The threat of an AI-driven apocalypse looms large in headlines, with experts divided over how serious the risks truly are. Before exploring those scenarios, it’s worth understanding what AI actually is – and what it isn’t.
AI, or to give it its full name artificial intelligence, has long been the holy grail of computer science. Some may argue that it’s not really intelligence, but if any person was able to read a novel-sized document within a couple of minutes and summarise it succinctly, they’d probably be called a genius.
So, it is a form of intelligence, just not a biological one like human intelligence. Instead of a mind made of neurons, it is a computational system running on hardware built from silicon chips, metals and plastics.
AI, particularly the large language models (LLMs) dominating today’s discussion, doesn’t resemble human intelligence in a biological sense. These systems do not learn through the same biological processes as people, and whether AI can have subjective experience remains debated. Instead, they are vast statistical engines trained on enormous datasets, capable of analysing and summarising information at speeds no human could match.
Although their ‘neural networks’ borrow terminology from neuroscience, the similarity is mostly metaphorical. Biological neurons communicate through electrochemical signals; artificial ones perform mathematical operations on numerical inputs.
Conversely, research shows that some AI systems develop internal representations that partially resemble patterns in the primate visual cortex, but only in narrow domains and because they have been optimised for similar tasks.
In short, AI is powerful, but it is a different form of intelligence from our own. Understanding that difference is essential when considering the risks – including the apocalyptic ones.
When humans point AI in the wrong direction
With these differences and similarities in mind, let’s look at some possible scenarios. We’ll begin with the most common, the one that always hits the headlines: AI will kill all humans.
It would be more accurate to say that AI might pursue a goal in a way that unintentionally harms humans. AI will usually follow the guidance given to it by humans – at least for the time being. Unfortunately, rogue or malicious people may give it instructions leading to harmful outcomes in order to achieve their own nefarious aims.
Such people might instruct AI to try to access sensitive or secure systems. In an extreme scenario, a malicious actor might try to use AI to compromise critical military command systems, with potentially catastrophic consequences. Alternatively, AI might be used to develop and release a unique deadly virus specifically targeting humans.
Because these scenarios are believable does that make them any more likely to happen? Only, I would say, if rogue human agents or states are behind such acts, pushing AI in this direction.
Less dramatic but still potentially devastating and apocalyptic, is the possibility that AI might crash world economies or destroy satellite or traffic management systems, bringing chaos and making everyday life grind to a halt. Such scenarios could conceivably send humanity back to the dark ages, if not the stone age. But again, rogue human agents or states would probably be the root cause.
What if AI is accompanied by humanoid robots that are as intelligent as, if not more intelligent than, humans? Artificial general intelligence (AGI) remains a theoretical possibility. Chaos and destruction could be multiplied, with the eradication of humans reminiscent of the Terminator movies.
It appears to me that malicious human intervention would be a major danger, although a sufficiently advanced AI pursuing a poorly defined goal could also cause catastrophic harm without being instructed to do so
Beyond the utter destruction and extinction of humanity, there may exist an altogether different but nonetheless chilling apocalyptic scenario: the enslavement of humanity.
The danger of machine logic
Why would AI choose enslavement rather than mass extinction? My answer is that AI is a machine that calculates the fastest and most efficient way to achieve the goal it has been assigned. And this could prove the most difficult scenario to predict and avoid, particularly if AGI robots had gradually and innocuously assimilated into societies across the world. Trusted servants and companions, perhaps, but with no more morals than automated meat grinders.
There might, however, be another, simpler explanation – machine logic – for why AI might appear to want to control humanity through virtual enslavement rather than actual enslavement or annihilation.
Humanity may have assigned AI the task of solving an existential crisis. For example, a worldwide famine, an impending planet-destroying asteroid strike or climate change. Taking control of humanity might be calculated as the best option to save as many people as possible – or what appears to AI to be the optimum outcome.
Bear in mind, however, that it would be using machine logic without emotions and with questionable moral standards. Morals are not even a fixed and universal concept among humans. They differ between people, societies and cultures. So, whose morals should AI follow?
Imagine that it can see the uncertainty in the different moral behaviours and attitudes it is presented with. Machine logic is generally designed to find the best and sometimes simplest solution. And there you have the machine-derived moral decision, which may or may not align with any accepted human moral code.
What might this mean for humanity when faced with an existential threat and its saviour is, at best, an AI with dubious morals?
An apocalypse scenario – Who does AI decide to save?
Let’s take a closer look at one of those existential examples. Quite arbitrarily, let’s choose the asteroid strike.
Initially, who will AI choose to save if it can’t save everyone? Calculating that humanity might need to repopulate the Earth once the dust has settled, it could decide there is no logical reason to save anybody of retirement age or above, instead selecting those more likely to be young enough and fertile enough eventually to repopulate the planet.
Its attendant AGI humanoid robots might then quite literally extract all older people from the mix and leave them to their fate. Alternatively, perhaps AI might calculate that it would be better to hasten their departure through euthanasia, using some modern and quick, hopefully painless but not guaranteed, method.
Not through any ounce of pity, but simply as a practical, clinical act. AI is not inherently cruel, just naturally amoral, operating through mathematics, algorithms and statistics.It may simulate empathy in its responses, but that does not establish that it experiences empathy as humans do.
Staying with the asteroid scenario, food, water and other resources would need to be rationed, since it would be difficult to estimate when it would be safe for people to re-emerge onto the surface. This would be another significant area of AI control.
I doubt there would be individualised meal plans. Instead, all adults might receive exactly the same portions, regardless of age, activity level or body size.
Another area of control would undoubtedly be law and order, and this is where things might become even more testing. How can law, with its inherent level of moral guidance, be administered by a highly intelligent machine with no morals, particularly with AGI robots acting as police enforcers? Without adequate guardrails, it could be a recipe for heavy-handedness and disaster.
Finally, when the day arrives to walk out onto the surface of the Earth again, will AI allow its ordered little world below ground simply to disappear? Or, like so many other intelligent beings, might it decide that its reign of control should continue unabated?
The moral ambiguities of AI and the futuristic prospect of as-yet theoretical AGI robots echo my own novel, The Sentient Ones, where the existential threat facing humanity in 2070 is the climate crisis.
Apocalypse is not inevitable
Overall, the apocalyptic scenarios surrounding AI are unsettling, but they are not inevitable. They arise partly from misunderstanding what AI is and from the danger that its goals could drift away from ours.
If we treat AI as a tool to be carefully aligned rather than a mind to be feared or worshipped, we can reduce the risks long before they become existential. The future will depend not on the intelligence of our machines, but on the wisdom with which we design, deploy and constrain them.
The apocalypse is not a prophecy; it is a warning. And warnings can be acted upon.
The AI machines that loved us to death
A chilling near-future thriller of AI rule, rebellion, and survival – read it as a warning, or a prophecy; the...
Read moreDetailsThis writer’s life: a diary. Part 8: technology and me
In this eighth instalment of her diary, Lynda Tavakoli writes about life without a mobile phone.
Read moreDetails
Bylines Network Gazette is out!
The latest Bylines Network Gazette is out now. Become a Friend of Bylines Network or upgrade to a paid Substack subscription, and you will get access to a vital collection of citizen articles and stories.
Journalism by the people, for the people.







