I do not dispute that we need regulation for artificial intelligence. We have lived through enough of the wild west of the dot-com era to know that leaving every decision to the market is not a serious plan for protecting people.

But I think we keep missing the point.

Technology is supposed to serve people. To extend what we can do, connect what was separated, and make parts of life better. Yet again and again, we create something extraordinary and then seem surprised when it begins to reflect the less admirable parts of us.

An investigation published by news.com.au raises deeply troubling concerns about AI and young Australians, including exploitation, deepfakes and the effect on learning. I think those concerns deserve to be taken seriously. A child harmed by a fabricated image does not need a philosophical lecture about innovation. They need protection, accountability and someone willing to act.

Still, after reading the reporting, I found myself thinking less about the machine and more about a pattern we have repeated for centuries.

The internet we thought we were building

I remember the early internet as a place of discovery. There was something almost generous about it. People gave away knowledge because they could. Someone you would never meet had taken the time to explain a problem you had spent two evenings trying to solve. A forum could be a classroom. A website could open a world.

That memory is real, even if it is not the whole history. The networks that preceded the public internet grew partly from government-funded research and defence interests. ARPANET connected researchers and computing resources, not some untouched utopia of human goodwill (DARPA's history).

But the possibility it offered was astonishing: knowledge could move without needing the permission of a publisher, a broadcaster or an institution.

Then we learnt to make money from attention.

The internet did not stop being useful. It became more useful in countless ways. But alongside the libraries and communities came surveillance, scams, exploitation and markets for things nobody should have to encounter. The dark web did not create human cruelty. It gave some of it new places to organise and hide. The wider commercial web found another way to distort things: make attention the product, and the incentives soon begin to favour whatever keeps us looking.

Australia's eSafety Commissioner warned in a May 2026 paper on recommender systems that systems optimised for engagement can amplify harmful or misleading material. That is not a failure of our ability to connect computers. It is a consequence of what we decided to reward.

We built a remarkable network for sharing information. Then we built businesses that could profit when information made people anxious, angry or unable to look away.

When we split the atom

Nuclear fission offers a harder comparison because its consequences are harder to ignore.

In December 1942, Enrico Fermi's team achieved the first controlled, self-sustaining nuclear chain reaction. From the same fundamental discovery came two paths: reactors capable of generating electricity and weapons capable of destroying cities (US Department of Energy).

Physics did not choose either path. People did. Governments funded them, scientists worked on them, institutions organised them, and wartime decisions gave one use a terrible urgency.

The bomb then became something else: a deterrent. Peace, or at least the avoidance of direct war between nuclear powers, was partly entrusted to the threat of destruction on an unimaginable scale. That is a strange achievement for a species that likes to describe itself as progressing.

Nuclear deterrence is not the whole explanation for the absence of a direct great-power nuclear war, nor is it a guarantee of safety. The very existence of the Nuclear Non-Proliferation Treaty and international safeguards tells us how much effort has been needed to contain the danger.

We discovered how to release energy from matter. We also found a way to make human survival depend on other humans exercising restraint.

That seems worth remembering whenever somebody tells us that a new technology is simply too important to slow down.

A road through the mountain, or a weapon

Alfred Nobel invented dynamite to make powerful explosives more practical to handle. It helped people carve tunnels, build railways and undertake work that had previously been dangerous or extraordinarily difficult. It also became a means of violence. The Nobel Prize's own history records both its construction uses and its use in warfare.

I find that duality unsettling. The same capability could clear a path through a mountain or destroy what stood on the other side of it.

There is a temptation to call the invention neutral and leave the matter there. But design, access and purpose are human choices too. We cannot place every moral responsibility on the person holding the finished tool while excusing the people who made it available, promoted it or profited from it.

The responsibility is shared. It is not evenly shared.

The medicine that taught us restraint

Antibiotics offer a quieter example. They transformed medicine by making infections treatable that had once killed people. Yet excessive and inappropriate use has helped drive antimicrobial resistance, weakening the effectiveness of the very medicines we depend on. The World Health Organization identifies misuse and overuse as major drivers of that threat.

There is no villain hiding behind every unnecessary prescription. Sometimes it is pressure, convenience, poor information, a system rewarding the wrong outcome, or the understandable wish to do something rather than nothing.

That may be the most uncomfortable parallel of all. We do not always create the monster through malice. Sometimes we create it through millions of apparently reasonable small decisions, each one taken without much thought for the whole.

