The Future Will Not Be Human by Default

We have spent decades asking whether machines can become more human, but far less time asking whether human progress will remain human as machines become more capable. That question is no longer philosophical in the abstract. It is entering offices, classrooms, hospitals, businesses and homes.

A 2026 global study by Boston Consulting Group found that 42% of regular frontline AI users report saving at least eight hours each week. An entire working day has appeared. Yet 66% receive limited or no guidance about what to do with that time, and more than half are not reinvesting it in more strategic work. The technology worked, the employee adopted it and the capacity appeared. What remained unclear was what to do with it.

That contradiction points to something larger than an adoption problem. We are becoming remarkably good at creating new capacity before deciding what we want to do with it. A manager can compress four hours of work into forty minutes, a teacher can generate lesson plans before breakfast, and a founder can automate most routine customer questions. Those gains are real, but none of them tells us what the extra time or capacity is for. More work? Better work? Lower prices? Fewer employees? Deeper learning, more care, more rest?

AI can work toward goals. It can increasingly recommend, coordinate and act. But whether those goals are worth pursuing, who should benefit, which risks are acceptable and what should not be sacrificed cannot simply be handed over to the machinery. Those are choices about direction, and institutions already have one. They have targets, budgets, incentives, hierarchies and definitions of success. AI does not arrive in an empty room. It enters an organization that already rewards certain things, ignores others and has its own ideas about what counts as value.

That matters because AI can strengthen those existing choices even as it changes how the organization works. If a company rewards only speed, AI can help it move faster. If it treats people primarily as costs, AI can make that logic more efficient. If it values learning, judgment and service, the same technology can strengthen those things instead. The tool matters, but so does the logic of the organization using it.

This is where the familiar phrase “human in the loop” starts to feel insufficient. A physician can challenge a recommendation that does not fit the patient in front of them. A public servant can review an exceptional case. A teacher can reject an answer that is polished but empty. Thoughtful human oversight matters, but participation can also become ceremonial. A person may still click the final approval button while lacking the time, information or authority to question what produced the recommendation. They remain present, but the choices that really matter were made elsewhere.

Presence, in other words, is not the same as power. A future does not remain human merely because people participate in it. It remains meaningfully human when people still have real influence over the goals, the limits, the consequences and who benefits. There is a useful word for that ability to understand, choose and act: agency. Can people understand what is happening? Can they question it? Can they choose among real alternatives, say no, challenge a decision and change direction? Do they still have enough authority to be responsible for the outcome?

None of this assumes that human judgment is automatically wise. People are biased, inconsistent and sometimes unjust, and slower systems are not necessarily better systems. The objective is not to preserve human decision-making simply because it is human. The harder question is whether new machine capability helps people make better decisions and carry real responsibility, or quietly takes both away.

Trust matters here for a practical reason. Important knowledge inside an organization does not live only in databases. It also lives in people: the technician who hears that a machine sounds different, the nurse who knows why one patient needs continuity, the employee who has learned which shortcut is safe and which one is dangerous. When people believe that sharing what they know will be used against them, they share less. When they feel unsafe questioning a system, errors remain silent. When instant answers replace difficult conversations, organizations can lose the kind of hard-earned knowledge that only comes from doing the work.

The evidence directly connecting trust to the technical quality of AI is still emerging, and we should not pretend otherwise. But we already know something important about organizations: trust and psychological safety affect whether people speak up, report errors, share what they know and challenge decisions. That means fear can make an organization less intelligent even as its tools become more capable.

This brings us back to the eight hours identified by BCG. They are not merely “time saved”; they are new space inside the working week, and what fills that space is a choice. Some organizations will use it for more meetings, more reporting and higher targets. Some will reduce headcount. Others will use it for learning, mentoring, experimentation, better decisions or more human attention. Who owns that recovered capacity? Who decides what it is for? What counts as value after the task becomes faster? These are not productivity questions alone. They are questions about power, incentives and who gets to decide.

For me, this question began long before artificial intelligence. I grew up in Costa Rica, a country whose modern civic story was shaped by a consequential choice about the kind of capacity it wanted to build. After the 1948 civil conflict, Costa Rica abolished its standing army, and the 1949 Constitution made that direction permanent. The decision did not invent the country's social investment, and it certainly did not solve the contradictions of development. Costa Rica still faces inequality, insecurity, educational gaps and institutional strain. But the choice mattered because it expressed something larger than a budget decision. It said something about the kind of country we wanted to build.

Abolishing the army was not only about what Costa Rica would stop funding. It reinforced a national direction centered on education, health, social protection and civilian institutions. I do not offer that history as a blueprint for how to govern AI; the contexts are too different. I offer it because it demonstrates something easy to forget when technology is moving fast: capability is not destiny. Societies and institutions make choices about what they build, what they measure, what they reward and what kind of life those choices are meant to support.

That idea later became more explicit in my work around social progress. The Social Progress Index was built on a deceptively important distinction: economic growth is not the same as social progress. A country can become wealthier without meeting basic needs, strengthening wellbeing or expanding opportunity. The Index therefore asks a different question: what is actually changing in people's lives? That distinction stayed with me because it exposes a recurring mistake. We often treat the thing we can measure most easily as if it were the thing we ultimately value.

AI risks repeating that mistake at extraordinary speed. If our measures of intelligence emphasize only autonomy, productivity, speed and cost reduction, those are the futures organizations and capital will learn to recognize. Dignity, the ability to choose, belonging, trust and meaning can disappear from view not because they lack value, but because the system has no column for them. Institutions are values repeated long enough to become structures; metrics are values given columns; budgets are values given power. What cannot be seen is difficult to govern, what cannot be explained is difficult to fund, and what cannot be measured is easy to sacrifice.

That does not make efficiency the enemy of humanity. Efficiency can remove drudgery, lower costs, expand access and return time to people. What matters is what happens next. Does automation free a professional to exercise better judgment, or merely increase the number of cases they are expected to process? Does AI give a teacher more time to teach, or simply raise the expected volume of administrative output? Does the time we recover become freedom, learning and better service, or just another input absorbed by the system?

These choices are not made once, because direction is not a single instruction we give a machine. It is something people and institutions have to keep choosing. That requires leaders willing to define value beyond efficiency, workers able to question without fear, measures that reveal what is happening to people, customers and citizens who can challenge decisions that affect them, and capital willing to recognize forms of value that do not arrive already translated into growth curves. It also requires us to decide what we refuse to sacrifice.

The central human question of the AI era, then, is not whether machines become more like us. It is whether increasing machine capability gives us more or less say over what progress is for. We have spent decades teaching machines to do more of what humans can do. The next stage of progress will depend on whether we become equally serious about the things we still have to decide: what is worth wanting, what kind of life our systems should support, who gets to choose, and what the recovered time of our lives is for. What kind of capability are we trying to build, and what kind of humanity should it make possible?