Human Failure

Market failure, government failure, democratic failure, murderous wars – in the end, they are all forms of human failure. At first, that sounds rather hopeless. Because if human beings are ultimately the problem, wouldn’t we have to change human beings in order to make the world a better place?

That conclusion seems obvious. And perhaps it is precisely the conclusion we should be most wary of.

Few ideas have more blood on their hands than the belief that human imperfection can be overcome through ideological transformation. If only everyone could be made to think correctly, embrace the right values, overcome their selfish interests and subordinate themselves to a higher social purpose – then surely, at some point, a better society would emerge.

History, unfortunately, suggests a different conclusion. The more obviously such attempts failed to achieve their goals, the less likely people were to question the idea itself. Instead, someone had to be blamed for standing in the way of its realization. Improving people became re-educating them, re-education became coercion, and coercion eventually became violence. From the guillotines of the Jacobins to the totalitarian social experiments of the twentieth century, there is a horrifying trail of blood left by attempts to create better human beings for a better society.

Perhaps, then, we should try exactly the opposite approach. What if we did not have to improve human beings at all? What if we simply accepted human imperfection and made that acceptance the starting point for designing society?

It sounds more modest. In fact, it could be quite revolutionary.

Our world is full of paradoxical and counterintuitive relationships. The old saying that “well meant” is not the same as “well done” merely hints at this. Sometimes the opposite of what was intended does not arise by accident. Sometimes consistently pursuing what appears to be a perfectly reasonable goal systematically produces the opposite result.

And this is where the first form of what I call human failure begins.

Rational human failure

Let us begin by imagining someone who behaves perfectly rationally. He looks after himself and his family, tries to secure his income, build some wealth and avoid unnecessary risks. If he is an entrepreneur, he tries to keep his business alive and competitive. None of this is particularly reprehensible. On the contrary, we would probably say that this person is behaving quite rationally.

Now let many such rational people interact with one another.

And suddenly, something strange can happen.

What makes sense for one individual does not necessarily make sense for everyone. Individual economic rationality and the rationality of the economy as a whole can diverge. Economics encounters this rationality trap again and again.

Sometimes it works out wonderfully. Adam Smith provided perhaps the most famous description of this phenomenon with his “invisible hand.” The baker does not bake bread purely out of love for humanity. An entrepreneur does not necessarily develop a better product because he wants to improve the world. Both are initially pursuing their own interests. Yet when markets function and many suppliers compete with one another, something highly beneficial to society can emerge from that self-interest. Products improve, processes become more efficient and prices fall.

Wolfgang Stützel later classified such relationships as classical competition paradoxes. And the fascinating thing is that we did not have to improve human beings at all. We simply took self-interest as it is, and under suitable conditions competition transformed it into something socially productive.

Unfortunately, the same mechanism can work in the opposite direction.

Marx described the destructive side of competition using the workers of his time as an example. When there are too few jobs, it may be perfectly rational for an unemployed person to offer his labour for less. Better to work for less money than not to work at all. But if the next person does the same, and eventually everyone does, wages fall across the economy. Demand for the products made by businesses then falls as well. In the end, even those who initially appeared to benefit from lower wages may lose.

Again, no one necessarily behaved irrationally. Precisely because everyone acted sensibly from their own point of view, the outcome for everyone together can be harmful.

This is the Marxian competition paradox. And it is precisely this process that I describe as rational human failure. It is not necessarily the individual who has failed. It is the outcome of human interaction that has failed when measured against the interests of society as a whole.

Seen this way, the seemingly endless argument between faith in markets and criticism of markets begins to look rather strange. The invisible hand and market failure are not mutually exclusive theories. They are two sides of the same reality. Competition can turn self-interest into social benefit – and the very same self-interest can produce social harm.

The interesting question, therefore, is not whether markets or governments are “better.” It is this: under what conditions does human self-interest produce a good social outcome, and what do we do when it does not?

At that point, we need institutions. And sooner or later, we arrive at government.

The problem, of course, is that governments are made up of human beings too. We have therefore not eliminated human failure. We have merely moved it to another level. Market failure can be followed by government failure. A public authority can try to solve a real problem and create a new one in the process. Politicians can introduce well-intentioned rules whose unintended consequences eventually cause more harm than good. And suddenly, people are deciding what other people should do and how other people’s money should be spent.

Perhaps there is something to be learned from this as well. Because of its many opportunities to fail, government should remain as lean as reasonably possible and use its particular strengths to correct market failures without simply replacing them with government failures.

One of those strengths is taxation.

