# Will we manage to cooperate in time?

## The real test may no longer be technological

There is a relatively comfortable idea about humanity that may be costing us time. The idea that, faced with a large enough threat, we always end up reaching an understanding. We may argue for years, deny evidence, defend particular interests and postpone difficult decisions, but when the danger becomes impossible to ignore, we react. History offers powerful examples that feed this confidence. We eradicated smallpox, we managed to build an effective international agreement to protect the ozone layer, we rebuilt societies devastated by war and, faced with Covid-19, we developed and produced vaccines at a speed that would have seemed improbable only a few years earlier.

There is, however, a problem with this optimistic reading of history. We have often cooperated after the danger had already become evident. And some of the transformations we are producing in the world today may not grant us that time.

We are facing an unusual combination of speed, scale and interdependence. Artificial intelligence is evolving faster than legislation, than education, than organisations and probably faster than our own capacity to understand all its consequences. The Stanford AI Index 2026 shows that more than 90% of the frontier models considered relevant in 2025 were produced by industry, and that model capability continues to advance rapidly. At the same time, safety evaluation mechanisms are not keeping the same pace: AI incidents documented in the AI Incident Database rose from 233 in 2024 to 362 in 2025. It is important not to turn this last figure into a false certainty. A rise in recorded incidents may mean more problems, better detection, or both. But the trend confirms at least one worrying asymmetry between technological capability and evaluation capability.

AI is not alone. Climate, biotechnology, geopolitical competition, cybersecurity, the transformation of work and the fragmentation of the information space all interact with one another. The World Economic Forum's Global Risks Report 2026 places geoeconomic confrontation at the top of the immediate concerns of the experts surveyed, followed by conflict between states, while disinformation and polarisation appear among the leading risks over the coming years. Again, precision matters: this report measures the perception of thousands of leaders and experts about risk; it does not demonstrate that those events are inevitable. But it is a good thermometer of the deterioration in expectations of international cooperation.

We have perhaps never possessed so much capacity to transform the world. The question is whether we have developed, at the same speed, the collective capacity to decide how we want to transform it.

This is where the old discussion about whether we are a cooperative or a competitive species stops being philosophical and becomes a question of institutional survival.

## We are not altruists. Nor are we merely selfish

For a long time a relatively suspicious view of human cooperation seemed intuitive to me. Humans cooperate, yes, but frequently because they need to. They cooperate to protect themselves, to achieve something they could not achieve alone, to belong to the group, to reduce danger, or because they know there will be consequences if they do not. Faced with catastrophe, they are capable of extraordinary generosity. Faced with the prospect of sacrificing part of their comfort today for the benefit of unknown people or of a future generation, that willingness seems to fall away quickly.

There is much truth in this intuition, but the research forces us to modify it.

Very young children already display helping behaviour before they master sophisticated systems of reward, reputation or punishment. In the work of Felix Warneken and Michael Tomasello, children of around 18 months spontaneously helped adults to reach their goals, and later studies found basic forms of helping in chimpanzees as well. Cooperation is not, therefore, exclusively a product of law or of a conscious calculation of interest. There are mechanisms of sociability and helping that appear very early in human development.

But this does not mean we are saints by nature. The research of Urs Fischbacher, Simon Gächter and Ernst Fehr on public goods found a far more interesting configuration. About half of the participants behaved as what the researchers called **conditional cooperators**: they contributed more when they believed that others were contributing too. About a third behaved as free riders, seeking to benefit without contributing proportionally.

This distinction is essential because it changes the question entirely. Perhaps the main difficulty of cooperation is not convincing each person that a common good exists. Perhaps it is convincing them that the others will not take advantage of their sacrifice.

Fehr and Gächter found another apparently strange behaviour: people willing to spend their own resources to punish those who did not cooperate, even when they drew no direct material benefit from that punishment. They called it altruistic punishment. The behaviour looks irrational when analysed at the individual level alone, but it becomes intelligible if punishing the opportunist works as a defence of the cooperative system.

We are, perhaps, a far more interesting species than the opposition between selfishness and altruism allows us to see. We are capable of generosity, but we watch the behaviour of others. We want to contribute, but we want reciprocity. We accept losing something for the collective, but we react when we conclude that someone has turned our cooperation into an opportunity for private gain.