The part of AI that looks back at us

I work with AI, and I build with it. I have seen what it can give someone who has an idea but lacks the resources to pursue it, or someone buried beneath work that consumes the hours they should spend exercising judgement. I would not want to abandon that possibility out of fear.

But I also know that AI is not quite like dynamite or a nuclear reactor. It increasingly mediates decisions, produces persuasive language, imitates human voices and can act through other systems. The outputs are not always predictable, and the people affected are not necessarily the people who chose to use it. Calling it neutral can become an excuse for refusing to examine its design.

And that is where the familiar pattern returns.

If we reward companies only for making AI more persuasive, more addictive, cheaper than a worker, and more deeply embedded in everything we do, we should not be surprised by what they optimise. If organisations replace judgement before understanding what that judgement was protecting, the failure will not belong to the model alone. If we let children learn to trust a fluent answer before they learn how to question one, we will have made another choice about the kind of society we want.

Australia has established an AI Safety Institute to examine emerging capabilities and harms and support regulators. That matters. Rules, evaluations, safeguards and consequences matter. We cannot rely on individual virtue to manage risks created at industrial scale.

But regulation tells us what we must not do. It cannot, by itself, tell us what we should be trying to become.

What are we choosing to optimise? Who benefits when the system works exactly as designed? Who carries the cost when it does not? And who still has the authority to say no?

Those are not questions for engineers alone. They belong to parents, teachers, employers, politicians and the rest of us who will live with the answers.

The lesson we did not teach

This is why I believe regulation is essential. We have had enough history to know that good intentions are not a safety mechanism. We regulate medicines, nuclear materials and explosives because their benefits do not erase their capacity to harm. We should be no less serious about a technology that can imitate trust, shape opinion and reach a child in the privacy of their bedroom.

But I keep coming back to education.

Look at what happened with social media. We gave young people a window into the world, then allowed much of what came through that window to be selected by its capacity to hold their attention. We worried about screen time, and sometimes forgot to ask what those hours were teaching.

The polished lives. The effortless wealth. The influencer who appears to succeed without learning a craft, building anything useful or showing the years of work behind the frame. Of course there are creators who teach, inspire and contribute something worthwhile. That is precisely why we should have taught children to recognise the difference.

Not every young person believed the illusion. Many were sceptical from the beginning. But too often we handed them the platform before we handed them the tools to understand it. We taught them how to use an app and assumed that meant they knew how to live with it.

I wonder whether we are preparing to make the same mistake with AI.

This time the feed will not only show us what to admire. It will be able to produce an endless supply of things to admire, fear, agree with or imitate. Images without photographers. Opinions without reflection. Articles without research. Confidence without knowledge. A flood of AI slop, cheap to produce and difficult to escape, competing with the careful work of people who still take the time to think.

And somewhere in that noise will be a child learning that a convincing answer is the same as a true one.

We can make deepfakes unlawful, require safer products, hold providers accountable and demand transparency. We should. These are not alternatives to education; they are the conditions that make education possible without leaving children to defend themselves against systems built to influence them.

Yet no law can sit beside a young person for every conversation, explain why an answer sounds plausible but is wrong, or teach them that popularity is not wisdom. That work belongs to families, schools and communities. It also belongs to the adults who should be willing to admit when they do not know something.

Education cannot simply mean teaching children to prompt an AI tool or preparing them for jobs that may change before they finish school. It means teaching them to ask where information came from, what is missing, whose interests are served, when to seek another human being, and when not to use the technology at all.

Perhaps the lesson is not to make our children afraid of AI. It is to make them harder to fool, less willing to outsource their judgement, and more interested in what is real than in what is impressive.

That would be a better inheritance than another set of terms and conditions nobody reads.

What will we make of this one?

The history of technology is not simply a story of good inventions corrupted by bad people. It is more complicated, and less comforting. We bring our ambitions, fears, generosity and greed into the things we make. Then the things we make change the conditions under which we live, and sometimes change us in return.

The internet did not invent greed. Nuclear fission did not invent war. Antibiotics did not invent short-term thinking. And AI did not invent our willingness to surrender responsibility when something easier comes along.

It may, however, magnify all of them.

I am not worried only about a future in which AI becomes powerful enough to act against humanity. I worry about a nearer future in which it becomes powerful enough to carry out our worst priorities exceptionally well, while giving us one more place to point when we are asked who made the decision.

We have become quite good at inventing remarkable tools. The harder work has always been deciding what deserves to be done with them.

Perhaps that is the question this generation must answer before the technology answers it for us.

Not whether AI will become a monster.

Whether we will teach it to be one.

Sources and further reading