AI systems make decisions. But the humans who build, deploy, and govern them make choices about what data to use, what to optimise for, and what trade-offs are acceptable. Those choices are ethical choices, whether they are treated as such or not.
When ethical dilemmas arise in AI development and deployment, having a structured way to reason through them is more useful than relying on instinct or compliance checklists. Five ethical lenses, drawn from established moral philosophy, offer exactly that.
1. The Rights Perspective
This lens focuses on moral rules, rights, and duties. It asks: does this system respect the dignity and autonomy of the people it affects? Are our choices the sort we could find universally acceptable?
In the context of AI, this perspective highlights the importance of informed consent, privacy, and the right of individuals to understand and contest decisions made about them. Some rights, like human dignity, are absolute and cannot be traded off against other considerations.
2. The Justice and Fairness Perspective
This lens asks whether the benefits and burdens of an AI system are distributed fairly. Are some groups systematically advantaged or disadvantaged by the system’s outputs? Are vulnerable or marginalised populations excluded from participation or opportunity?
Fairness in this sense goes beyond legal compliance. It requires active attention to who is affected by the system, how, and whether those effects are proportionate and equitable.
3. The Utilitarian Perspective
This lens evaluates actions by their consequences, specifically whether they produce the greatest good for the greatest number. In AI, this means considering the full range of stakeholders affected by a system, not just its immediate users.
Utilitarian thinking is useful for broadening the frame of analysis. But it requires caution: the consequences of AI systems extend far into the future and across large populations, making confident calculations difficult. Moral consequences also go well beyond economic outcomes.
4. The Common Good Perspective
Where utilitarian thinking aggregates individual benefits and harms, the common good perspective asks a different question: what is the effect of this system on the health of communities and institutions?
An AI system that makes most individuals better off but undermines trust in public institutions, concentrates power inappropriately, or damages community cohesion may still be ethically problematic. This lens keeps the wider social fabric in view.
5. The Virtue Ethics Perspective
This lens asks not what rules we should follow, but what kind of people and organisations we want to be. It is situational and context-sensitive, requiring mature moral judgement rather than mechanical rule application.
For AI development, virtue ethics means building habits of honest, careful, and empathetic design. It means asking: would we want future generations of data scientists to use our current practices as an example to follow?
Using the Lenses Together
No single lens is sufficient on its own. The value of this framework lies in applying all five perspectives to a given situation and using the tension between them to surface considerations that any single approach would miss.
Concretely, this means asking: which option produces the most good and least harm? Which best respects the rights of all stakeholders? Which treats people fairly? Which best serves the broader community? Which reflects the kind of organisation we want to be?
This is not a process that slows down AI development. It is one that makes AI development more robust, more trustworthy, and more likely to produce systems that hold up over time.