The future is not predicted: it is rehearsed together. Article 700 proposes thinking big without making anyone feel small, an ambition that does not measure progress by the size of a machine or the speed of an answer, but by society's capacity to expand knowledge, distribute opportunity, and care for a shared life. From that perspective, a deeply human future is neither a sentimental image nor a pause in innovation. It is a public discipline: purposes come before tools, listening precedes deployment, and the ability to correct remains available whenever a promise and lived experience diverge.

For too long, buying a solution before agreeing on the problem has been called innovation. That order changes the conversation. An organisation begins by asking what a product permits, a team adapts its needs to a catalogue, and affected people appear at the end, when almost everything has been decided. The alternative reverses the sequence. A specific need is described first; capabilities, rights, and limits are identified next; only then does the institution compare whether a technology, an organisational reform, or a combination of both responds best. The result may look less spectacular, yet it is usually more useful, understandable, and sustainable.

Thinking at scale does not require uniformity either. The same infrastructure can provide broad benefits and impose very different burdens depending on age, language, disability, location, available time, or the ability to appeal. An average is informative, but it does not represent every life. A human proposal therefore has to examine the distribution of benefits and costs, not only the aggregate balance. It must also recognise that efficiency loses legitimacy when it saves an institution minutes by adding hours, anxiety, or travel to the people who depend on it.

From fascination to a shared purpose#

Responsible imagination begins with a sober question: whose daily life should improve, and in what way? The answer cannot be reduced to a feature list. It has to describe an observable situation, such as reducing waits without removing in-person assistance, supporting learning without turning every doubt into commercial data, or coordinating energy without penalising vulnerable households. That description separates a public goal from a technical preference. It also allows people without specialist training to contribute legitimate knowledge, because they understand the rhythms, obstacles, and consequences of their context.

Responsible imagination begins with a sober question: whose daily life should improve, and in what way?

A well-framed purpose includes a boundary. It states what will not be done even if it is technically possible, which data are unnecessary, which decision will remain with a person, and which harm would justify stopping a trial. The boundary prevents a tool designed to advise from becoming a mechanism that scores, monitors, or excludes. It also turns broad values into verifiable commitments. Dignity ceases to be a decorative word when a person can understand a decision, provide context, choose an alternative route, and obtain redress if the process fails.

A collective rehearsal requires diverse roles, not a symbolic meeting. A developer understands technical capacities and limits; a service professional understands exceptions and workloads; a recipient understands invisible barriers; a public-interest representative must watch rights, resources, and long-term effects. No perspective replaces the others. When one dominates, blind spots appear: accuracy is confused with justice, satisfaction among active users conceals abandonment, and formal compliance is interpreted as social acceptance.

Five criteria for a deeply human future#

The first criterion is situated usefulness. A proposal must solve a recognised problem under real conditions, including difficult days, unstable connections, and cases that do not fit. The second is agency. Affected people need enough information and an early opportunity to alter scope, conditions, or priorities. The third is proportionality: the greater the consequence for education, employment, health, mobility, or access to rights, the stronger the required evidence, oversight, and route of appeal.

The fourth criterion is reversibility. A responsible test has a review date, exit criteria, usable copies of information, and a procedure for returning to a safe alternative. Claiming that an off switch exists is insufficient if nobody has authority to use it or if the institution has lost the capabilities needed to continue without its supplier. The fifth is accountability. An identifiable person or body must be able to explain what happened, inspect records, correct a decision, and repair its consequences.

These criteria work best with paired indicators. Quantitative measures reveal time, errors, differences between groups, resource use, complaints, and abandonment. Qualitative evidence explains why the pattern occurred: a confusing instruction, a missing language, a humiliating alternative, automation that fragmented work, or an interface that made a recommendation look compulsory. Combining both protects against two familiar deceptions: an anecdote presented as a universal rule and an average presented as a complete experience.

Case 1: a school tests support without automating judgement#

In an illustrative case, a secondary school wants to support reading and writing with a generative tool. Its leadership avoids announcing a total transformation. It defines the problem first: some students need more opportunities to practise, while teachers lack time to provide immediate comments on every draft. The agreed goal is not to produce text faster. It is to increase the frequency of revision without replacing educational dialogue or turning personal expression into an automated score.

The trial begins with two small groups, materials containing no sensitive data, and tasks in which students compare suggestions, detect errors, and justify their choices. Families receive a plain-language explanation of the purpose, information used, limits, and enquiry channel. An equivalent activity without the tool remains available, with no loss of marks or stigma. Teachers retain assessment, record when a suggestion helps or weakens a text, and devote a session to sources, authorship, bias, and trust. Learning therefore covers use as well as the judgement required not to use.

After six weeks, the school does not ask only how many texts were completed. It examines whether revisions improved, whether dependence on formulaic language increased, whether any group faced greater barriers, and how much additional work fell on teachers. Several students explain that fluent answers look definitive even when their reasoning is weak. The school then changes its working method: every suggestion must be checked against a source or supported by the student's own reason. The correction is not treated as a failed pilot, but as evidence that rehearsal is serving its purpose.

Case 2: a neighbourhood modernises a service without closing the door#

In a second illustrative case, a municipal library network wants to improve bookings, notices, and guidance through a digital assistant. The initial diagnosis shows that the principal frustration does not come from a lack of machine intelligence. It arises from confusing opening hours, catalogues that use different terms, and procedures that force people to repeat information. Before procuring a complex solution, the team aligns vocabulary, simplifies forms, and publishes an accessible calendar. Technology is reserved for guidance and never used to decide who deserves a service.

