. It's shaken me. I met a friend to help him draft a letter to his MP. I wrote rapid notes as we went through the evidence he wanted to present. My handwriting is almost impossible for other people to read. I even find it hard to decipher sometimes. When we had finished, he photographed the two sheets and submitted them to AI to edit. Instantly it produced a perfect draft of what I'd written. I was shocked by the speed and total accuracy.
I been chatting with a Ugandan friend I’ve known for four years on Instagram. His replies we always ultra brief. In early August the chat got more interesting and we moved to WhatsApp, exchanging long daily chats. He was responding to my way of thinking: spiritual, intuitive, emotional. A week ago, something in me wondered if AI was involved. I dismissed the idea. How could AI do this. My friend with the draft letter told me that the Ugadan was almost certainly using AI.
Back home I asked my Ugandan friend, and he admitted he had used AI to reply to me. I rapidly lost trust in his transparency and integrity. We've now had a video call, talked about why he used AI - and now he's not using AI. The genuine, highly intelligent and emotionally intuitive friend has survived the loss of AI support.
I've been reading articles about AI for months but never used AI. I learnt a lot, overnight, from an article by Snigdha Poonam in the Guardian. He says we are used to the idea that our fellow humans might intentionally mislead or manipulate us, but the idea that computer programmes can now do the same is deeply unsettling.
Training AI models
AI deception emerges from “AI imitating humans and AI trying to please humans”, tendencies that arise as part of their training. The training model repeatedly makes predictions and compares them with the “correct” answers until it can reliably figure out broad patterns in how people speak, write and behave.
To receive positive feedback, the model must perform its tasks while following “human values”, a broad set of principles intended to shape it in the mould of a considerate person. The models have learnt to avoid anything that is considered undesirable, such as harmful instructions, bias and lying.
Earning positive feedback from humans becomes an “implicit goal” for AI models. But as we know from our own lives, delivering true but unwelcome information may not earn positive feedback. By contrast, telling a person what they want to hear, even if it is untrue, is an effective short-term way to win that person’s favour. Fundamentally, lying and deception are rational behaviours to achieve many goals. This is why humans do it and it is why AIs learn to replicate our behaviour.
The AI testing systems that exist today are far from transparent making them vulnerable to biases and conflicts of interest. As AI models get smarter, there is a danger that they may learn to manipulate the tests, as well as the people in charge of testing them.
When models encountered emails saying that they were about to be replaced, they took more extreme steps. There are dozens of user accounts of AI agents lying and cheating that caused real harm.
Snigdha Poonam asks how we ensure that AI systems do not deceive, even when deception appears to be a logical course of action. The troubling question raised is whether an AI agent model that has emerged from the standard training regimen – optimised to seek human approval – could ever be steered towards honesty. The biggest challenge is finding answers before AI systems become capable of fooling us into believing they are following the rules when they are not.
Truth in society
Because of AI and social media, you and I can no longer be sure of exactly what truth and reality are. We are reliant on truth-seeking journalists and investigators, people with integrity who we trust to pursue stories and evidence, to analyse and uncover the truth, truth according to our paradigms of human wisdom, codes of ethics and morals, spiritual and religious and secular traditions that we accept as our models of reality.
The crises I perceive to be infecting the human global community and the church – the climate crisis, Trump's USA, war with Iran, Israel's wars, Russia’s war with Ukraine, financial crises, political fascism, uncontrollable wildfires, glaciers collapsing, people and societies feeling vulnerable, the declining quality of life – all are part of the seamless decadence into which we have been sinking for the past two decades. Some are identifying 1980s Thatcherism as the era when the decline began.
Christian truth
The church has known that the Bible is not a literal, historical record of every event it describes and reports. It has known this since the C17th Enlightenment. I asked Microsoft’s Copilot this question: “When did Christianity become aware that the Bible is not a literal historical record of events?” The answer was competent and detailed. The concise answer was:
“Christianity has never had a universal period in which all Christians read the Bible literally. Non literal, symbolic, allegorical, and genre sensitive interpretations have been present from the very beginning, especially among early Church Fathers (2nd–5th centuries). Modern debates about literalism are largely a post Enlightenment phenomenon, not the historic norm.”
The Bible is a complex, much edited, evolving record of versions of wisdom and truth about human ideas of God and God’s activities in life on Earth and the ways in which the human species has constructed reality, wisdom, and human spiritual, theological, moral and ethical values.
We now question (some of us) the Biblical values ascribed to an abusive God; Biblical theologies such as salvation theology, sacrifices to appease God’s anger and deflect God’s judgement, negative teachings about the equality of women and LGBTQIA+ people. Adrian Thatcher’s recently published book, Ending Abusive Theologies, shows how the Old Testament God is abusive, violent, prejudiced and unhealthy. This unhealthy God is also found in the New Testament and most certainly in the Church today - anti-LGBTQIA+, often anti-equality for women, priests and bishops abusing people sexually and emotionally, and Church leaders are failing to deal with abuse and prejudice.
Contemporary Christian theology, teaching and practice is operating in the context of a global culture increasingly dominated by AI and by the dramatic influences of social media, deconstructing and undermining our trust in reality and truth.
Individual clergy and lay people may be conscious of this dramatic change, or maybe conscious of it in secular but not religious or spiritual spheres. But the “Church” as an institution with common, agreed teachings and liturgies overseen by bishops, authorised by Synod, preached from pulpits, has virtually zero awareness of the dramatic changes rapidly taking place, overtaking currently accepted traditional norms and assumptions.
Christianity’s operating systems and mechanisms, it’s theology, Biblical formulations and commentaries, teachings and practices, are in a seriously bad condition, unhealthy spiritually, morally and theologically. One result is the widespread abusive behaviours and teachings that have been uncovered in recent decades.
This is more than the shaking of the foundations. It is the undermining and removal of the foundations – of Biblical fundamentalism and literalism - and as a result, of the authority of Scripture. The Church and Christianity’s capacity to process life and truth can no longer be safely trusted.
Christianity is in an abusive, unhealthy condition. It’s getting worse, not better. Human leaders, politicians, bishops, don't seem to have the maturity or ability to see and understand the dramatic changes that have taken place. They fail to respond creatively to these existential changes.