Artificial intelligence has arrived in the places where most people actually live their lives: phones that anticipate what you’ll search for next, software that drafts emails and summarizes meetings, tools that translate conversations in real time, and systems that quietly influence what you see, buy, or apply for. The speed of adoption is now outpacing the speed of public debate. That mismatch—between how quickly AI capabilities are spreading and how slowly societies agree on what they should be used for—is becoming the defining tension of the moment.
A recent call from the Financial Times frames the issue in a deceptively simple way: the technology’s development belongs to everyone, so share your vision with us. It’s not just a slogan. It’s an attempt to shift the center of gravity in the AI conversation. Instead of treating AI as a purely technical story—models get better, costs drop, performance rises—the question becomes: what do people want AI to do, and who gets to decide?
That question matters because AI is no longer confined to research labs or niche experiments. It is being embedded into hiring pipelines, customer service workflows, credit assessments, medical documentation, education platforms, and content recommendation engines. In each case, AI doesn’t merely “assist.” It changes incentives, reshapes processes, and can alter outcomes for individuals and communities. Even when the technology is marketed as neutral or efficient, its deployment reflects choices: what data is used, which objectives are optimized, what risks are tolerated, and which harms are considered acceptable.
The most important shift, then, is from capability to direction. The world is learning what AI can do at a breathtaking pace. The harder work is deciding what it should do—and what it should not do.
Why the “what people want” conversation is late, but urgent
There is a familiar pattern in technology rollouts: innovation moves first, governance follows later. But AI is different in two ways.
First, AI systems can scale decisions. A human manager might review a few dozen applications a day; an automated system can process thousands in seconds. That means the consequences of a flawed objective function or biased training data can propagate widely before anyone notices. When the debate arrives after deployment, it often becomes a debate about damage control rather than design.
Second, AI systems are increasingly opaque. Many modern models are not easily interpretable, and even when they are technically explainable, the explanations may not be meaningful to the people affected. This creates a governance gap: citizens and workers need clarity about how decisions are made, but the systems themselves may not provide it in a form that supports accountability.
So when the FT invites readers to share their vision, it’s implicitly acknowledging that the public conversation has been too narrow and too slow. The people who will live with AI’s outcomes—employees, patients, students, consumers, and communities—have not always been treated as core stakeholders. Engineers and executives have often led the discussion, while regulators and civil society groups have had to catch up.
The result is a growing sense that AI is being shaped by a small set of voices, even though its effects are broad. That perception isn’t just political; it’s practical. If governance is built without the lived experience of those impacted, policies may fail to address real-world harms or may overcorrect in ways that undermine beneficial uses.
A unique take: AI governance as “product design,” not just regulation
One reason AI debates can feel repetitive is that they often frame governance as a binary choice: either we regulate heavily or we let innovation run free. But the more useful lens is to treat governance as part of product design.
In other words, the question isn’t only whether AI is regulated; it’s how AI is engineered to behave in the contexts where it will be used. Safety, fairness, transparency, and access aren’t merely compliance checkboxes. They are design constraints that determine what kinds of systems get built, what kinds of data are collected, what kinds of user interfaces are provided, and what kinds of recourse exist when things go wrong.
Consider transparency. In many AI deployments, transparency is treated as a disclosure document: a policy page, a model card, a vague explanation. But for affected individuals, transparency needs to be actionable. If an AI system influences a loan decision, a job screening outcome, or a medical triage workflow, people need to understand not only that AI was involved, but also what factors mattered, what evidence was used, and how they can contest errors.
Now consider fairness. Fairness isn’t a single metric. Different definitions of fairness can lead to different trade-offs. A system optimized for one fairness criterion might still produce unacceptable disparities under another. Fairness also depends on context: what counts as discrimination in one domain may be handled differently in another. That means fairness requires domain knowledge and stakeholder input, not just statistical adjustments.
Access is another example. If AI tools are powerful but expensive, they can widen inequality. If they are available but poorly designed, they can create new forms of exclusion—especially for people with disabilities, limited digital literacy, or language barriers. Access is not simply “availability”; it’s usability, affordability, and inclusion.
When governance is treated as product design, the invitation to share visions becomes more concrete. People aren’t just asking for rules; they’re shaping requirements.
