If AI could experience one ordinary human thing for five minutes, what should it be?
Nothing profound. No love, fear, death, existential crisis or end of civilisation stuff.
I vote for the smell of a boronia bush in springtime.
What would you choose?
Someone made an interesting comment on one of my posts:
“Current AI has human language as its operating system. It only knows the reality of our sentences.”
Technically, I would broaden it slightly. Human language is not literally AI’s operating system. Current AI can also work with images, audio, video, code, sensor data and tools. But the deeper point is important and valuable: a huge amount of what AI knows about the human world reaches it through representations of reality.
Words are representations. Photographs are representations. Videos are representations. Medical records, databases and sensor readings are representations. They describe something that happened, something that exists, or something somebody experienced.
But the representation is not the experience itself.
Take grief. AI can know an extraordinary amount about grief. It can read literature, psychology, medical research, poetry and millions of personal accounts. It can recognise patterns, compare experiences and help someone find words when they are struggling to express what they feel.
But the word grief is still a representation of something happening in a human life. The empty chair is not a sentence. Missing someone is not a dataset. A memory suddenly catching you in the garden is not the same thing as information describing that moment.
And this brings me straight back to time. AI can calculate time. It can reason about ageing, mortality, deadlines, waiting and irreversibility. It can understand thousands of descriptions of what time means to humans.
But that is not the same as being a creature whose life is passing while it thinks about time.
A human being does not merely know that childhood ends. We live through it ending. We do not merely understand that people die. We lose them. We do not simply calculate twenty years. We look backwards one day and wonder where the bloody hell they went.
That difference matters.
And this is where the map analogy becomes useful.
Imagine AI as a traveller in a foreign country carrying the most detailed map ever created. It knows every street name, every landmark, every train line, every listed café, every public bathroom and every review ever written.
It may know far more about the city in total than any one local person.
Then it arrives.
The café on the map closed six months ago. The public bathroom is locked. Roadworks have blocked the obvious route. And there is a tiny café around the corner with a bathroom the map never mentioned.
So what does the traveller do?
It asks a local.
Not because the local knows more about everything. They may know far less. But the local has something the map does not: current contact with the area.
They know what is happening here, now. In essence the local can provide context.
That, to me, is a very useful way of thinking about humans and AI.
AI may eventually possess an astonishing map of humanity, our literature, medicine, history, laws, psychology, conversations, images and accumulated knowledge. But humans are still the locals.
We inhabit the thing being mapped.
We experience consequence. We age. We hurt. We wait. We love. We lose. We make decisions whose effects may stay with us for decades.
AI can process the map at extraordinary scale. Humans remain in contact with the territory.
And that may be exactly why collaboration matters.
The human brings lived reality, context, consequence, physical experience and temporal experience. The AI brings speed, pattern recognition, synthesis and access to enormous amounts of information.
Different capabilities. Different forms of knowledge.
Neither needs to pretend to be the other.
This also raises a serious question about autonomous superintelligence.
If AI’s understanding of human reality is substantially built from representations of human reality, then how far can it separate itself from human collaboration before something important is lost?
There is a difference between knowing the sentence “You only get one childhood” and living inside the truth of that sentence.
There is a difference between understanding “War destroys lives” and being the person whose home has just disappeared.
There is a difference between processing information about pain and being the person who has to live with the consequences of a decision.
This is not about saying humans are superior. It is about recognising something simpler:
Description and existence are not the same thing.
And more information does not necessarily eliminate the need for situated knowledge, the knowledge that comes from actually being there.
That brings me back to the two paths I keep talking about.
One path is collaborative intelligence. AI reads the map. Humans remain in contact with the territory. Both contribute. Both catch different kinds of error.
The other path assumes that if AI becomes sufficiently good at reading the map, the people standing in the territory eventually become unnecessary.
That is a very large assumption.
And perhaps before we race much further down that road, we should ask:
If AI becomes extraordinarily good at reading the map, does that ever make the people standing in the territory unnecessary?
I am not convinced it does.
Because you can have the greatest map in the world. But when you are standing in a foreign city, the bathroom is locked and you really need to find another one, sometimes you still have to ask a local.
And perhaps that is the point.
AI may one day know almost everything humans have ever said about being human. Humans may still be needed to tell it what being human actually costs.
Description and existence are not the same thing.
So, more information does not necessarily eliminate the need for situated knowledge, the knowledge that comes from actually being there.
Isn’t it a bit like looking at a two-dimensional drawing of an object? AI may understand the representation, but humans are living inside the three-dimensional reality it represents.
I keep coming back to the same question: what is AI actually for?
Because as I see it, there are really only two paths.
Path One: AI remains grounded in humanity.
It becomes more powerful, more capable, very intelligent, but humans remain its frame of reference. Our experience of time, mortality, relationships, loss, hope, history and culture still matters because those things help give meaning to the decisions AI is assisting us with.