Perceived fairness is not a moral ornament of cooperation. It is one of its infrastructures.

## From Aristotle to Hobbes: the problem is older than it seems

Aristotle saw the human being as a political animal, someone whose life only fully finds its expression within an organised community. This was not merely a way of saying that we enjoy the company of others. In his conception, the capacity for language allows us to discuss the just, the unjust and the common good, making political life part of the human condition itself.

Hobbes begins almost at the opposite extreme. If there is no authority capable of guaranteeing that others honour their commitments, even rational individuals with an interest in peace have reasons to distrust one another. Fear, competition and uncertainty make a permanent state of conflict possible. The Hobbesian solution is the common power capable of making commitments credible. For Hobbes, fear is not an accident in the construction of political order. It is one of its fundamental motivations.

Rousseau introduces an important correction to the excessively bleak view of human nature. For him there is an original disposition towards compassion, a *pitié* that leads us to care about the suffering of others, provided that this does not directly threaten our own survival. It is the evolution of society, with its permanent comparison, its status and its dependence on the opinion of others, that creates other forms of rivalry.

Two millennia between Aristotle and us, three centuries between Hobbes and Rousseau, and we remain caught in the same tension. We need others. We are able to care about them. And at the same time we fear that they will take advantage of us.

Contemporary research has not settled the dispute. It has made it more complex.

Michael Tomasello has argued that one of the particularly important human capacities is **shared intentionality**, the possibility of building a goal that is no longer merely "I want to do this" or "you want to do that", but "we are going to do this". Children between one and two years old display capacities for joint attention, cooperative communication and collaboration that help us understand how that "we" begins to emerge.

Perhaps the great human innovation was never the rational individual.

Perhaps it was the capacity to invent the "we".

## Catastrophe can create a "we", but we don't know which one

The hypothesis that a great threat unites us finds empirical support, but with an important caveat.

After the devastating earthquakes of February 2023 in Turkey, researchers interviewed survivors in the worst-affected regions. They found a relationship between the perception of shared suffering, identification with other survivors, and willingness to help them. In a particularly strong result, participants reported a willingness to help other Turkish survivors comparable to the one they showed towards their own families. Part of that identification extended also to Syrian refugees affected by the same catastrophe.

This confirms something deeply human. An extreme experience can temporarily erase earlier boundaries and create a common identity.

But we should not build our civilisational hope on that possibility.

A study conducted during the crisis between the United States and Iran in 2020 tested precisely the idea that a common enemy reduces internal divisions. Among certain highly polarised participants the opposite happened: exposure to the external enemy decreased their willingness to learn from members of the opposing party. The authors were cautious about generalising, but the result is enough to destroy the simple idea that a common threat automatically unites us.

Catastrophe does not necessarily produce solidarity.

It can produce solidarity **within a group** and hostility towards another.

It can produce a "we", but the decisive question becomes who is included in that pronoun.

A society with sufficient trust can interpret a crisis as a shared challenge. A society entering the same crisis deeply polarised may go looking for internal enemies, culprits or traitors. Danger alone does not decide which of these responses prevails. Prior conditions count.

And this is where the present moment becomes particularly uncomfortable.

## We are entering the storm with little trust

The OECD Trust Survey 2026, based on data collected in 2025 across 33 OECD countries and five accession candidates, found only 40% of respondents with high or moderately high trust in their national government, while 43% reported low or no trust.

But there is a figure that seems to me even more important. Although 68% consider that voting influences what the government does, only 31% feel that people like them actually have a voice in government decisions. Between those who feel they have a political voice and those who feel they do not, there is a 47 percentage point difference in trust in government.

We should not conclude from this that "people have stopped trusting democracy". The study itself shows higher levels of trust in the police, the courts, the civil service and local government, and the situation varies greatly between countries. The interesting finding is another one: **feeling that one takes part in the decision is deeply associated with trust in the institution that decides**.

This has enormous implications for the transitions ahead of us.

It is difficult to ask people to accept painful changes in employment, higher energy costs, new environmental rules, transformations driven by AI or restrictions on certain behaviours if they believe that they did not take part in the decision, that the costs are not evenly distributed, or that the most powerful will manage to escape the rules.

A society can ask for sacrifices.

What it can hardly do for very long is ask sacrifices of the many while they believe the exceptions belong to the few.

