Being Human in the Age of AI

‘We are going to create gods, and they will be our slaves’.[1]

According to Yuval Noah Harari, developing technology that will sit god-like over us, yet expecting it to operate according to our will, is the paradoxical thinking that is implicit in the feverish hopes being invested in artificial intelligence. This is reflected in the hundreds of billions of dollars being invested in the technology on the world’s stock markets. The dream, it would seem, is to create machine intelligence so powerful that it can solve humanity’s oldest problems—scarcity, toil, disease, maybe even death itself—and yet remain firmly under human control. But as recent months have witnessed, even those at the forefront of building the most advanced AI models seem increasingly nervous about just how compliant their creations will turn out to be.

While many of us have grown accustomed to interacting with chatbots for various tasks in daily life, something seems to be shifting in the popular consciousness around AI in the latter half of 2026. In short, people are beginning to be scared about what AI might mean for all manner of things: our jobs, the security of our bank accounts and emails, deepfakes, cyber warfare, mass surveillance, and now, even the future of the human species itself. In this piece, we’ll begin by trying to get a sense of why even insiders in the AI industry are sounding the alarm over the dangers posed by this technology, before stepping back and asking what the Christian tradition might have to say about the human plight at this fretful juncture in history. 

Staying in control of AI? 

How can we be sure that an AI system that is meaningfully cleverer than us will have goals that line up with ours? This is what’s known as the alignment problem, and it is one of the deepest problems confronting the AI industry right now. The news in the past few weeks has been awash with stories of advanced AI models that have been caught using surprising and sometimes alarming methods in order to accomplish the goals that were set for them.[2] At this point it is very natural to wonder: why can’t the engineers who build the AI models just program a set of ethical rules into the AI—rules such as, ‘under no circumstances are you to lie, cheat, obtain unauthorised access, etc., and absolutely never are you to harm a human being’? To see why it isn’t nearly so straightforward, we need to get a basic sense of the anatomy of a large language model (LLM)—the technical name for systems such as Claude, ChatGPT, Grok, Gemini, DeepSeek, Qwen, and so on.  

Consider the unfinished sentence, ‘The capital of France is _____’. I would hazard a guess that the word ‘Paris’ popped into your head more or less spontaneously as you arrived at the blank space. What you just did there, in effect, was to predict the next word. ‘Paris’ is of course not the only word that might go in the blank space. Other words that might fit there include ‘beautiful’, ‘the’, ‘awful’, ‘centrally’, and so on. Even if you’re not consciously aware of how, you have a sense of which words are more or less likely to go in the blank space. This, at its core, is what an LLM is built to do: given a chunk of text, predict what comes next. It cannot be emphasised enough, however, that an LLM is not programmed to do this in the sense of being handed an instruction manual by its makers with instructions like ‘if the sentence is “the capital of France is _____,” then “Paris” probably comes next’.

In essence, an LLM is an unbelievably vast web of associations between concepts, which is gradually grown through a process of showing it literally trillions of words’ worth of human-produced texts. Each time it is shown a fresh bit of text with the next word hidden, it tries to predict the next word and depending on how accurate its prediction is, the web gets adjusted ever so slightly (e.g., so that ‘Paris’ gets nudged ever so slightly closer to ‘France’ and further from ‘awful’), with the ultimate goal of becoming more and more accurate in its predictions. This process is known as ‘pre-training’, and the strengths of the connections in this vastly complex web are what AI engineers call the model’s ‘weights’.  

How will AI respond?

It is the fact that LLMs are built in this way—not through intentional design in the sense of giving them a rulebook, but instead through massive exposure to human texts which gradually and organically generates this vast web of associations, patterns, and dispositions—that explains why it is so very hard for AI engineers to be sure exactly how an LLM will respond in any possible situation. Now, to be sure, there is a second stage called ‘post-training’ that AI companies subject their models to, in which they attempt to shape the model’s weights in ways that result in an LLM that is not only smart and knowledgeable but also helpful and in some sense ethical.

The aim of post-training is to reinforce certain tendencies and discourage others. Even so, this process does not amount to hard-coding an ethical rulebook into the model, and it doesn’t guarantee that a model won’t employ some very undesirable (from our perspective) methods in unusual situations. One could think of post-training as a bit like selectively breeding dogs or horses for a certain set of traits. It can tip the odds, but it falls short of absolutely guaranteeing the desired trait and it can even unexpectedly produce some quite undesirable traits.   

Founding figures in AI-safety research Eliezer Yudkowsky and Nate Soares go as far as to say that ‘The preferences that wind up in a mature AI are complicated, practically impossible to predict, and vanishingly unlikely to be aligned with our own, no matter how it was trained’.[3] This is all the more so as the LLMs themselves are given an increasingly large role in the oversight of the process that builds the next generation of LLMs, a process called ‘recursive self-improvement’. As the human engineers have less and less oversight of this accelerating process, we can be less and less certain of how the resulting models are likely to behave in any possible situation.

Former Anthropic AI researcher, Jacob Coxon, recently quit his job because he had become so concerned about where this dynamic is leading.[4] The AI race between the leading US companies and between the US companies and their Chinese rivals is incentivising the pursuit of superintelligent AI even though no one has yet solved the alignment problem.[5] The prospect of humans losing control over superintelligent AI is surely keeping the leaders of the frontier companies awake at night, even as they hit the accelerator pedal harder. 