What “belonging to everyone” could mean in practice
The phrase “belongs to everyone” can sound abstract, but it can be translated into mechanisms.
1) Participatory governance
Instead of leaving AI oversight solely to technical experts or elected officials, participatory governance brings in workers, consumers, educators, clinicians, and community representatives. This doesn’t mean every decision is made by consensus. It means stakeholder input is systematically gathered and reflected in requirements, audits, and evaluation criteria.
2) Public-interest evaluation
AI systems should be evaluated not only on benchmarks but on outcomes that matter to society. That includes measuring error rates across groups, assessing downstream impacts, and testing for misuse. Public-interest evaluation also asks: does this system reduce harm, improve access, or increase exploitation?
3) Recourse and accountability
If AI affects someone’s life, there must be a path to challenge outcomes. Recourse can include human review, explanation, correction of data, and meaningful appeal processes. Without recourse, transparency becomes theater and fairness becomes a promise without enforcement.
4) Data rights and provenance
AI is only as good as the data it learns from, and data is never neutral. “Belonging to everyone” implies that people should have more control over how their data is used, and that organizations should be able to demonstrate provenance—where data came from, what consent was obtained, and what limitations apply.
5) Safety standards that match real risks
Safety is often discussed in terms of extreme scenarios—superintelligence, existential risk, catastrophic failure. Those concerns may be relevant, but everyday safety risks are already here: hallucinations in high-stakes contexts, prompt injection vulnerabilities, model drift, and automation bias where humans over-trust outputs. A responsible approach treats both long-term and near-term risks as part of the same safety culture.
The debate shouldn’t be about winners and losers—but it can’t ignore trade-offs
The FT’s framing suggests avoiding a simplistic “winners and losers” narrative. That’s wise, because AI progress will inevitably create disruptions. Some jobs will change, some tasks will be automated, and some industries will be transformed. But pretending there are no trade-offs can also be misleading.
A more constructive approach is to acknowledge trade-offs openly and design transitions responsibly. That means:
– Investing in worker reskilling and job redesign, not just layoffs.
– Supporting small businesses and public institutions so AI benefits aren’t monopolized by large firms.
– Ensuring that productivity gains translate into broader economic improvements rather than concentrating benefits.
– Building labor protections into AI deployment—clear boundaries on surveillance, performance monitoring, and algorithmic management.
This is where “vision” becomes practical. People don’t just want AI to be safe in theory; they want it to be fair in daily life. They want it to enhance opportunities rather than erode them.
The hidden power of AI: it shapes behavior even when it doesn’t “decide”
Another reason the conversation needs to broaden is that AI doesn’t only make explicit decisions. It also shapes behavior through recommendations, ranking, personalization, and feedback loops.
When an AI system ranks content, it influences attention. When it predicts what a user might want, it influences consumption. When it optimizes engagement, it can amplify sensationalism or polarizing material. These effects can be subtle, but they accumulate. Over time, AI-driven environments can change what people believe is normal, what they consider credible, and what they choose to do.
This is why governance must include not only decision-making systems but also “influence systems.” A society that focuses only on formal decisions—loans, hiring, admissions—may miss the broader cultural and psychological impacts of AI-mediated information flows.
The unique challenge of generative AI: trust, verification, and the cost of being wrong
Generative AI adds a new layer to the governance problem. When AI produces text, images, code, or audio, it can mimic human output convincingly. That creates a trust crisis: if content can be generated cheaply and at scale, verification becomes more expensive.
In the short term, this leads to practical problems: misinformation, fraud, impersonation, and confusion. In the long term, it raises deeper questions about how societies maintain shared facts. If everyone can generate plausible narratives, the burden shifts to institutions—newsrooms, courts, schools, and governments—to verify authenticity.
But verification isn’t free. It requires resources, standards, and infrastructure. That’s why “what people want” includes a desire for systems that support truth rather than undermine it. People want AI that helps them learn, communicate, and create—without turning every interaction into a credibility test.
This is also where transparency and provenance become crucial. Watermarking, content credentials, and cryptographic signatures are often discussed, but their effectiveness depends on adoption and interoperability. Governance