Government is quite good at collecting taxes and establishing binding rules. Things become much more difficult when it tries to decide which car millions of different people should drive, which heating system they should install or which technology will prove to be the right one ten years from now.

A steadily rising and long-term predictable carbon tax, with all revenues returned to citizens on an equal per-capita basis, would be a good example of a different approach. The government would not have to educate better citizens or select the better heating system. It would simply change the economic framework. After that, everyone would once again be free to make decisions according to their own interests.

And suddenly, the very self-interest we had regarded as the problem becomes part of the solution.

Rational human failure may therefore actually be a solvable problem. We do not have to change people. We merely have to learn how to design the rules of society so that individually rational behaviour produces socially rational outcomes as often as possible.

That is actually a rather hopeful conclusion.

Unfortunately, it immediately raises another question.

If this is possible in principle, why have we not built such wonderfully designed social systems already?

And that brings us to the second – and probably much more treacherous – form of human failure.

Irrational human failure

Human beings are not merely self-interested. They are also emotional.

That is hardly a new discovery. Freud struggled with this phenomenon, and later psychological models such as Transactional Analysis attempted to describe different modes of human thought and behaviour. For our purposes, a simple observation is enough: human beings do not always think in the same way.

We are capable of remarkably rational thought. Give an engineer a technical problem, a mathematician an equation or a programmer a bug in some code, and with remarkable reliability an analytical process begins.

Now give the same people a problem involving justice, wealth, poverty, migration, power, war or peace – and something changes.

We no longer merely think. We feel and judge. Experience, fear, desire, moral beliefs and group identities enter the process. Transactional Analysis offers one useful model for understanding this. Simplifying considerably, we might speak of rational thinking, wishful thinking and judgemental thinking, between which we switch largely without noticing.

Again, there is nothing inherently wrong with this. It is simply human.

Unfortunately, almost every major question concerning the design of society is about people.

So precisely where we most need to design our social institutions intelligently, we find ourselves dealing with subjects particularly likely to activate our wishful and judgemental thinking.

Perhaps this explains one of the most astonishing contradictions of our civilization. We manufacture computer chips with structures only a few nanometres wide, send spacecraft billions of kilometres through the solar system and build machines whose complexity can no longer be fully understood by any single human being. Yet when we design our social systems, we repeatedly stumble over problems that in some cases have been known for centuries.

And occasionally we even kill one another over them.

Perhaps what we lack is not primarily knowledge. Perhaps what we lack is an institution that helps us deal with the way our own minds work.

And this is where, in recent years, something genuinely new has appeared.

Artificial intelligence.

Of course, there have always been people capable of thinking unusually clearly about counterintuitive social relationships.

For simplicity, let us call them nerds.

The nerds, however, had a problem. They could think things that hardly anyone wanted to hear – and quite often they were not particularly good at explaining them either. Anyone who has understood a paradoxical relationship eventually finds themselves standing in front of another person whose entire intuition insists that the relationship simply cannot be true.

The nerd talks. The other person shakes their head. End of social innovation.

And suddenly, between the two, there is a machine developing a remarkable ability:

It can translate.

The nerd can think a relationship through and present it to the AI. The AI can examine it, search for counterarguments, look at it from different perspectives and then try to explain it in a way that is accessible even to someone whose intuition initially points in exactly the opposite direction.

Consider an example.

Almost everyone thinks it is good if government has as little debt as possible. At the same time, most people also think it is good if they themselves possess as much safe financial wealth as possible.

Both desires sound perfectly reasonable.

But financial claims are not real assets that simply exist somewhere on their own. Every financial claim is matched by someone else’s financial liability. So whenever we talk about the indebtedness of one sector, we are necessarily also talking about the financial claims held by others.

In terms of accounting balances, this is not particularly complicated. Intuitively, however, it feels completely different. “Less debt” sounds good. “More wealth” sounds good too. And our wishful thinking has no difficulty demanding both at the same time.

Yet there is another way to become genuinely wealthier.

We can save in real terms. We can create better homes, machinery, energy systems and infrastructure, accumulate knowledge and preserve the natural foundations of our lives. In doing so, we create real wealth without having to build ever larger financial claims against one another.

The difficult part is not necessarily the logic. The difficult part is making that logic thinkable against our own intuitions.

Perhaps this is precisely where artificial intelligence could acquire a future role as a social institution.

Not as a ruler over human beings. Not as a ministry of truth. And certainly not as an instrument for finally creating the “right” kind of person after all. That would merely repeat the old mistake using a new technology.

AI could become something much more modest – and perhaps much more important: an institution that helps us take our own imperfections into account when thinking about society.

Then, finally, we would no longer have to change human beings.