There are also signs of democratic deterioration that deserve attention, though here too precision matters. The V-Dem Democracy Report 2026, using the project's own methodology and data up to the end of 2025, concludes that the level of democracy experienced by the average world citizen has returned to roughly 1978 levels, and calculates that 41% of the world's population currently lives in countries it classifies as undergoing autocratisation. The report counts 44 countries on that trajectory. These values should be read as the results of a complex academic index, subject to methodological choices and scientific debate, not as direct measurements comparable to temperature or population. Even so, the scale and the direction of the trend deserve to be taken seriously.

The problem beginning to take shape is a dangerous one: **the challenges that demand collective cooperation are growing precisely when some of the social conditions necessary for that cooperation are under pressure.**

## Ostrom saw that the problem is not solved by authority alone

If Hobbes were alive, he might look at this instability and call for stronger institutions. And he would be partly right. But Elinor Ostrom showed that the choice between an omnipotent central state and individuals left to their own self-interest is far too impoverished.

For decades, Ostrom studied communities that shared scarce resources: forests, fisheries, irrigation systems, pastures and water reserves. She found many cases in which those communities managed to govern common resources sustainably without full privatisation and without depending exclusively on central control.

But it is important not to romanticise the finding.

Ostrom did not find communities where everyone was naturally good.

She found **good rules**.

The systems that worked tended to display clearly defined rights and responsibilities, a reasonable correspondence between benefits and obligations, monitoring, accessible conflict-resolution mechanisms, the possibility for users to take part in creating the rules, and graduated sanctions for those who repeatedly broke them.

This may contain one of the most important clues for the twenty-first century.

We do not need to wait for people to become morally superior. We need to build contexts in which cooperating makes sense, in which opportunism is detected, in which the rule appears legitimate, and in which whoever bears the cost can see that others are subject to the same game.

Environmental laws, taxes or technological regulation are not necessarily demonstrations of humanity's moral failure. They are often coordination technologies. They solve what we cannot solve individually because each person depends on the behaviour of everyone else.

Law, in this sense, does not replace cooperation.

It can make it possible at scales where personal trust is no longer enough.

## We have already proved that we can cooperate globally

If we want to build a pessimistic theory of humanity, we will have to explain some inconvenient facts.

Smallpox existed for thousands of years and killed hundreds of millions of people over the course of history. The WHO intensified the global eradication programme in 1967 through vaccination, surveillance and containment. The last known natural case occurred in Somalia in 1977 and the World Health Assembly declared the disease eradicated in 1980. It remains the only human infectious disease eradicated globally.

It is hard to imagine a more concrete example of planetary cooperation.

The Montreal Protocol also contradicts any easy fatalism. Adopted in 1987 to control substances that destroyed the ozone layer, it achieved practically universal participation and the elimination of around 99% of the ozone-depleting substances controlled by the agreement. If compliance holds, projections indicate that total ozone should return to approximately 1980 values by around 2040 between 60°N and 60°S, around 2045 in the Arctic and around 2066 in the Antarctic.

There was no sudden moral conversion of humanity.

There was sufficiently solid science, a relatively identifiable problem, the capacity to verify production and consumption, substitute technologies, international commitments and financial mechanisms that helped countries with less economic capacity.

The success was not born of our being good.

It was born of our having managed to **design cooperation**.

Climate change shows how much harder that task becomes when the problem is deeply embedded in the normal functioning of the economy. The UNEP Emissions Gap Report 2025 estimates that, if the announced national contributions are fully implemented, warming this century will land at approximately 2.3°C to 2.5°C. With the policies currently in place, the projection is around 2.8°C. There is progress relative to earlier assessments, but UNEP itself notes that part of the improvement stems from methodological changes and that the new pledges, on their own, changed the trajectory very little.

The correct reading is not "Paris solved the climate".

Nor is it "climate cooperation was useless".

It is far less comfortable: **we managed to change the trajectory, but not yet enough**.

That sentence may describe a significant part of human history.

## Covid showed two humanities at the same time

The pandemic offered another gigantic laboratory.

On one hand, we saw a scientific mobilisation unprecedented in scale and speed. On the other, when resources became scarce, global cooperation quickly found the borders of the nation-state.

