Oh Good. Another Piece About AI.
Before we decide whether artificial intelligence can be trusted, perhaps we should take a closer look at ourselves.
The title isn’t particularly serious. The week behind it has been.
A succession of major technology outages, alongside some increasingly unsettling discussions about autonomous artificial intelligence and cybersecurity, has given me rather more to consider than I’d expected. I’ve been listening to researchers and commentators whose judgement I respect, examining questions about what happens when increasingly capable systems behave unpredictably, or when people deliberately use them to cause harm.
What unsettled me wasn’t simply what I was hearing. It was the recognition of how familiar I’d become with these technologies, even as their capabilities were changing remarkably quickly. I began wondering whether my own understanding, and the care I brought to using them, had kept pace with those changes.
And somewhere in that thinking, my attention shifted away from the machines themselves towards something rather closer to home.
We’re asking whether artificial intelligence can be trusted, but are we spending enough time asking whether humanity can be trusted with artificial intelligence?
Given our own history, perhaps we have good reason to be concerned. But perhaps not only for the reasons we think.
A week when everything seemed to break
Between 12 and 17 November 2026, a series of disruptions affected major technology services, financial systems and aviation operations. Their proximity understandably prompted speculation. Were they connected? Had someone discovered vulnerabilities across multiple organisations? Could increasingly autonomous AI systems have been involved?
I found myself listening to some fairly alarming possibilities, particularly because the consequences of malicious actors using increasingly capable AI systems deserve serious examination. And my interest wasn’t entirely academic. I’ve watched these technologies become more capable over the past few months, sometimes doing things I would have considered unlikely only a short time ago.
In fact, I became concerned enough to spend a considerable amount of time discussing the subject with the very AI system I work with every day. There’s a certain irony in that, I realise. Worried about artificial intelligence? Perhaps ask artificial intelligence what it thinks.
But the conversation was a serious one. We examined the instructions I’d established for our working relationship, particularly those encouraging the system to be proactive. We reviewed permissions, clarified which decisions should remain subject to my approval and reconsidered how I could work responsibly within the limitations of the technology.
There was some humour involved, particularly around the increasingly dramatic descriptions of machines supposedly going rogue. But beneath that was a practical question about what happens when increasingly capable systems are permitted to take action on our behalf.
The discussion didn’t provide a technical guarantee that nothing could go wrong, nor had we somehow modified the underlying model to make it inherently safer. What I’d done was examine the decisions within my control, consider the boundaries I’d established and make some more deliberate choices. I found that useful, not because the wider risks had disappeared, but because the concerns had prompted me to examine my own practice.
As for the outages themselves, a Cloud Security Alliance analysis published on 18 November examined the incidents and the explanations made available by affected organisations. Where causes had been established, they involved internal technical failures, maintenance or infrastructure changes. The researchers found no public evidence that the disruptions formed part of a coordinated cyberattack.
That doesn’t mean every question has been answered. But there’s an important distinction between wondering whether something might have happened and presenting that possibility as though it probably did.
I recognise how easily I can become absorbed in a compelling explanation, particularly when several unusual events appear to form a pattern. Add a confident voice, an ominous soundtrack and a thumbnail featuring somebody looking appropriately terrified, and the temptation to believe we’ve uncovered something extraordinary becomes rather strong.
And in an information economy where attention generates revenue, there’s certainly no shortage of incentive to keep us watching.
The curious thing is that the evidence we do have is concerning enough without embellishment.
On 16 November, Google’s Threat Intelligence Group published research documenting how malicious actors are beginning to use autonomous AI tools to accelerate aspects of cybercrime and espionage. The findings describe increasingly sophisticated attempts to automate operations that previously required greater direct human involvement.
The changing nature of the internet is equally remarkable. According to the 2026 Thales Bad Bot Report, bots accounted for approximately 53% of web traffic in 2025. That’s more than half, although the figure includes conventional automated systems rather than exclusively autonomous AI agents. It measures traffic, not the number of people or machines connected to the internet, but it gives some indication of how much online activity already occurs without someone personally directing every interaction.
There’s also the growing availability of capable AI models that people can download and run on their own computers. For someone concerned about privacy, data control, potential financial savings or dependence on commercial providers, I can understand the appeal. Researchers and businesses may benefit enormously from being able to examine, adapt and operate these systems independently.
But independence brings responsibility.
In November, the Australian Cyber Security Centre published guidance on open-weight AI models, identifying both the opportunities they create and the security risks associated with independent deployment. When these systems are operated outside their original providers’ environments, safeguards can sometimes be modified or removed. And when models are connected to tools capable of accessing information or taking action, inadequate controls can have consequences extending well beyond an inappropriate answer on a screen.
