The person who does not fit the form
A woman is trying to report a change in her circumstances to a public authority.
The digital service is clean. The typography is accessible. The steps are numbered. Each field validates correctly. The interface has passed a usability test.
Her life still does not fit it.
The form assumes one address. She has two temporary places to stay. It assumes a stable employment category. Her work has changed three times in two months. It asks for documentation she cannot produce through the route the system recognizes.
The service is easy to use in the narrow sense.
It is difficult to live through.
This is the distinction human-centred technological development has to hold.
Technology can be usable without being humane. It can be efficient without being accessible to the people whose circumstances fall outside its model. It can invite feedback after the important decisions have already been made. It can put a person in the loop while keeping the problem definition somewhere else entirely.
A system is not human-centred because the human appears in the final usability test.
Definition
Human-centred technological development is an approach to designing, deploying, governing, and revising technology in which human needs, capabilities, rights, values, contexts, and lived consequences shape the problem definition as well as the solution.
It involves people early enough to influence what is built, examines the wider system in which technology will operate, and evaluates success through human and social outcomes alongside technical performance. In AI contexts, it also asks whether meaningful human agency, oversight, contestability, and competence are preserved after deployment.
The definition is intentionally broader than interface design.
A button can be perfectly placed inside a harmful process.
A model can be accurate against its benchmark and still operate inside a badly framed objective.
A service can score highly on satisfaction among the users who complete it while excluding those who leave before the survey appears.
Human-centredness therefore has to begin before the screen.
Human-centred is not one method
The literature contains a family of overlapping traditions rather than one standardized recipe.
Human-centred design and user-centred design emphasize understanding users, tasks, environments, iteration, and evaluation. Participatory design goes further by asking who gets to shape design decisions and how power is distributed in that process. Human factors and ergonomics examine interactions among humans and system elements and use design to improve both human well-being and system performance. Value-sensitive design brings moral and social values into technology design. Human-centred AI attempts to adapt these concerns to systems that learn, infer, predict, recommend, and sometimes behave in ways that are not fully specified in advance.
These traditions are related.
They are not interchangeable.
A project can interview users without being participatory. It can be participatory without adequately addressing rights. It can optimize human performance without expanding agency. It can describe itself as human-centred while leaving the institution's own incentives untouched.
The label should therefore never substitute for the method.
What the human-centred AI literature currently shows
Human-centred AI has grown rapidly, but the field is still consolidating its concepts and methods.
A 2025 scoping review by Bevilacqua and colleagues screened more than a thousand records and included fourteen studies from 2018 to 2023 that met its criteria. The review found recurring emphasis on human needs, trustworthiness, and augmentation rather than substitution, but also found that principles and guidelines were often vague, practical applications were limited, and standardized methodologies were lacking.
That is an important finding for this Knowledge Library.
The right response is not to abandon the term.
It is to use it more precisely.
Human-centredness should be visible in decisions: who participated, when they participated, what authority they had, what harms were considered, what populations were missing, what objectives were changed because of human evidence, and what happened when deployment produced consequences the design process did not predict.
If none of these can be shown, “human-centred” may be branding rather than an accountable development method.
The problem begins before design
Many technology projects begin with a solution-shaped question.
How can AI reduce support costs?
How can automation speed this workflow?
How can personalization increase engagement?
A human-centred process asks whether the problem has been framed too early.
Perhaps the support burden comes from a policy people cannot understand. Perhaps the workflow is slow because staff are correcting upstream data errors. Perhaps engagement is not the outcome users would choose if they understood how the system shaped their attention.
Problem framing is therefore one of the deepest human-centred acts.
It keeps the human from becoming the object that must adapt to a technological decision already made.
Capability, not only preference
Human-centred work often uses preferences, interviews, satisfaction, and behavioral observation.
These are valuable.
They are not sufficient.
People can prefer designs that are convenient in the short term and still lose capabilities they care about over time. They can adapt expectations to systems they cannot change. They can value a recommendation while being unable to understand or contest the data relationships that produced it.
This is where a capability lens strengthens human-centred development.
The question becomes not only “Do people like this?” but “What are people genuinely able to do, understand, refuse, challenge, learn, and become in relation to this system?”
That shift matters especially in public services, education, employment, health, finance, and other environments where exit is difficult or power is asymmetric.
The Nordic public-sector laboratory
Northern Europe offers a useful contemporary example because human-centred digitalization is being discussed at the level of public administration, not only product design.
The Nordic Council of Ministers' 2025 report A small step for digitalisation, a giant leap for humanity? examines human-centredness in Nordic public-sector digitalization through governance, public services, and innovation and reform. The work emerged from the Nordic DigiGov Lab collaboration involving public digital authorities in Sweden, Norway, and Finland.
One of its most useful contributions is conceptual humility.
“Human-centric” is not treated as self-explanatory.
The report asks what the term means in practice, how institutions organize around it, and how public services can preserve human concerns inside digital transformation.
For the Institute, that is exactly the level of seriousness required.
A Swedish citizen interacting with a public system is not simply a customer choosing among competitors. Rights, dignity, accessibility, accountability, and the ability to receive help become part of system quality.
Human-centredness becomes civic infrastructure.
Industry 5.0 and the European turn
European policy has also used human-centricity as part of the Industry 5.0 framing, alongside sustainability and resilience.
This moves the conversation beyond the Industry 4.0 emphasis on digitalization and automation toward the role of industry in society and the place of workers within technological transformation.
Policy language should not be mistaken for empirical validation. The existence of a European framework does not show that organizations using the language have changed their practices.