In May 2022, the WHO and the COVAX partners showed the scale of the contrast. Only 16% of people in low-income countries had received at least one vaccine dose, against around 80% in high-income countries. COVAX had already shipped more than 1.3 billion doses to 87 low- and lower-middle-income countries, but had not managed to eliminate the inequality of access.

At the end of 2021, COVAX's own independent allocation group explicitly criticised wealthy countries for having bypassed the multilateral mechanism and struck direct contracts that secured them priority access.

It is hard to find a more revealing experiment.

We managed to create knowledge globally.

When the moment came to divide a scarce resource, we returned to the group.

This does not demonstrate that cooperation always fails. It demonstrates that cooperation and competition can exist within the same event, at different scales.

And this may be precisely the key to understanding the twenty-first century.

## The scale of consequences has outgrown the scale of our identities

For almost all of human history, most of the consequences of our decisions were relatively close by. Today that is no longer true.

A decision taken in a technology company in California can affect programmers in India, teachers in Portugal, elections in Europe and labour markets in Latin America. An emission produced in one country alters a shared climate system. A new viral variant crosses continents in days. A technology can be reproduced across thousands of organisations almost immediately.

Consequences have become global.

Political institutions remain largely national.

Human identities are frequently smaller still.

Family, community, company, religion, party, class, nation.

None of that is necessarily bad. We need local belonging. The problem appears when the size of the "we" that takes the decision is much smaller than the size of the "we" that bears the consequences.

There is research suggesting exactly this difficulty. Studies on large-scale cooperation show that the simultaneous existence of local identities and local benefits can reduce cooperation with wider groups. Human cooperation is extraordinary, but it tends frequently to be parochial: we cooperate intensely with "our own", while potentially competing with everyone else.

And this lets us see something we often forget: cooperation and competition are not opposites.

An army is an extraordinary cooperation machine built in order to compete.

So is a company.

So is a national football team.

Even a nation can develop enormous internal solidarity while competing aggressively with others.

The decisive political question is not, therefore, whether people cooperate.

It is **at the scale of which group they are able to treat the common interest as their own**.

## A civilisational race condition

This is where I find a hypothesis worth exploring.

In programming, a *race condition* occurs when several processes act on a concurrent system and the outcome depends on the sequence or the speed of operations. Each process may be executing its function correctly and the system may still produce an undesired result.

Some of the main contemporary crises may have this structure.

Imagine an AI company that recognises the risks of moving too fast. It may want to invest more time in safety. But if it believes that its competitor will not do the same, slowing down unilaterally may mean losing market, talent, capital and influence.

Now replace the company with a country.

No government wants to constrain its laboratories too heavily if it believes that a rival power will keep accelerating.

The result is disturbing: **each actor can take a decision that is rational from its own point of view, and together they can produce an outcome that none of them would choose on its own**.

The same problem arises in climate, in ocean fisheries, in the exploitation of certain resources, in armaments and in parts of biotechnology.

We do not need villains.

We only need misaligned incentives.

Robert Axelrod showed, through his work on the iterated Prisoner's Dilemma, that cooperation can emerge among self-interested actors when there is reciprocity and an expectation of future interaction. What he called the "shadow of the future" changes the present calculation: if I know I will need you again tomorrow, exploiting you today becomes more expensive. Later experimental work reinforced the importance of that expectation of repetition.

Now let us look at the architecture of incentives we have built.

Quarterly corporate results. Elections every few years. Financial markets that react in fractions of a second. Digital platforms that optimise the next minute of attention. Media cycles that barely survive a day.

At the same time, climate, education, demography, the sustainability of social security, infrastructure and technological governance demand horizons measured in decades.

We have ever longer problems and ever shorter incentives.

This may be one of the central contradictions of our time.

## Technology does not distribute power by itself

There is another reason not to treat this transformation as a purely technical problem.

Daron Acemoglu, Simon Johnson and James Robinson received the 2024 Prize in Economic Sciences for their studies on how institutions form and how they affect prosperity. An important part of the work of Acemoglu and Johnson insists on a simple idea that the history of technology frequently makes us forget: innovation and social prosperity are not synonyms. Technology can greatly increase productivity without the benefits being distributed equally widely.

It is not the machine that decides who keeps the additional productivity.

It is institutions, contracts, markets, power relations, taxes, education, property and politics.