Entrepreneurs could remain self-interested. Employees could continue to put their families first. Savers could continue to want to protect their wealth. And all of us could remain emotional, contradictory, sometimes judgemental and sometimes prone to wishful thinking.

We would merely have to stop designing social systems for a kind of human being who does not exist.

When dealing with rational human failure, we would accept that individually sensible behaviour can produce collectively irrational outcomes and design our institutions accordingly. When dealing with irrational human failure, we would additionally accept that the people designing those institutions are not purely rational beings either.

Perhaps, then, alongside markets and government, we really do need another institution: an artificial intelligence whose task is not to make decisions for us, but to help us recognise when we are about to trip ourselves up.

Because perhaps a better world does not begin with better human beings.

Perhaps it begins with a much simpler realization:

We do not need new human beings. We need to finally learn how to work with the ones we already have.

This idea will also be explored further at friedliche-intelligenz.de.

Related Approaches and Scientific Connections

The distinction proposed in this article between rational and irrational human failure, and its connection to artificial intelligence as a possible tool for social synthesis, can be linked to several existing strands of research. The proposed synthesis itself should therefore not be understood as arising in isolation, but rather as an attempt to connect insights that have so far largely been discussed in different academic contexts.

An important point of connection is the research on Integrative Complexity. It distinguishes between the ability to recognize different perspectives in the first place (differentiation) and the ability to subsequently connect them within a more comprehensive understanding (integration). This ability has been studied for decades, particularly in political psychology, leadership research and conflict resolution.

A second important connection is the extensive research on Motivated Reasoning. It shows that people do not process political and social information solely according to its factual content. Emotions, prior beliefs, moral judgements and social identities influence which information is accepted, rejected or subjected to particularly critical scrutiny.

Taken together, these approaches point towards a possible explanation for why the synthesis of imperfect theories in the social sciences can be so difficult. Emotionally driven judgement tends to classify competing ideas as right or wrong, good or bad. Synthesis, by contrast, requires us to hold different – and sometimes apparently contradictory – partial truths simultaneously and ask under which conditions each of them is valid.

There are also increasingly interesting scientific connections to the possible role of artificial intelligence proposed in this article.

Research on AI-enhanced Collective Intelligence explores how human and artificial intelligence can complement one another rather than treating AI simply as a replacement for human intelligence. The aim is to combine their different capabilities in ways that may create collective problem-solving capacities beyond those of either humans or AI alone.

A particularly concrete example is the Habermas Machine, presented in Science in 2024. In experiments involving more than 5,000 participants, an AI system helped groups with differing political views formulate statements representing common ground. The AI was not intended to decide which participants were right. Instead, it helped identify and formulate perspectives that people with differing views could accept.

Another related approach is Cognitive Dissonance Artificial Intelligence (CD-AI), proposed in 2025. Instead of resolving contradictions as quickly as possible, this approach seeks to preserve competing propositions long enough to encourage critical and dialectical thinking. The underlying idea is particularly relevant to the problem discussed here: apparently contradictory perspectives may contain valuable partial truths that should not be discarded merely because they do not immediately fit together.

These approaches do not appear to formulate the same distinction between rational and irrational human failure proposed in this article. Nor do they appear to combine these two forms of human failure with AI as an institutional aid for synthesizing imperfect social theories in precisely the way suggested here. They do, however, provide important scientific points of connection.

The interesting possibility may therefore lie in their synthesis: individually rational behaviour can produce collectively irrational outcomes, while emotionally influenced human reasoning may simultaneously make it difficult to integrate the different partial insights required to understand and institutionally correct those outcomes.

Artificial intelligence could potentially assist precisely at this second level – not by deciding which theory is “right”, but by helping us keep contradictory partial truths visible, examine the conditions under which each is valid, and search for a more comprehensive synthesis.

Selected references

• Integrative Complexity: research tradition originating with Peter Suedfeld, Philip E. Tetlock and colleagues, examining differentiation and integration of competing perspectives in political thought and decision-making.
• Motivated Reasoning: Ziva Kunda, “The Case for Motivated Reasoning”, Psychological Bulletin, 1990.
• AI-enhanced Collective Intelligence: current research on combining human and machine intelligence for collective problem solving.
• Habermas Machine: Christopher Summerfield and colleagues, “AI can help humans find common ground in democratic deliberation”, Science, 2024. DOI: 10.1126/science.adq2852.
• Cognitive Dissonance Artificial Intelligence (CD-AI): proposed in 2025 as an approach in which AI preserves rather than prematurely resolves contradictory propositions in order to support dialectical and critical reasoning.