But it indicates a significant shift in what technology strategy is being asked to answer for.
The worker is no longer only a variable in production.
The citizen is no longer only an end user.
The person becomes part of the purpose against which the system is evaluated.
Human-centredness has critics
There is a serious criticism hidden inside the phrase itself.
Who counts as “the human”?
Design traditions have often universalized the needs of the populations easiest to recruit, the users most legible to institutions, or the cultural assumptions of the designers themselves. A generic human can erase disability, class, language, race, gender, age, geography, migration status, and other conditions through which technologies are actually encountered.
Human-centred approaches can also become anthropocentric, treating nonhuman life and ecological systems merely as resources for human preference.
Recent philosophical critiques of human-centred AI therefore ask whether “human” is too broad, too flattering, or too politically neutral a category to do the ethical work assigned to it.
That criticism should improve the field rather than close it.
Human-centred development needs to specify which humans, in which roles, with which powers, across which timescales, and with what obligations to other forms of life and future generations.
Participation without power is not yet human-centred
There is a quieter failure mode inside the language of human-centredness.
People can be consulted without being able to change anything.
A workshop can collect stories. A design team can map journeys, record pain points, and invite representatives into the room. A public agency can describe a process as participatory because citizens were asked what they thought. Then the decisive constraints remain untouched: procurement has already fixed the system, the success metric has already been chosen, the deployment date cannot move, and the people most affected have no meaningful route to refuse.
This is why participation has to be examined as a distribution of power, not only as a research method.
The relevant questions are practical. At what stage can affected people alter the objective? Can they contest the categories used to describe them? Can they see what happened to their objections? Can participation change a requirement, delay a deployment, or expose a value conflict that the original brief concealed? Are people invited because their knowledge is necessary, or because their presence will make the process easier to legitimize?
Human-centred technological development becomes more credible when the human is not reduced to a source of usability data. People are also rights holders, workers, citizens, patients, students, neighbors, experts in their own conditions, and sometimes people with good reasons not to want the system at all.
That makes refusal part of the design landscape. A genuinely human-centred process has to remain capable of discovering that the humane version of a technology is a smaller version, a slower version, a different version, or no deployment in a particular context.
How should it be evaluated?
No single human-centredness score is credible.
Evaluation should follow the lifecycle.
Problem definition: Were affected people involved before the solution was fixed? Were alternative nontechnical responses considered?
Design: Were relevant populations represented? Were accessibility, rights, capability, and power considered alongside preference?
Development: Can designers trace how human evidence changed requirements, objectives, or constraints?
Deployment: Are there meaningful routes for correction, appeal, human review, and refusal?
Use: What happens to workload, agency, trust, learning, safety, dignity, and task performance in the real environment?
Longer term: Which capabilities are strengthened, bypassed, or redistributed as the technology becomes normal?
Methods may include participatory workshops, contextual inquiry, usability studies, ethnography, experiments, accessibility testing, safety analysis, workload measures, complaint data, qualitative follow-up, audits, and longitudinal research.
Different questions require different evidence.
What it is not
Human-centred development is not making a machine sound warm.
It is not adding a user interview after procurement.
It is not maximizing satisfaction.
It is not the assumption that human judgment is always superior to machine performance.
It is not simply keeping a human in the loop.
It is not a promise to preserve every existing job or workflow.
It is not opposed to automation.
And it is not ethically complete by itself. Human rights, justice, ecological responsibility, safety, and democratic legitimacy can require questions that user preference does not answer.
What the evidence does not show
The evidence does not show that projects labeled human-centred consistently use the same methods.
It does not show that user involvement automatically produces safer, fairer, or more effective AI.
It does not show that trust is always a desirable outcome. A poorly calibrated system may deserve less trust.
It does not show that satisfaction predicts long-term capability or well-being.
It does not show that human oversight is meaningful merely because a person can approve or override an output.
It does not show that one group of users can stand in for everyone affected by a system.
The field needs stronger operationalization, clearer outcomes, and more longitudinal evidence about what human-centred processes actually change.
What this changes for Inner Technology
Inner Technology adds one question to human-centred development that is easy to miss in conventional usability work.
What is the technology training in the human over time?
A system can meet immediate needs while changing attention, confidence, memory, responsibility, dependence, social expectation, or the felt ability to act without it.
That does not make the system harmful.
It makes development formative.
Human-centred technological development and Inner Technology therefore meet at the point where design becomes part of human formation.
The woman at the public-service screen still cannot complete the form.
A genuinely human-centred system does not merely make the error message kinder.
It asks why her life was treated as an error condition in the first place.
Research references
Bevilacqua, R., et al. (2025). Framing the Human-Centered Artificial Intelligence Concepts and Methods: Scoping Review. JMIR Human Factors, 12, e67350.
DOI: https://doi.org/10.2196/67350
International Ergonomics Association. What Is Ergonomics (HFE)?
URL: https://iea.cc/about/what-is-ergonomics/
European Commission, Directorate-General for Research and Innovation. (2021). Industry 5.0: Towards a sustainable, human-centric and resilient European industry.
DOI: https://doi.org/10.2777/308407
Nordic Council of Ministers. (2025). A small step for digitalisation, a giant leap for humanity? Reflections on human-centredness in the public sector. TemaNord 2025:554.
DOI: https://doi.org/10.6027/temanord2025-554
Ryan, M. (2024). We're only human after all: a critique of human-centred AI. AI & Society.
DOI: https://doi.org/10.1007/s00146-024-01976-2