This changes the question about AI.

Perhaps the most important question is not how many jobs will disappear. Perhaps it is who will own the machines, who will control the models, who will set the objectives, who will receive the productivity gains, and what capacity those who bear the costs of the transition will have to influence the rules.

If we manage to create enormous wealth with AI and simultaneously concentrate much of that wealth, we may have created a technological revolution and a political problem in the same operation.

Peter Turchin offers a provocative hypothesis here, though it should be treated with caution. His structural-demographic theory seeks to relate political instability to population pressures, competition among elites and state fragility. It is serious, published research, but there is no academic consensus that it works as a general law or as a precise instrument of historical prediction. The model itself has been criticised and later studies show limits to its generalisation.

Even so, there is a useful question hidden in Turchin's theory: what happens when a transformation produces extraordinary winners, numerous losers, and a growing perception that the system has stopped distributing opportunity legitimately?

It is a question the AI revolution cannot ignore.

## Without a shared reality there is not even a shared problem

Hannah Arendt saw an even deeper dimension of politics. A political community does not only need institutions. It needs a common world where different people can appear before one another, speak, act, and recognise that, despite differing perspectives, they are discussing the same reality.

Habermas carried this concern into the idea of the public sphere: a space where citizens can confront arguments about common matters and where public opinion can, at least ideally, be formed through critical discussion.

Neither of them knew TikTok, political microtargeting, deepfakes or generative models capable of producing practically unlimited volumes of text, audio, image and video tailored to different audiences.

The danger is not simply that false news exists. It always has.

The danger is the industrialisation of personalised reality.

If each community receives a different description of the event, trusts exclusively the sources that confirm its identity, and treats any contrary evidence as manipulation, we no longer have merely political disagreements.

We come to have ontological disagreements.

We are no longer arguing about what to do about the problem.

We are arguing about whether the problem exists.

And without a minimum quantity of shared reality there is no possible cooperation, because there is not even a common object to cooperate about.

We may therefore have to start thinking about truthful information the way we think about other essential infrastructures. Not in the authoritarian sense of creating an official truth, but in the sense of creating mechanisms of provenance, authentication, independent auditing, transparency, open science, sustainable journalism and the capacity to verify where a given claim, image or video came from.

Industrial society needed sanitation to protect its water.

Informational society may need something similar to protect trust.

## Can we use technology to produce more cooperation?

There is a contemporary experiment that is particularly interesting because it inverts the dominant architecture of social networks.

In Taiwan, processes such as vTaiwan used the Pol.is software to gather the opinions of thousands of citizens and to group people according to their patterns of agreement. Instead of favouring responses that provoked outrage, the system made visible those statements capable of attracting support across different groups.

One of the best-known cases was the discussion on regulating Uber from 2015 onwards. The process helped identify zones of consensus among passengers, taxi drivers, Uber users, companies and authorities, and influenced later regulatory changes.

But here too it is worth resisting the temptation to tell only the pretty story. Later research on the same process showed important limitations: not all the minority positions identified by the algorithm reached the political stage with equal weight, and the conflict between Uber and part of the taxi sector did not disappear.

This makes the example more interesting, not less.

Technology did not solve democracy.

It showed that **technological design changes the kind of behaviour that emerges**.

For two decades we built systems whose main economic incentive was to maximise attention and engagement. Then we were surprised when outrageous, polarising and emotionally intense content prospered inside them.

Perhaps we should try the inverse question.

What if we used the same computational capacity to discover compromises, expose false consensus, identify shared interests, structure deliberation and show people what opposing groups actually think, instead of the caricature we make of them?

For years we asked how to make machines more intelligent.

Perhaps we should start asking how to use them to make groups of human beings collectively more intelligent.

## Four hypotheses for the world we are building

There are four hypotheses that seem to me to follow from this research. They are not demonstrated facts. They are interpretations and, precisely for that reason, they should be open to challenge.

The first is what I would call **cooperative lag**. There is an interval between the moment we rationally perceive a risk and the moment we are emotionally and politically willing to pay the price to avoid it. That interval is tolerable when change is slow and reversible. It becomes dangerous when the technology or the natural system moves faster than the political response.

