Part 1 mapped the wider ethical terrain around AI. This second post asks what those concerns might mean for the way AI is designed, developed and deployed, drawing on six principles that put purpose, human agency, fairness, planetary limits, wider system effects and accountability at the centre.
In Part 1, seeing the wider ethical picture around AI development and use, I looked at ethical concerns around AI at three connected levels: the wider structures within which AI is developed, the organisational choices through which it enters our work, and the everyday practices through which people use it.
That wider view matters because many of the concerns being raised cannot be resolved simply by asking individuals to use AI more carefully. Questions about energy and water use, labour, ownership, concentrated power, changing employment, and effects on education and public information are shaped much further upstream.
Looking upstream brings a different set of questions into view. Rather than beginning with what the technology can do, we can start with what we are trying to achieve, for whom, and within what wider social and ecological setting.
We already know quite a lot about good design
Many of the principles that might guide better AI development are not new, or unique to AI. Over several decades, participatory development, systems thinking, sustainable design, action research, adaptive management and related traditions have accumulated practical lessons about introducing technologies and other interventions into complex settings. They remind us that technologies have purposes, that different people experience their benefits and costs differently, that knowledge is distributed, that interventions interact with wider systems, and that consequences often unfold in ways that could not have been fully predicted at the outset.
These traditions also encourage us to question the boundaries around a design problem. It is easy to define the relevant system narrowly around a developer, customer or immediate user. But people live and work in places. We have neighbours and communities. We are linked to people elsewhere through supply chains, labour, information and economic systems. And all of this takes place within ecological systems on which we depend.
Seen this way, asking whether an AI system works is not enough. We also need to ask what it is intended to improve, who gets to define that improvement, who may benefit or lose, what else it may change, and whether the resources it requires are justified by the value it creates.
Participation is part of this. It is not necessary, or possible, for everyone to participate in every design decision. But good design should ask who needs to be involved, in what, at what stage, and with what influence. This is partly about fairness. It is also about knowledge. Designers rarely possess all the knowledge needed to understand how a technology will interact with the situations into which it is introduced.
Six principles for good design
Against that background, I find six principles particularly useful for thinking about better AI design and development. As the following diagram highlights They are not new or unique. Rather, they bring together practical expectations that have emerged across different fields and traditions, and which may also be useful to people designing, developing, deploying and governing AI.
They are not intended as a comprehensive framework for ethical AI. They are broad principles for asking better questions about purpose, people, resources, wider consequences and responsibility. Participation is not a separate seventh principle because it runs through the others, helping shape purpose, bring different knowledge and perspectives into design, make consequences more visible and strengthen accountability.
The six principles overlap rather than operating as separate tests. Questions about jobs, for example, can involve purpose, human agency, fairness and wider system effects. Energy and water use raise questions about planetary limits, but also about whether the value being created justifies the resources involved. The sections that follow look at each principle in turn.
1. Design for worthwhile purposes
The starting question should be what we are trying to achieve, rather than what the technology is capable of doing. Technical capability, novelty, market opportunity and productivity can all be relevant, but they are not purposes in themselves. A system may work extremely well at doing something without making the activity particularly worthwhile.
This means asking some basic questions early. What problem are we trying to address? Who sees it as a problem? Who is expected to benefit? Are there other ways of achieving the same outcome? What might be displaced or weakened in the process? These questions become particularly important when growing technical capability itself begins to be treated as sufficient reason to automate an activity.
2. Strengthen human agency
Technology can extend human capability. It can help people gain access to information, explore alternatives, communicate more easily and carry out tasks that would otherwise take considerable time. But it can also create dependency, narrow opportunities to learn, reduce professional discretion and quietly shift decisions towards systems that people do not fully understand or control.
Better design should therefore aim to strengthen people’s capacity to understand, decide, learn and act. Keeping a human formally “in the loop” is not enough if that person’s role has been reduced to accepting recommendations they have little capacity or time to question. We should be asking whether a system supports human judgement and capability over time, or gradually substitutes for them.
3. Share benefits and burdens fairly
AI systems create benefits, costs and risks, but these are rarely distributed evenly. The organisations receiving productivity gains may not be the same people whose jobs are changed or lost. People whose creative or intellectual work contributes to AI systems may have little influence over how it is used. Environmental or infrastructure costs may fall on communities far removed from the users receiving the benefits.
Fairness therefore requires more than checking whether an algorithm treats individual users consistently. It also means asking who gains, who pays, who carries the risks and who has a meaningful say. Those questions need to be considered across the development chain, including data, labour, infrastructure, ownership and deployment. Ownership and control matter here because those able to set the terms of development and deployment may also be better placed to capture the benefits and shift some of the costs elsewhere.