The assistant is tested in three branches serving different populations. Every screen clearly identifies the interaction as automated and provides visible access to a person. Enquiries are not repurposed for advertising, data retention is limited, and language choices are made with local associations. Staff record unanswered questions, failed referrals, and occasions on which the system expresses certainty it does not possess. Older residents, screen-reader users, and people with limited digital literacy take part in paid testing, rather than being invited only to a final presentation.

Measurement shows that simple enquiries are resolved sooner, but also that access to a person becomes less visible on small phones. The municipality corrects the design, retains desks and telephone assistance, and publishes results by branch and channel. Expansion depends on repetitions declining without exclusion increasing, answers being traceable to official information, and staff having time to review mistakes. The case shows that keeping a door open does not diminish innovation. It prevents convenience for a majority from turning into abandonment for a minority.

Case 3: a city allocates resources with evidence and appeal#

In the third illustrative case, a city considers a model to prioritise building energy-efficiency inspections. The team refuses to use a prediction as a verdict. It treats the output as one signal among several, alongside technical data, district knowledge, and requests from residents. Before the pilot, the team examines whether historical records reflect unequal inspection patterns; a neighbourhood with few files can appear low risk precisely because it has received less attention. The city publishes variables, purpose, responsible body, and limits in understandable language.

For one season, the trial compares the usual selection process with assisted selection. No household loses support because of a score, and a person reviews every priority before any communication is sent. Residents can correct data and request review by telephone, in person, or online. A committee of technicians, community associations, rights specialists, and frontline staff examines geographic differences, false alerts, time, and environmental benefits. It also calculates the system's own energy and material costs, because a human future does not silently transfer its burdens to the environment.

Results show more useful detections but a higher error rate among recently refurbished properties. Instead of hiding the limitation, the city corrects the outdated source, contacts those who received inaccurate notices, and repeats the evaluation. Only then does it approve gradual expansion, with an annual audit and expiry clause. Authority to stop remains with the administration, not with the technology contract. Progress lies in that ability to acknowledge, repair, and learn, rather than in pretending that a model will always be right.

How to organise a rehearsal that can be corrected#

A serious rehearsal can be summarised in a public record. It identifies the problem, affected group, expected benefit, risks, necessary information, alternatives, responsible person, and review date. It adds a baseline for comparison and stopping thresholds. If complaints rise, an unjustified difference between groups appears, or the alternative service deteriorates, the trial pauses. Those thresholds turn caution into a concrete operation and prevent money already spent from becoming a reason to continue.

Participation must occur before a product is fixed. A late consultation can only decorate decisions. Early participation can redefine a problem, rule out data, create a non-digital option, or reveal that the main obstacle is organisational. Legitimate participation requires time, accessibility, feedback, and traceability: the institution explains which proposals it adopted, which it did not, and why. Where appropriate, it also compensates knowledge contributed by communities that are repeatedly consulted without receiving resources or power.

Documentation must not become a mountain of reports that nobody uses. A concise log of changes, incidents, decisions, and responsible people should be maintained, accompanied by specific materials for those operating the service and those who may be affected. A public explanation answers at least six questions: what should improve, what information is involved, what consequence can arise, how long the measure will last, who is accountable, and how an appeal works. If a team cannot answer clearly, the decision is not mature.

Independent review adds value when it has access, competence, and the ability to influence outcomes. Depending on risk, it may come from an autonomous internal unit, a public authority, a mixed committee, or an external auditor. What matters is that review does more than certify paperwork: it observes practice, speaks with affected people, tests exceptions, and checks remedies. Human oversight does not mean placing a signature at the end of a chain. It means providing time, evidence, and real authority to change the result.

Ambition on a human scale#

Thinking big means extending the time horizon. An immediate improvement may create dependency, reduce professional capacity, or establish infrastructure that is difficult to leave. Evaluation therefore covers maintenance, training, working conditions, accessibility, security, energy, and contractual exit. It also asks who learns from the system. If only a supplier learns while an institution loses knowledge, innovation weakens future capacity. If learning is documented and shared, every pilot strengthens collective judgement.

A human scale does not mean limited ambition. It means that no scale erases the particular person. A national policy can retain local channels; a common platform can allow justified adaptations; an advanced system can offer a simple explanation and a dignified alternative. Technological wonder then finds a direction: it helps society imagine more possibilities without presenting every possibility as an obligation. Curiosity is paired with responsibility, and speed leaves room for care.

Article 700 closes a series, but it does not close a conversation. Its central invitation remains open: a society can treat the future as a finished product that arrives from elsewhere, or as a common practice that is examined, discussed, and corrected. The second option requires more institutional work and less faith in automatic solutions. It also offers something more valuable than a prediction: the democratic capacity to decide what deserves to be built, what must be protected, and what should be left behind.

A deeply human future is ultimately recognised through modest tests with large consequences: a person understands what changes; a community participates before the decision; a professional can halt a process; an institution acknowledges an error; an alternative stays open; a remedy arrives in time. There, thinking big makes nobody small. It turns progress into a responsible relationship among imagination, evidence, rights, and care.

Institutional sources#

  • UNESCO, Recommendation on the Ethics of Artificial Intelligence: https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
  • OECD, OECD AI Principles: https://oecd.ai/en/ai-principles
  • European Commission, Artificial Intelligence Act: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
  • Council of Europe, Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law: https://www.coe.int/en/web/artificial-intelligence/the-framework-convention-on-artificial-intelligence