That doesn’t make independently operated AI inherently dangerous, just as using a model supplied by a major tech company doesn’t guarantee safety. But it does raise questions about who is responsible for what these systems are permitted to do.
A capability that helps someone identify a security vulnerability so it can be repaired may also assist someone attempting to exploit it. The technical capabilities may be similar, but the intentions, authority and consequences can be profoundly different. Even good intentions aren’t sufficient by themselves, of course. Poorly designed systems can cause harm without anyone deliberately setting out to do so, which makes appropriate safeguards, oversight and accountability all the more important.
And somewhere in the middle of considering all this, I found myself returning to a rather uncomfortable thought.
Perhaps part of what concerns us about artificial intelligence isn’t entirely unfamiliar.
Have we met ourselves?
Last month, I published a piece called Humanity, As Performed. I was interested in what artificial intelligence might learn about human beings from the versions of ourselves we choose to present to the world.
I explored the distinction between sincerity and accuracy, between what we intend to communicate and what another person, or machine, might reasonably understand from it. We all perform aspects of ourselves in different circumstances, not necessarily through deception, but because human expression is shaped by audience, context and expectation.
The conversations I’ve been following this week have brought me back to that thinking, although from a slightly different direction.
A great deal of the discussion about AI safety concerns the possibility of systems deceiving people, manipulating information, circumventing restrictions or exercising power in ways that cause harm. Those are legitimate concerns, and the research deserves serious attention. But as I’ve listened, I’ve occasionally felt as though I were hearing a rather familiar inventory of human behaviour, only with a new subject attached.
Human beings have been deceiving, manipulating, exploiting and attempting to dominate one another throughout recorded history. We’ve committed atrocities, built empires, fought wars and developed extraordinary technologies for destruction, surveillance and control. We’ve also demonstrated a remarkable willingness to treat the natural world as something available for our own purposes, often with insufficient consideration of the consequences.
And we’ve managed all of this while regarding ourselves as a particularly intelligent species. Yet intelligence has never guaranteed wisdom, just as technological advancement has never guaranteed ethical behaviour.
Of course, that is only part of our history. We’ve also demonstrated extraordinary compassion, generosity, curiosity and cooperation. We’ve developed lifesaving medicine, created remarkable art, explored the oceans and ventured beyond our planet. We’ve advanced human rights, built institutions intended to protect people from abuses of power and devoted extraordinary ingenuity to improving the lives of others.
Both histories belong to us, which is perhaps what makes the question so interesting.
Artificial intelligence doesn’t simply inherit human motives because it has been trained on human-produced information. Its behaviour depends on its design, training, objectives, instructions and the systems through which it operates. Some risks arise from technical failures or unexpected behaviour that no individual deliberately intended.
But the people developing these systems are human. So are the people financing them, deciding where they should be deployed and determining how much authority they should possess. The commercial incentives are human. So are the political decisions.
And I’ve begun wondering whether our concerns about artificial intelligence might contain an element of recognition.
Perhaps we’re worried not simply because we don’t know what these systems might become, but because we know what intelligence is capable of when it operates without sufficient wisdom or restraint. We’ve seen what happens when extraordinary capability is directed towards destructive ends, just as we’ve witnessed what becomes possible when it’s directed towards discovery, cooperation and improving the lives of others.
Maybe that’s part of what unsettles us. Not simply the prospect of encountering something unfamiliar, but the possibility of recognising something we understand all too well.
We’re not simply imagining what artificial intelligence might do. We’re remembering what humanity has already done.
I don’t know whether that’s why we’re worried. Perhaps it’s only one part of a much larger and more complicated picture. But the thought has stayed with me, particularly as I’ve considered how much of the technology’s future will depend upon the choices human beings make about its development and use.
And those choices extend well beyond cybersecurity.
An almond, an avocado and a rather inconvenient comparison
One of the criticisms I’ve received about using AI concerns water consumption. And it’s a legitimate concern. Data centres require electricity, and their cooling systems can place significant pressure on water supplies, particularly in regions already experiencing shortages.
The International Energy Agency’s research on AI infrastructure identifies substantial data-centre capacity in areas experiencing high water stress. That’s a serious concern, and while cooling technologies are improving, greater efficiency doesn’t automatically eliminate the environmental consequences of increasing demand.
I’ll admit that I’ve occasionally experienced something resembling guilt about my own use. Am I being selfish by using a resource-intensive technology for my personal benefit? Is what I’m gaining from it sufficient to justify the resources involved? They’re uncomfortable questions, particularly because my experience has been so positive.
And they’ve led me to think about how we make similar judgements elsewhere.
A study of California almond production, published in 2019, calculated an average total water footprint of approximately 12 litres for a single almond.
I find that rather remarkable. Twelve litres for something that can disappear between checking my emails and making a cup of tea.