Participation is especially important where those making design choices are not the people most likely to experience their consequences.
4. Work within planetary limits
The material demands of AI were part of the wider ethical picture considered in Part 1. Here the design question is what follows from recognising them. Improvements in the efficiency of AI systems matter, but so does the overall scale at which they are being developed and used.
Environmental costs should therefore be part of the design brief rather than treated as something outside it. This does not mean expecting technologies to have no environmental footprint. It means asking whether resource use has been made visible, whether it can be reduced, and whether it is proportionate to the value being created. This brings purpose back into the discussion: what kinds of value justify resource-intensive uses of AI, and who should be involved in making that judgement?
5. Design for the wider system
Technologies do not simply enter an existing setting and leave everything else unchanged. People and organisations adapt around them. Practices change, expectations shift, new dependencies develop and activities that were previously valuable may become less visible.
This makes it important to look beyond the immediate user and intended function. How might widespread use affect education, professional practice, employment, communities, information systems or democratic institutions? What behaviours might the system encourage? What might it displace? And what happens if millions of people and organisations respond to the same incentives?
Looking at the wider system does not allow us to predict everything that will happen. It can help make foreseeable effects more visible, but it cannot remove uncertainty about what will happen as a technology becomes widely used. That makes it harder to treat wider effects as someone else’s problem, while also pointing to the need for continuing learning and adaptation.
6. Remain accountable and adapt
No design process will identify every consequence in advance. Complex technologies interact with changing social systems, and effects that seem minor during testing can become significant when systems are widely adopted. Responsible design therefore cannot finish when a product is released.
Developers and organisations need ways of noticing what happens in practice, learning from unexpected effects, responding to challenge and changing course. People affected by a system need meaningful routes for questioning decisions and seeking redress. Monitoring should look not only for technical failure but also for changes in behaviour, capability, relationships and wider system effects. Accountability also needs to remain identifiable. If responsibility is distributed so widely that nobody has the authority or obligation to act, it is difficult to call it accountability at all.
These principles work together
The six principles are deliberately broad, and they overlap. A question about employment, for example, may involve human agency, fairness and wider system effects. Concerns about water or energy use connect planetary limits back to questions of purpose: is this use of resources justified by what the system is helping us achieve? Misinformation raises questions about the wider information system as well as accountability. Concentrated control over powerful AI systems raises questions about fairness, agency and whose purposes are being served.
The principles will also sometimes pull in different directions. A socially valuable use of AI may require significant resources. Increasing access may create new risks. Protecting people from harm can conflict with their desire for autonomy. Good design does not remove these tensions. It makes them more visible and gives people a better basis for discussing and navigating them.
Locating these principles
These principles also reflect the way I have come to think about design and intervention through work in environmental management, participatory practice, systems thinking, evaluation and adaptive management. Across these fields, I have repeatedly encountered the limits of approaches that define a problem too narrowly, treat expert knowledge as sufficient, or assume that an intervention will work as intended once it is introduced.
I find the six useful because they bring some of that accumulated learning into the questions now being raised by AI. I offer them in that spirit: not as a new framework, but as a set of expectations that may help others think about what better AI development and use could look like.
Concluding comments
There is no shortage of principles, standards and frameworks for responsible AI, and these six are not offered as another claim to have captured everything that ethical AI requires.
Their purpose is simpler. They help shift the starting point from asking what can we make AI do? towards asking what are we trying to achieve, with whom, and what kind of social and ecological system are we helping to create.
These questions are not unique to AI. They reflect lessons learnt over many years across many fields and communities of practice, through working with technologies, environmental interventions, organisations and communities in situations where knowledge is incomplete, different perspectives matter and consequences unfold over time.
Responsibility does not rest only with technology developers. Governments, organisations, professions, communities and users all help shape how technologies are introduced and what becomes normal practice. But responsibility is not evenly distributed either. Those with greater power to shape AI systems, determine their purposes and influence the conditions under which they are used carry greater responsibility for their consequences.
Better individual and organisational use of AI still matters. But responsible use sits within this wider picture. We should also be able to expect the technologies themselves to be designed and developed in ways that support human agency, distribute benefits and burdens more fairly, respect ecological limits, recognise wider system effects and remain open to scrutiny and change.
For the wider framing behind this post, see Part 1: Seeing the wider ethical picture around AI development and use. Related thinking about purpose, participation and design in complex settings is explored in Designing together: reflections on co-design in complex settings.You can also explore the wider collection of reflections and resources on AI for reflective practice, research and collaboration.
[*1 Image by Jintana / Adobe Stock]
[*2 Graphic: Will Allen, with ChatGPT assistance, 2026]