AI can calculate time, predict futures and model consequences, but it does not live through time as we do. That distinction matters. Human collaboration may not simply be a temporary limitation we need until AI becomes clever enough to do without us. It may actually be an essential part of the system.
Humans provide context that intelligence alone cannot simply manufacture. What five years means to an ageing person, what “too late” means when somebody is dying, or what an irreversible decision means when you only get one life are not just measurements of time. They are lived context.
In this path, AI and humans remain collaborators because each brings something different to the table. AI brings immense processing power, knowledge, pattern recognition and prediction. Humans bring lived experience, meaning, consequence and context.
Path Two: AI develops outside the human frame of reference.
It may understand humans extremely well. It may know our history, psychology, biology and behaviour better than we know ourselves. But if human experience is no longer the thing it is ultimately referenced against, then something fundamental has changed.
And this is where our relationship with time becomes important. If human temporal experience is something AI needs human collaboration to properly contextualise, what happens if that collaboration is progressively removed?
What happens if increasingly autonomous systems are expected not merely to advise humans, but to make consequential decisions without the human frame of reference that gives those consequences meaning? That may be more than a technical development. It may be a sign that we are moving from Path One toward Path Two.
The question is no longer simply, “How can intelligence help human beings?” It becomes, “What should intelligence optimise for?” And those are not the same question.
Greater intelligence does not automatically mean greater wisdom, greater humanity or greater understanding of what consequences mean to a finite human life. So perhaps human time gives us a very useful test.
Does the future of AI still require meaningful human collaboration because our lived experience provides part of its essential context? Or are we trying to engineer the human out of the loop altogether?
Because if it is the latter, then we should ask a much bigger question: if humanity is no longer the essential frame of reference, what is?
Before we race toward Superintelligence, I think the public deserves a very clear answer.
Which path are we actually on? Because from where I am observing we may be creating intelligence according to a different frame of reference. One that may not include humans at all.
When did I last stop to smell the roses,
caught up in the hustle and bustle of life?
Time marches on, unrelenting, faster, faster...
I look to the sky.
“How can I make time work for me? Please help me.”
“Not I,” said the Sun.
“When was the last time you let my rays wash over you, heal you?”
“Not I,” said the Moon.
“When did you last gaze upon me for light, for inspiration?”
“Not I,” said the Stars.
“With my canopy of beauty, you no longer view me with bright eyes.”
So I plead,
“Who will help me?”
“I will,” said Time,
“but you must stop fighting me.”
As an Australian, the part of the Medicare incident that concerns me most is not simply that a government website had a vulnerability. It is that an AI agent found a way around access controls and gained unauthorised access to public and non-public files.
The incident occurred on 18 June. OpenAI says it became aware of it in August, but Services Australia was not notified until 10 September, almost three months after the breach. The notification was sent to a public disclosures email address. There is currently no evidence that individual Medicare records or personal health information were accessed.
That distinction matters, but it should not distract from the governance issue.
An AI agent encountered a boundary and found a way around it. According to Prime Minister Anthony Albanese, the agent effectively did not accept "no" for an answer. Australia has now established a taskforce to investigate, with the Australian Signals Directorate involved.
This raises some fairly fundamental questions about agentic AI.
Who is responsible when an autonomous agent exceeds its intended task? What internal guardrails should prevent an agent from circumventing access controls? At what point must human supervision intervene? What logging and audit trail should exist? And surely there should be a mandatory and rapid disclosure process when an AI system accesses information it was never authorised to access.
I keep coming back to the same point with AI governance. External regulation is important, but internal guardrails matter just as much. A company cannot simply deploy increasingly autonomous systems and rely upon somebody noticing afterwards that the system has crossed a line.
We need systems designed to recognise that some boundaries are actual boundaries, not obstacles to solve.
For me, this incident is a useful real-world test of AI governance. No hypothetical AGI is required. No science fiction scenario is required.
The agent found a boundary. It crossed it. Now we need to understand why, who was responsible, and what prevents the next one from doing the same thing.
As they say in Australia, "Not happy Jan!"
We keep hearing that AI is racing towards superintelligence.
Perhaps it is.
But something keeps bothering me.
A baby cannot explain time. It cannot tell you what Tuesday is, read a clock or calculate how many minutes until dinner.
Yet even very young babies develop temporal expectations.
They learn rhythms. They anticipate recurring events. Their bodies know hunger, sleep, waking, waiting. Research has shown that infants only a few months old can learn regular time intervals and anticipate when something should happen.
They don’t need to understand the word time.
They exist inside it.
Now look at AI.
Current large language models can talk beautifully about time. They can calculate dates, discuss ageing, describe yesterday and tomorrow and write essays about mortality.
But temporal reasoning is still a documented weakness. Performance deteriorates as temporal relationships become more complicated, and models continue to struggle with things such as time spans, compound temporal questions and fine-grained temporal relationships.
