Debates about AI often sound polarised, but much of the disagreement comes from people beginning with different assumptions about what is at stake. This post maps two overlapping levels of concern – the structural conditions shaping AI systems, and the practice-level tensions that surface when we actually use them. Making these layers visible does not resolve the debate, but it can clarify what is actually being argued about.
Earlier posts in this series have focused on how AI is being used in reflective, collaborative, and collective work. They have explored AI as a thought partner, and as something that can sit, carefully, within group processes.
This piece steps back from questions of use. It recognises that many concerns about AI sit beyond technique, and that people are often responding to very different ethical worries.
Some of those worries relate to infrastructure, labour, and power. Others arise in the everyday experience of working with these tools. Public debates can sound sharply polarised, ranging from careful adoption to outright refusal. But much of that disagreement reflects different starting points about what is at stake.
It helps to distinguish structural concerns that sit at the level of wider systems from those that show up in day-to-day practice.
The structural level
The first set of concerns relate to the wider systems within which AI is developed and used. These are not mainly about individual behaviour. They are about infrastructure, labour, and power.
AI as environmental and material infrastructure
From this perspective, AI is understood primarily as physical infrastructure rather than abstract software. It depends on data centres, energy supply, water use, mineral extraction, and global supply chains. These material systems face growing pressure in some regions, and concerns about local grid and water impacts have been documented.
For example, recent analyses of data centre water and energy use in AI training have highlighted rising local pressures on electricity grids and freshwater supplies. A recent U.K. report on water use in AI and data centres provides one such illustration. At the same time, hardware and model efficiency continue to improve, even as overall demand grows.
Of course, not all AI systems operate at the scale of large foundation models. Some run locally or on smaller infrastructures. But the current wave of generative systems that shape public debate and organisational practice does depend on substantial centralised computing resources.
AI as appropriation of labour and knowledge
A second set of concerns focuses less on infrastructure and more on what AI systems draw on. Training practices range from fully licensed datasets to highly contested forms of large-scale scraping, and it is this variation that fuels much of the current debate. For many critics, this raises questions about whether these systems amount to a form of appropriation of labour and knowledge.
Here, ethical attention centres on authorship, consent, creative labour, and responsibility for interpretation. Questions arise about how training data is gathered, whose work is being used, and whether communities, researchers, or creators have meaningful agency in those processes. Public debates over scraped web text, artists’ work used without consent, and the use of open-source code in model training have made these questions more visible in recent years.
In research contexts, additional questions arise about how AI is used in analysis and interpretation, and where responsibility for meaning sits. Critics within this lens argue that drawing firm boundaries around what is acceptable use is an act of care rather than resistance to change. Refusal, in this context, protects the integrity of labour, craft, and responsibility.
AI as reinforcement of existing power structures
A third lens treats AI less as a tool and more as a form of governance. The most widely used generative systems are developed and controlled by a relatively small number of organisations. Platform design decisions shape defaults, and those defaults shape behaviour. Over time, this influences how knowledge is produced and circulated.
The issue here is not only what individual users do. It is who decides how systems are built, what values are embedded in them, and whose interests are served. In this sense, governance refers not only to formal regulation, but also to the quieter ways platforms shape defaults and expectations. From this perspective, regulation and collective oversight become central questions. The concern is about concentration of power and long-term institutional direction.
These three structural concerns help explain much of the wider debate. They show why optimisation, restraint, refusal, and regulation can all appear reasonable, depending on what is being protected. Other concerns, such as bias, discrimination, and harmful misuse, are also central to current debates, but are not explored in detail here. Questions about long-horizon existential risk, model alignment research, or detailed technical benchmarking sit beyond what I am trying to cover in this piece.
Alongside these debates, a number of organisations have articulated principles for responsible AI. For example, the UNESCO Recommendation on the Ethics of AI outlines considerations such as avoiding harm, fairness, privacy, transparency, human oversight, and accountability. These kinds of principles provide a useful orientation, particularly for organisations developing policy or governance frameworks. At the same time, they are typically expressed at a high level. In practice, the challenge lies in how they are interpreted and applied in specific situations, especially where trade-offs are involved.
But by themselves, these wider structural issues do not fully describe what happens when someone actually sits down and uses these systems.
The practice level
The wider system matters. Infrastructure matters. Labour and governance matter. Yet when we work with AI in writing, policy thinking, or facilitation, other questions appear as well. These concerns are more immediate. They arise in everyday use. Over time, they may also influence the wider system. Two issues in particular keep resurfacing in my own work and in conversations with others.
Reliability of reasoning
The first issue is about what the system produces. Generative AI systems write in fluent, confident language. They provide summaries, numbers, examples, and arguments. Often this is helpful. It can clarify structure or open up new angles. But the same fluency can also make weaknesses harder to spot. For example:
- References that appear plausible but cannot be traced.