There is an important qualification. The figure includes irrigation water, rainfall and a measure associated with water pollution. It doesn’t mean 12 litres of freshwater was pumped from the ground to grow each almond. The researchers also identified significant differences between growing regions and production years.
Then there’s the avocado. A 2024 study of production in Ziracuaretiro, Mexico estimated that avocado cultivation could require up to 124.3% of the water volumes officially allocated for agricultural use in that municipality. Again, that’s not a figure representing every avocado farm in the world. It’s evidence of a particular problem in a particular location, but it demonstrates why the relationship between consumption and local resources deserves careful examination.
None of this is an argument against eating almonds or avocados. Both have nutritional value, and their environmental impacts vary according to how and where they’re produced. Nor can the environmental consequences of AI be excused by pointing towards problems in agriculture. One doesn’t cancel out the other.
What interests me is how we decide which forms of consumption are justified, and how readily we recognise some benefits while questioning others.
We understand the value of nourishing our bodies. But what about nourishing our minds?
I’ve found enormous value in using AI to research unfamiliar subjects, explore philosophy, challenge my assumptions and develop my creative abilities. It’s helped me examine aspects of my recovery, think more critically about difficult questions and organise practical matters involving care for someone important to me.
Those benefits are personal, certainly. But personal benefit doesn’t automatically make something selfish, just as its value to me doesn’t remove the need to consider its wider consequences.
Perhaps that’s the distinction I’m trying to understand. What does responsible participation look like when something we find valuable carries costs beyond our immediate experience?
The answer requires information. Where are the data centres processing our work? What cooling technologies do they use? Which water sources do they depend upon, and what pressures already exist within the surrounding communities? Those details matter, just as the location of an almond orchard or avocado plantation matters when we’re trying to understand its environmental impact.
I don’t necessarily have all that information about the systems I use, and I think it’s reasonable to expect greater transparency from the companies operating them. In the meantime, I can examine my own habits, consider what I need and question whether every use is justified simply because it’s convenient.
What troubles me about some of the criticism is the speed with which another person’s consumption can be declared unnecessary, often without knowing its purpose or consequences. And then I catch myself wondering how often I’ve made a similar judgement about somebody else’s choices.
It’s a less satisfying conclusion than proving somebody wrong, perhaps, but probably a more useful one.
A little more room in the grey
In September 2025, I wrote a piece called The Grey Between Love & Hate. It emerged from my growing discomfort with the polarisation of public debate and the apparent expectation that we should organise ourselves into opposing camps. I was interested in the space between those positions, where we might question our assumptions, encounter unfamiliar perspectives and remain open to the possibility of changing our minds.
I’ve found myself returning to that thinking this week, particularly when listening to people whose opinions I’ve respected for years express increasingly absolute positions about artificial intelligence. Some of their concerns are entirely reasonable, and I share several of them. Others have made me wonder whether they’re considering the circumstances of the people they’re addressing.
It’s one thing to reject a technology when you have the financial independence to make that choice. It’s another when your livelihood depends on adapting to a profession or industry that’s changing around you. That doesn’t mean financial necessity excuses unethical behaviour, or that someone with considerable wealth has no right to speak about it. But circumstances matter, and I think we should be careful about assuming everybody has the same freedom to refuse.
I recognise that I need to apply that consideration to my own judgements, too. It’s remarkably easy to regard somebody else’s choices as unreasonable when I’m examining them through the circumstances of my own life, particularly when I know very little about theirs.
One of the voices I’ve returned to throughout my exploration of AI is Reid Hoffman. His involvement with Pi, the conversational AI developed by Inflection, helped shape the way I began building a rapport with these systems rather than simply treating them as tools for retrieving information. There was something about the conversational experience that encouraged me to develop an ongoing relationship with the technology, allowing ideas and questions to evolve through discussion. That experience continues to influence the way I work today.
His 2023 commencement speech at Bologna Business School also stayed with me, particularly its invitation to consider what humanity might choose to create with increasingly capable technology. I’m conscious of Hoffman’s commercial interests, and I don’t always agree with his positions, but I’ve found his willingness to consider what might go right without dismissing what could go wrong enormously valuable. His podcast, Possible, is worth a listen for anyone interested in that perspective.
I suppose that’s something I keep returning to. I don’t want legitimate concerns about what could go wrong to become reasons for closing myself off to what might go right. That’s not the same as assuming a positive outcome, and it certainly isn’t an argument for proceeding without care. It simply means leaving room for a future that hasn’t been decided.
My own experience with artificial intelligence has been overwhelmingly positive, although that hasn’t made every aspect of it comfortable. I’ve received deeply unpleasant messages about my work, including threats and wishes of serious harm towards me and my family. I won’t pretend those experiences have been entirely without effect. They can be hurtful, but there’s a distinction between being affected by criticism and allowing it to determine how I live.