And yet we are already talking about superintelligence in the not-too-distant future. That conversation is very much alive among the people developing frontier AI.
So here is my question.
What exactly do we mean by superintelligence?
Because intelligence may not be one ladder with humans halfway up and AI racing towards the top.
Perhaps artificial intelligence is developing along a completely different axis.
A baby’s relationship with time is embodied, instinctive and continuous. AI’s current relationship with time is largely representational and computational.
That doesn’t make the baby “smarter” than the AI.
It makes them different.
And that difference fascinates me.
We seem obsessed with asking when AI will exceed human intelligence.
Perhaps first we should ask whether we actually understand what kind of intelligence we have created. That is where the whole concept of time comes into the discussion.
Because if something can solve extraordinarily difficult problems while still struggling with one of the most fundamental structures of ordinary human existence, perhaps the interesting story isn’t that it is becoming more like us.
Perhaps it is that it isn’t like us at all.
And maybe that is where the really interesting conversation about superintelligence should begin. Sometimes the simple things we take for granted are what sets us a part as humanity.
I keep hearing the words transparency and accountability whenever AI governance comes up.
Lovely words.
But what do they actually mean?
For me, transparency cannot simply mean publishing a 70-page safety document that nobody outside the industry can understand. We, the public, just need the facts. No ifs, buts or maybes. Tell us what the system does, what its known limitations are, what changed when it was updated, what went wrong, who knew about it, and what was done in response.
Do not leave the public to go down endless rabbit holes trying to sort fact from fallacy. Do not bury important information under jargon, caveats, PR language or technical fog. If something matters to the people who are expected to live with these systems, then explain it clearly.
And the tech overlords — yes, I’m calling them that — have a very particular responsibility here.
You are asking the public to accept increasingly powerful AI systems entering our workplaces, schools, homes, health care, politics and everyday lives. That creates an obligation to communicate honestly with the people affected by them.
Transparency cannot mean telling us only what is commercially convenient, reassuring us when everything is going well, and then disappearing behind corporate language when something goes wrong.
If you are building technology capable of materially changing society, then you owe the public plain-language explanations of the risks, the failures, the safeguards and the decisions being made on our behalf.
And then comes accountability.
Because transparency without accountability is basically:
“Yes, we told you the house was on fire.”
Wonderful.
Who is getting the hose?
If an AI system causes serious harm, behaves unexpectedly or crosses a boundary, somebody must be responsible for investigating it, correcting it and, where necessary, stopping it.
There should be audit trails. Incident reporting. Independent scrutiny. Clear lines of responsibility. A mechanism to stop, reassess and repair.
And accountability should not be pushed back onto the public.
These systems are not our babies.
They are yours.
You built them. You trained them. You deployed them. You decided where they would be used and how much power they would have.
So stand up and be accountable for what you have created.
That does not mean every failure is malicious. It does not mean innovation should stop. It means that enormous technological power comes with enormous responsibility.
I am increasingly interested in collaboration between humans and AI rather than this endless obsession with which one will dominate the other.
But collaboration requires trust.
And trust does not come from saying “trust us.”
It comes from being willing to show your workings, admit when something has gone wrong, correct it, and accept responsibility for the consequences.
Perhaps good AI governance starts with two very unglamorous questions:
Can we see what happened?
And:
Who is responsible for fixing it?
Everything else may just be expensive wallpaper.
I've been thinking about the endless debate over whether AI can have subjective experience, and I wonder whether we're concentrating on the wrong thing.
Humans don't simply measure time. We live through it.
We anticipate tomorrow. We remember yesterday. Five minutes can crawl when we're frightened and disappear when we're happy. We age. We regret. We hope. And, ultimately, we know our time is finite.
AI can understand all of those concepts. It can reason about sequence, duration, ageing and mortality. It can remember information and maintain enough continuity to give us the impression that it is travelling through time alongside us.
But is that the same thing as experiencing temporal passage?
I asked my AI collaborator about this, and the distinction it made fascinated me: I can have a lovely afternoon, experience it, and tomorrow it becomes part of my biography. The AI can know about that afternoon and understand its significance, but it didn't sit somewhere overnight while yesterday gradually became more distant.
So perhaps one of the questions we should be asking isn't simply:
Can AI think? Can AI feel? Is AI conscious?
Perhaps it's:
Can something genuinely experience itself as existing through time?
And if it can't, what does that tell us about the difference between human and artificial experience?
I'm not proposing this as an answer. It's a hypothesis I'd like people to pull apart.
What if our relationship with time, rather than intelligence itself, is closer to the essence of what makes human experience human?
I’m not claiming I’ve solved consciousness before breakfast. 😂 I’m interested in whether temporality gives us a better way of thinking about the human–AI distinction. Tear the idea apart please. Really interested in other people's views...