AI is a choice.
Notes on Virginia Dignum's work on responsible AI, and what it means for organisations adopting AI now.
Know the field.Design how business and technology serve people.
Filmed by David Pickup.
AI is designed. People in organisations choose what it does, whose values it carries and how fast it moves. Virginia Dignum, Professor of Responsible AI and director of the AI Policy Lab at Umeå University, puts it simply: AI does not happen to us. These notes follow her argument and what it asks of the organisations we work with.
The trouble with "AI first"
Strategy
Many organisations now run an AI-first strategy: adopt AI everywhere, and ask questions later. In their 2025 submission to the United Nations, Dignum and colleagues at the AI Policy Lab point to five problems with it:
- It moves the organisation from governance to unchecked deployment.
- It ignores the evidence that AI's productivity gains are often limited.
- It skips the two questions that come first: why AI, and whether AI at all.
- It creates a democratic deficit, because the people affected are not consulted.
- It builds pressure to adopt without evaluation.
Field One starts from those first questions: who the service is for, what they need, and where AI earns its place.
Question Zero
Before adoption
Before any build, ask whether AI is the right option at all. Dignum calls this Question Zero: ask why before asking how.
Every design decision then carries three further questions. Who should decide? Which values should be considered? Whose values, and how do we prioritise them?
Who decides, what evidence, which remedy
Accountability
In July 2026, over 1,300 employees of frontier AI companies signed Pacing the Frontier, a statement asking the US government to support an international effort to deliberately pace frontier AI development. Dignum's response, "Pacing is not direction", sets out three tests for any such call.
Who decides?
Identifiable people in identifiable organisations. Describing their choices as a force of nature makes accountability disappear.
What evidence?
When the labs define the risk, only the labs hold the evidence. Independent audit with real access is the test.
Which remedy?
Binding rules, verification and liability, set up so that the safe choice is also the cheap one.
"AI is not the weather. Speed is one number; direction is a choice." Virginia Dignum
We govern what we count
Measurement
Most of AI's impact goes uncounted. Dignum describes three levels of impact:
The consequences for governance follow. Accountability follows measurement, so whatever goes unmeasured has no one answering for it. What is reported becomes what is designed for, which makes choosing a metric a policy decision. Measurement should also reach the planetary level: planetary boundaries, equitable access and societal cohesion. And Question Zero applies to metrics too: agree on what counts, and who audits it, before building.
Well over 1,000 sets of principles
Principles
The world already has more AI principles than any organisation can read.
| Global | The UNESCO Recommendation on the Ethics of AI, adopted by all member states in 2021. The OECD AI Principles, with 47 adherents and updated in 2024. The Global Partnership on AI, now run jointly with the OECD. The G7 Hiroshima Process principles. The Council of Europe Framework Convention, the first legally binding international AI treaty, signed by 21 parties including the EU, the United States, Japan and Canada. |
|---|---|
| Regional | The EU AI Act across 27 member states. The Santiago Declaration for Latin America and the Caribbean. |
| National | More than 930 initiatives in 71 countries. |
| Others | Over 200 guidelines from NGOs, academia and private bodies. |
They agree on a common core, set out in the EU High-Level Expert Group's seven requirements: human agency and oversight, technical robustness and safety, privacy and data governance, transparency, diversity and fairness, societal and environmental wellbeing, and accountability. The IEEE's Ethically Aligned Design turns these into questions any team can ask: does the system respect human rights, does it improve wellbeing, are its makers and operators accountable, and is it transparent?
What can be done now
Action
Governance is wider than regulation, and none of this needs to wait for global agreement.
- Enforce what exists. GDPR, the Digital Services Act, product liability and sector law already apply to AI.
- Buy with conditions. Public and private buyers can require documentation, testing, auditability and liability without any new law.
- Build independent capacity. Mandatory disclosure, third-party audit with real access, compute reporting and public evaluation research.
- Ask Question Zero first. Is AI needed here at all, is it the best option, and what are the alternatives?
Where the work happens
Field One
An AI system lives inside a socio-technical ecosystem: an organisation, its services, its people and the rules around them. Ethics, regulation and governance apply to that whole ecosystem, and no technical fix removes the harms on its own.
Field One brings the principles into everyday decisions: procurement conditions, what gets measured, service design, sign-off, and who answers when something goes wrong. Humane Technology by Design →
Governance is how we decide what to build, who answers for it, and how we correct course.
Virginia Dignum, Umeå University AI Policy Lab
Sources
- Virginia Dignum, Responsible Artificial Intelligence (Springer) and The AI Paradox.
- Virginia Dignum, "Pacing is not direction", Thinking in Public, 31 July 2026.
- Virginia Dignum, "No, AI will not kill us all", Thinking in Public, 12 September 2026.
- Dignum, V., Dignum, F., Fjaestad, M., & Tucker, J. E. (2025). Submission to the United Nations to identify the terms of reference and modalities for the establishment and functioning of the Independent International Scientific Panel on AI and Global Dialogue on AI Governance. AI Policy Lab, Umeå University.
- Kaack et al., Nature Climate Change 12, 518–527 (2022). Gaffney et al., Nature Sustainability 8, 467–469 (2025).
- Pacing the Frontier (July 2026). OECD AI Principles. Council of Europe Framework Convention on AI.