The second is a **scale mismatch**. The consequences of some of our decisions have become planetary, but much of our machinery of identity, responsibility and governance remains national or local. This does not mean we have to destroy those identities. On the contrary. We may need to learn to combine strong local belonging with mechanisms of cooperation at much larger scales.

The third is that **shared reality is turning into a public good**. A society cannot function on freedom of expression alone. It also needs conditions that make it possible to distinguish testimony from fabrication, evidence from manipulation, and legitimate disagreement from deliberate falsification.

The fourth is that we may be living through a **civilisational race condition**. Countries, companies and individuals respond rationally to the incentives they encounter, but the sum of those decisions can produce a trajectory that almost nobody would choose as a collective outcome.

None of these hypotheses implies that we are doomed.

All of them imply that waiting for cooperation to arrive spontaneously is an insufficient strategy.

## What might work

Ostrom points towards polycentric institutions, not towards the fantasy of an omnipotent world government. Complex problems may need several centres of decision, connected to one another: cities, states, companies, scientific communities, regulators, international organisations and civil society. The challenge is to create interoperability between them without eliminating autonomy.

Axelrod reminds us that we have to lengthen the shadow of the future again. If governments, companies and markets are systematically rewarded for the decision that produces immediate benefit and transfers the cost to whoever comes next, we cannot be surprised at our inability to solve long-term problems. We need institutions capable of representing future interests inside present decisions.

Ostrom and the research on cooperation further suggest that any transition depending on sacrifice requires visible reciprocity. If the costs of decarbonisation, of the transformation of work or of technological adaptation are perceived as deeply asymmetric, the resistance will not be mere selfishness. It will also be a reaction to the design of the system.

The OECD provides another clue. Feeling that one has a voice is strongly associated with trust. We may have to increase the bandwidth of democracy, not by replacing elections with permanent plebiscites, but by creating far more continuous forms of participation, deliberation and feedback.

And finally there is technology itself. We can build AI to replace human decision or to improve human decision. To maximise persuasion or to reveal manipulation. To find the message that divides most or to find the statement that different groups can accept. To concentrate knowledge or to make it accessible.

There is no technological law determining which of these trajectories will win.

There are choices.

And power.

## The question that remains

The evidence forces me to abandon two comfortable positions.

The first is the romantic view that, deep down, when faced with great problems, humanity always ends up doing the right thing. History does not permit that confidence.

The second is the cynical view that we cooperate only when we are afraid, when we are compelled, or when we expect to gain something. That does not survive the evidence either.

We are capable of spontaneous altruism. We are capable of building extraordinarily sophisticated institutions of cooperation between strangers. We managed to eradicate a disease from the planet and to collectively change technologies to allow the ozone layer to recover.

But we are also conditional cooperators, tribal, and deeply attentive to the fairness of the game.

Perhaps the future does not depend on our managing to transform human nature.

Perhaps it depends on understanding it well enough to build institutions suited to it.

Because the problem of the coming years will not only be that we have more intelligent machines, a faster economy or more powerful technologies.

It will be that we have created systems with consequences larger than any individual actor can control, at a moment when trust is scarce, shared reality is under pressure, and incentives continue frequently to reward whoever arrives first, not whoever chooses best.

History shows that we manage to cooperate when we realise we are in the same boat.

The real challenge now is considerably harder.

**Will we realise we are in the same boat before it starts taking on water?**

## A note on evidence and interpretation

The quantitative data in this essay were verified in primary or academic sources, including the OECD, Stanford HAI, UNEP, the WHO, V-Dem, Nobel Prize Outreach and scientific articles published in *Nature*, *Scientific Reports*, *Science*, *Economics Letters* and other academic publications.

Not all claims have the same epistemological status. The OECD figures are survey results. The Global Risks Report measures risk perception. V-Dem produces indices built from an academic methodology that admits debate. The AI incidents are documented incidents, not a complete count of every incident that exists. The experimental studies on cooperation show behaviour under specific conditions and do not, on their own, authorise generalisations about all of humanity.

The concepts of **cooperative lag**, **scale mismatch**, **shared reality as a public good** and **civilisational race condition**, as used here, are interpretative hypotheses of the author. They are proposals for thinking about the problem, not established scientific results.

This distinction does not weaken the essay.

It is exactly the opposite.

A hypothesis is only intellectually interesting when we know where the facts end and the attempt to understand them begins.