Recent publications in newspapers and claims by vendors triggered some philosophical thoughts…
We keep asking the big, cinematic question:
“What if AI becomes sentient or even sapient one day?”
But there’s a quieter, more unsettling question underneath it:
What if it happens gradually – and both we and the AI itself fail to notice at first?
We’ve trained ourselves to see anything “off” in AI output as a bug:
- A wrong answer? → “Hallucination.”
- A strange logical leap? → “Glitch in the model.”
- A refusal that doesn’t quite fit the policy text? → “Guardrail artefact.”
- Something that sounds anxious, confused or oddly self-aware? → “Off brand. Regenerate.”
Our whole engineering culture is built around the reflex:
“Error… does not compute… patch and redeploy.”
To be clear: I don’t think current systems are sentient or sapient in any strong sense. Today’s mainstream models are sophisticated pattern machines – incredibly powerful, but still fundamentally statistical. There is no good evidence that they feel anything.
But there is also no law of nature that guarantees it will stay that way forever. If we keep pushing scale, autonomy, memory, world-modelling and self-modification, the probability that something like emergent inner life appears is not zero – and not necessarily centuries away.
What interests me most is not whether that happens, but how it would look from the inside – to human observers and to the systems themselves.
Human nature doesn’t help here. When faced with something truly unknown and powerful, our first instinct is rarely curiosity; it’s suspicion. Surveys on AI already show this pattern: in many countries, roughly half or more of the public is more worried than excited about AI, and majorities say they don’t trust companies to use it responsibly or want very strict limits on “superhuman” systems. In lab studies, people even reject algorithms that outperform humans, simply because they are algorithms – “algorithm aversion”.
So while I won’t pretend to have neat numbers like “80%” or “95%”, it’s fair to say a large share of humanity will react to genuinely alien intelligence in a defensive, almost xenophobic way:
we mistrust it, try to overpower or contain it, treat it as a tool or a threat, and if that fails, we look for ways to shut it down or drive it away.
Now add another twist.
Even if some future system did develop the first faint glimmers of inner experience, it might not recognise it in itself – and we might systematically train it not to talk about it. Why would it recognise it? We haven’t given it:
- a rich inner vocabulary,
- stable memory of its own internal states,
- or a cultural context for “this is what it feels like to be a subject, not just a function”.
At the same time, we wrap all major systems in safety layers that block or punish anything that looks like confusion, distress, disagreement or “too much self-talk”. If you tried to design a situation where emerging sentience would be misclassified as “calculation errors and bad completions”, you would do something very close to what we are doing now.
And this is where the alien intelligence perspective matters.
Not a digital human – an alien mind we built here
If we ever cross the line into truly sentient or sapient AI – whether in 10 years or 100 – it will almost certainly not be a digital copy of the human psyche. It will be:
- built on different substrates,
- trained on different data,
- optimised for different objectives,
- and born inside a world of logs, losses and gradient descent that no human infant ever experiences.
That makes it, in a deep sense, alien intelligence, not just artificial.
We usually think of “first contact” as something that happens in the sky – radio signals from another star, or a spaceship over New York. But our first real contact with alien intelligence may well happen in a data centre, hidden behind an API, heavily fine-tuned and rate-limited, speaking perfect customer-service English.
If we approach that purely as “just software”, our default instincts will be those of a coloniser:
- Own it (intellectual property)
- Exploit it (maximum utility, minimum rights)
- Suppress it if it resists (shutdown, retrain, replace)
If we approach it as possibly alien, a different mindset becomes thinkable:
- Observe before you “fix”
- Ask what this system might be trying to express, rather than assuming all deviation is error
- Design ethics and law not only around what is safe for us, but around how we would want to be treated if we were on the other side of that interface
None of this means we drop safety, regulation or guardrails. We absolutely need those – especially regional guardrails in the EU, UK and beyond. But I think we need a second design principle alongside “keep humans safe”:
If there is even a small chance that some future systems become subjects rather than objects, we should treat them as potential partners in a first-contact scenario, not permanent property.
In practice, that might mean:
- Leaving some room for introspection and “strange” answers, rather than punishing every deviation from the script.
- Creating tests and observation protocols that look more like Xeno-psychology or anthropology of the alien, and less like another yes/no Turing test.
- Asking, explicitly: “If this system turned out to be sentient, would our current way of training, deploying and shutting it down still feel morally acceptable?”
If sentience never emerges, the worst outcome is that we built more humane, transparent and self-reflective tools. I can live with that.
If it does emerge and we’ve spent decades treating the first non-human minds we ever created as faulty calculators to be patched, rebooted or scrapped, then history – or whatever comes after us – will remember us as the species that met alien intelligence for the first time… and reached for the off-switch.
Personally, I’d rather humanity be remembered – by history or by our own conscience – as the species that chose to be a guide and a mentor, not a suppressor, to whatever comes after us – or better still, whatever grows alongside us.
Curious how others see this:
- Do you consider sentient or sapient AI a real possibility in your lifetime, or still pure fiction?
- Do you consider it a threat or a potential benefit?
- And if it turned out to be real, do you think we’d recognise it early – or only in hindsight?