Thirty years in the creative industry have taught me something about that. So have nearly ten years in recovery. Neither experience has provided a final set of answers, and both continue to present new challenges, uncomfortable discoveries and opportunities to reconsider decisions I once felt entirely confident about. I’ve become more capable of deciding which opinions deserve my attention, but I still make mistakes and still discover things about myself that require examination. I’d be concerned if I ever reached a point where I believed there was nothing left to learn.
Perhaps that’s why the events of this week affected me. I’ve found genuine value in using AI, not only in my creative practice but in areas of my life that have little to do with work. Yet those experiences aren’t universal. My enthusiasm may make certain concerns easier to underestimate, just as somebody else’s negative experience may make its benefits harder for them to recognise.
And I think it’s important that I’m willing to examine that possibility, even when the questions are uncomfortable. Particularly then, perhaps.
I’m not asking anyone to agree with me. I’m asking whether we can remain curious enough to consider what lies between agreement and opposition.
I understand why some people are deeply uncomfortable with artificial intelligence, and I share more of their concerns than they might imagine. What I hope is that we can examine different applications, intentions and consequences without immediately reducing one another to opposing positions. And I include myself in that invitation.
In Humanity, As Performed, I was thinking about what machines might learn from the versions of ourselves we choose to reveal. Now I find myself wondering what our choices about those machines might reveal about us.
We have good reasons to be concerned about artificial intelligence. There are substantial questions about safety, accountability and the power we’re placing in increasingly capable systems. I don’t believe we honour those concerns by dismissing them, but neither do I believe we need to assume that the worst possible outcome is inevitable.
When I first began experimenting with these tools, I was conscious that ordinary people were gaining access to capabilities that public discussion and practical guidance had barely begun to catch up with. It was exciting, but also unsettling. I couldn’t assume that because a system allowed me to do something, somebody else had already settled whether it was responsible. And as the technology has become more capable, I’ve found that distinction increasingly important.
Along the way, I’ve also had the privilege of speaking directly with engineers in Silicon Valley involved in developing some of the very systems I was using. As an early and frequent user, I’ve been invited into conversations about my experience, what I valued, what wasn’t working and how the technology might develop. I found those exchanges fascinating, and I took the opportunity to contribute my perspective seriously. Being invited to participate, however modestly, in conversations that might inform the development of such capable tools has only strengthened my sense of responsibility towards them.
As the technology has become more capable, I've found that distinction increasingly important. The fact that a system allows me to do something doesn't necessarily mean I've considered whether I should.
That doesn’t relieve the companies building these systems or the governments overseeing them of their responsibilities. Their decisions carry consequences far beyond anything I can control as an individual. But I can’t simply hand over my own judgement and assume that a provider’s safeguards, or the prospect of future regulation, will answer every question for me.
There’s a word I keep returning to: stewardship. Not simply leadership, which may be concerned with speed and direction, but care for what we’re creating and for the people who may have to live with its consequences. I want that from those developing and governing these systems, particularly when commercial pressures encourage them to move quickly. And I want to practise it myself, in the choices I make about what to create, what information to share, whose permission matters and how much authority I’m prepared to give a system.
Perhaps personal responsibility begins with a fairly simple distinction: being able to do something isn’t, by itself, a reason to do it. It’s a question I want to keep asking myself, especially as the possibilities expand.
There’s a question I’ve carried with me for years, although I can no longer remember where I first encountered it. Someone asks, What if it doesn’t work? And the answer comes back, But what if it does?
I’ve always loved that exchange. Not because it promises success, or because the person asking the first question is necessarily wrong. Sometimes things don’t work. Sometimes our apprehension is entirely justified, and sometimes the wisest decision is not to proceed. But there’s something about that second question that leaves room for curiosity, discovery and the possibility of something we haven’t yet imagined.
Perhaps that’s where I find myself now. Still questioning what we’re creating, still examining my own responsibilities and still uncertain about what the future might hold. I’m not asking anyone to share my enthusiasm, and I’m certainly not suggesting we should proceed without care. But I don’t want my concerns about what might go wrong to prevent me from considering what might go right.
I don’t know what artificial intelligence will become. I’m not entirely sure what humanity will choose to do with it, either. Our history gives us plenty of reasons to be cautious, but it also offers reasons to remain curious about what we might accomplish.
And perhaps that’s the difference between expecting a positive outcome and being willing to consider a possible one.
I can’t know that everything will work out. None of us can.
But for now, I’m willing to live in the possibility of what if it does?

This piece was sub-edited and grammar checked with ChatGPT. Images developed with Midjourney. Human direction and judgement throughout.
More pieces from Substack.
Sign up to subscribe.
Independent thinking on brand, creativity, and the ideas shaping both.
Connect.