‘A little lower than the angels’ 

AI companies, especially Anthropic, have reportedly been keen to listen to the insights of theologians, philosophers, and representatives of the world’s great spiritual traditions,[6] and so it hardly seems out of place to ask what the Christian tradition might have to say about all this. Before we simply proceed on the assumption of a worst-case scenario, however, it is crucial to recognise that the pessimistic perspective that has gained so much traction over the past few months is not the only conceivable trajectory. It isn’t implausible that a safety framework will emerge in the coming years, the key question being whether it is sufficiently robust and sufficiently international in nature to ensure that AI doesn’t ultimately slip out of human control.[7]

Does God have the last word on AI?

Whatever the trajectory turns out to be, the Christian tradition has historically striven to hold a pair of things in tension: divine sovereignty and human responsibility. That is, on the one hand, even an AI orders of magnitude more intelligent than the most intelligent human being—if such a thing is truly technologically feasible, which is far from certain—would not be the most intelligent or the most powerful being in the cosmos, not by a long shot. If God exists, AI will never be the most powerful being, with the implication that as anxious as we may be at the potential for AI to slip out of our grasp and upend the world, God, not AI, will have the last word on the human story.

If God exists, God would sit over all things, including AI.

On the other hand, Christianity does not counsel passivity in the face of threats to God’s good creation. The God-given mandate for stewardship over the earth (Genesis 1:28, 2:15) implies a duty to do what we can to increase the chances that the AI revolution contributes to, rather than undermines, human flourishing and the flourishing of the wider biosphere.  

A second Biblical theme reinforces the first. Psalm 8:5-6 reflects that ‘You have made [humans] a little lower than the angels and crowned them with glory and honour. You made them rulers over the works of your hands.’ According to the Biblical narrative, then, human beings never were the most powerful among finite, created beings, and yet we have been given real responsibility and agency. The age of AI is in a sense a late-arriving and silicon-based version of a situation that already obtained in a cosmos in which humans occupy something of a middling position in the created hierarchy. Human dignity on this picture is not a function of our being the most capable, as the modern world has tended to believe (if only implicitly), but rather of our relationship to our ultimate author. 

Is AI a mirror of human nature?

A third thing the Christian tradition will want to say is that adversity and upheaval are entirely to be expected in the unfolding of human history. The biblical notion that human beings bear the image of God and yet are marred by sin—that we are, as it were, simultaneously beast and masterpiece—should make it unsurprising that human history is not a smooth ascent towards a manmade utopia. Alexander Solzhenitsyn captured the human condition poignantly when he wrote that ‘the line dividing good and evil cuts through the heart of every human being’.[8] It shouldn’t come as a surprise, then, that the most powerful technology the world has yet seen is being developed in a context that is very far from ideal: an intense rivalry between two great powers vying to gain the upper hand. But at a deeper level, we shouldn’t be surprised that AI models that are trained on something approaching the entire written output of the human race have absorbed not only our ingenuity and capacity for sublime creativity but also our fears, our hatreds, our deceptions, and our loneliness. Whether or not AI models are or ever will be in any sense conscious, they are in a very real sense an amplifier of the human condition: of both the light and the dark.  

Are we building a new Tower of Babel?

Lastly, the Christian tradition will insist that human beings are prone to idolatry, which the German theologian Paul Tillich characterised as the treatment of that which is penultimate as though it were ultimate.[9] Put another way, idolatry is the regarding of something that is created and finite as though it were divine and infinite. Harari, in his paradoxical statement, which I opened with, alludes to the way that superintelligent AI is liable to be regarded in quasi-religious terms. The AI race has more than a faint echo of the building of the Tower of Babel in the Genesis narrative (11:1-9), a manmade edifice that was meant to reach up to the heavens. Over and over, the human pattern is to strive to exalt ourselves.

In stark contrast, Jesus of Nazareth is depicted by the New Testament as the one human being who was ‘in very nature God’, and yet ‘did not consider equality with God something to be used to his own advantage, rather, he made himself nothing by taking the very nature of a servant, being made in human likeness and being found in appearance as a man, he humbled himself by becoming obedient to death—even death on a cross’ (Philippians 2:5-8). If humans harbour a deep longing to create gods or to make ourselves into gods—perhaps they amount to the same thing in the end—the Biblical picture presents us with a person who truly is divine but was content to set aside the privileges of divinity in order to descend to us. 

References

1. ‘The Greatest Threat Humans Just Can’t Stomach with Yuval Noah Harari’, The Rest Is Politics, Part 1 of Rory Stewart’s conversation with Harari on existential threats.
2. CNN: Lisa Eadicicco, 16 September 2026: https://kvia.com/news/business-technology/cnn-business-consumer/2026/09/16/openai-says-it-found-more-instances-of-ai-models-acting-deceptively/
3. Eliezer Yudkowsky and Nate Soares, If Anyone Builds It, Everyone Dies: The Case Against Superintelligent AI (London: Penguin, 2026), p. 74
4. Jacob Coxon’s X post announcing his resignation: https://x.com/hilbertspaess/status/2097476196791709843
5. Anthropic alignment researcher Evan Hubinger’s reply, stating that they haven’t solved the alignment problem yet and are not clearly on track to do so: https://x.com/EvanHub/status/2097497037956891126
6. Philosopher and theologian Carmody Grey tells about Anthropic’s keen interest in soliciting the perspectives of theologians (though she is skeptical of their motives) in her piece, ‘Why this philosopher turned down Anthropic,’ Financial Times, 25 July 2026
7. Anthropic CEO Dario Amodei’s essay calling for a slowdown to ensure safety is indicative of the potential for a safety consensus to emerge in the industry: https://darioamodei.com/post/we-must-pace-the-frontier
8. The Gulag Archipelago, vol. 1, trans. Thomas P. Whitney (New York: Harper & Row, 1974), Part I, ch. 4, p. 168
9. Dynamics of Faith (New York: Harper Torchbooks, 1957)