- Statistics presented confidently without a clear source.
- Arguments that move quickly from description to conclusion.
These behaviours have been widely documented in current large language models and are often referred to as “hallucinations” or fabrication – though the term “hallucination” can be misleading, implying randomness rather than a predictable feature of how these systems generate plausible text.
When AI-generated material enters research, evaluation, or policy advice, we need to ask straightforward questions. Can this be checked? Is the evidence visible? Are limits and uncertainties acknowledged? In many settings, the risk is not dramatic failure. It is gradual drift. If unsupported claims are accepted because they are well written, standards can quietly shift. And over time, this will affect how trust is built in professional work.
What happens to our own thinking
The second issue is about what happens to us when we use these tools regularly. When drafting and synthesis are readily available, it becomes easier to hand over early thinking. This can save time. It can help move work forward. But there may also be changes in how we think.
- We may spend less time wrestling with difficult material.
- We may read more quickly and less deeply.
- We may move from developing arguments to editing generated text.
These shifts are subtle. They do not happen overnight. Yet they can influence how judgement is exercised and how groups deliberate together. If such changes become widespread, they do not remain personal habits. They begin to feed back into organisational and structural patterns. Expectations of expertise can shift. Decision-making cultures may adjust.
Some of these questions are also being explored by others working in practice. In a recent piece, Sophie Lambin reflects on the relationship between AI, speed, and the responsibility to think clearly. She points to how increased speed can subtly reshape how we process information and exercise judgement, raising questions that sit close to the practice-level concerns described here.
If fewer people retain deep subject knowledge, and more organisational work depends on AI-generated outputs, authority may gradually reorganise – much as we have seen influence shift toward dominant digital platforms over the past two decades. The analogy is not exact, but we have seen how dominant search and social platforms reshape how knowledge is accessed and valued. Similar shifts are at least plausible as generative systems become embedded in everyday work. In that sense, what begins as a practical convenience can contribute to wider governance questions.
Seeing both levels in practice
Holding both levels in view makes disagreement easier to understand. Some people focus on environmental impact, labour, and power. Others are concerned with reliability and the effect on human judgement. These levels are analytically distinct, but they are not sealed off from one another. Recognising this mapping helps avoid talking past one another and clarifies why neither careful individual use nor regulation alone resolves every risk.
If AI shapes the wider system and also sits inside everyday work, then responding responsibly requires more than taking a position for or against it. It requires deliberate examination:
- Not only asking whether we should use AI, but seeing how it responds to difficult or ambiguous prompts.
- Not only debating governance, but checking how it handles evidence, uncertainty, and edge cases.
- Not only naming risks, but noticing where its arguments become overstated, under-supported, or unstable.
That move, from mapping concerns to examining model behaviour in practice, is the next step.
Locating my own practice
My own engagement with AI is shaped by both the wider structural concerns and the practical issues described above. None of these disappear through careful use, and none feel abstract in my work. What this has led me to is not a fixed position, but a set of working habits. For example:
- AI is used purposefully, mainly as a support for thinking and preparation rather than as a substitute for judgement, relationship, or craft. Attention to scale and default use matters, and I am wary of practices that add speed or volume without improving sense-making.
- Source integrity remains important. When AI surfaces articles, statistics, or references, I check them before drawing on them. It helps keep evidential standards visible, rather than assuming fluency equals reliability.
- Questions about AI use need to stay visible in collective settings. That includes naming uncertainty, acknowledging uneven impacts, and recognising where responsibility sits beyond individual users.
Alongside this, I have become more deliberate about examining how these systems behave. When I use AI to explore an argument, I probe the output. I ask for sources. I check numbers. I test alternative framings. At times I ask for the assumptions behind a claim, or for the strongest case against it, to see whether the argument still holds together. This is less about catching dramatic failure than about understanding patterns of instability.
These responses do not resolve the wider tensions described above, and they are not sufficient on their own. But they point toward something that may be worth stating plainly before closing. Before asking how to use AI well, it may help to recognise what people are worried about when they object to it. Ethical clarity, in this space, may come less from finding the right answer and more from being explicit about which concerns we are responding to, and which we are not.
This post is part of a small set of reflections on using AI in practice, from individual use through to shared work and wider ethical considerations. You can explore the full set on the AI for reflective practice, research, and collaboration hub. A few related posts from that collection include:
- AI as a thought partner: reflections on collaborative practice and systems work
- AI prompts for shared thinking: a light framework for purposeful prompting
- Working with AI in the room: authorship, responsibility, and collective judgement
- AI in place-based practice: what is shifting
[* Image by Jintana / Adobe Stock]
