We should stop Asking Who Is to Blame

Blame is a useful social shortcut. It is not a metaphysical explanation.

When something goes wrong, the first question is usually: who is to blame?

A person is harmed.
A company fails.
An AI system makes a destructive recommendation.
A self-driving car kills someone.
A platform radicalises users.
A leader creates a toxic culture.
A society drifts into dysfunction.

Almost immediately, our attention moves toward blame. We want a responsible party, a guilty person, a defective decision-maker, an object onto which moral tension can be discharged.

This response is not stupid. It is not merely a failure of reason or a moral defect. It is a useful heuristic, and like many useful heuristics, it exists because it helped solve real problems under real constraints. For most of human history, we did not live in large bureaucratic societies mediated by software, corporations, supply chains, machine-learning systems, legal abstractions, and institutional incentive structures. We lived in small social groups where trust, reciprocity, reputation, and danger had to be assessed quickly.

Someone stole food. Someone betrayed trust. Someone was violent. Someone failed to reciprocate. Someone put the group at risk. In such an environment, slow causal analysis would often have been too expensive. The group needed to know: is this person dangerous? Can they be trusted? Should they be punished? Should they be excluded? Should they be forgiven? Should others imitate them or avoid them?

Blame helped answer those questions quickly.

It compressed a complex causal field into a social judgment. It turned behaviour into character. It allowed the group to say: this person is generous, that person is selfish; this person is brave, that person is cowardly; this person is trustworthy, that person is dangerous. Praise and blame became tools for regulating cooperation, enforcing norms, discouraging harmful behaviour, and coordinating expectations inside the group.

That is why the blame reflex feels so natural. It is not an accident. It is a social technology with evolutionary roots.

But a useful heuristic is not the same as a metaphysical truth.

The Blame Reflex Is a Shortcut

The blame reflex works by simplification. It sees an action and quickly infers an agent. It sees repeated action and quickly infers character. It sees harm and quickly infers responsibility. To a degree, this is useful. If someone repeatedly lies, cheats, or behaves violently, it may be adaptive to treat them as untrustworthy or dangerous rather than reconstructing every causal influence that shaped them.

The problem is that the shortcut can become mistaken for a deep explanation. To say “he did it because he is cruel” may be socially useful. It may help the group identify danger. It may help victims name what happened. It may help others avoid the same person. But philosophically, it explains very little. What produced the cruelty? What shaped the person? What incentives rewarded the behaviour? What constraints failed? What conditions made the act more likely? What could have interrupted it?

The blame reflex usually stops before those questions begin.

This is where the fundamental attribution error becomes important. We tend to explain other people’s behaviour by reference to stable inner traits, while underestimating the role of situation, context, pressure, incentive, history, and circumstance. Someone cuts us off in traffic, and we think: reckless person. Someone fails to respond, and we think: rude person. Someone behaves badly under stress, and we think: bad character.

There may be truth in these judgments. Character matters. Patterns matter. Some people really are more dangerous, more selfish, more impulsive, or more dishonest than others. But character itself is not an uncaused essence. Character is a pattern produced by causes. It is biology, development, habit, memory, trauma, attention, imitation, social reinforcement, culture, incentives, and repeated choice sedimented into a person.

The fundamental attribution error is therefore not simply irrational. It has understandable roots. In social life, we often need fast compression. We cannot calculate the full causal history of every person in every interaction. We need rough models. We need to know who is safe, who is dangerous, who is reliable, who is manipulative, who will cooperate, and who may defect.

So we attribute character. And often we do so too quickly.

The Evolutionary Logic of Moral Agency

In my thesis, I discussed one possible evolutionary explanation for why our sense of agency and free will is so powerful. Liam Clegg’s paper Protean Free Will suggests that something like free will may have roots in the evolutionary advantage of unpredictability. A creature whose behaviour is too predictable becomes easier prey, easier competition, and easier to manipulate. A creature able to generate some degree of pseudo-randomness in its behaviour may gain an advantage simply by being harder to model.

This does not give us metaphysical free will. But it helps explain why the experience of agency might be so strong and so useful. A system that experiences itself as choosing, adjusting, resisting, deciding, and acting is a system that can respond more flexibly to its environment. It can revise itself. It can surprise others. It can become harder to exploit.

The same logic extends socially. In dealing with other moving beings, it is efficient to assume agency (and human-like consciousness). If something moves as if it has goals, we treat it as goal-directed. If something behaves as if it intends harm, we react as if it intends harm. This intentional stance is fast, cheap, and often adaptive.

We do this even with very little information. In the classic Heider and Simmel experiments, people watched simple geometric shapes moving around a screen and spontaneously described them in social and moral terms. One shape was bullying. Another was helping. Another was trying to escape. We do not need much before we start seeing agency. Movement plus pattern is enough for the social mind to construct intention.

That tells us something important. Our minds are built to see agents. And once we see agents, we are built to assign praise and blame.

This makes evolutionary sense. In a social group, it is useful to track who helped, who harmed, who defected, who protected, who shared, who exploited, and who violated the implicit rules of the group. Moral agency, in this sense, is not first a philosophical theory. It is a social operating system.

The problem begins when the operating system is mistaken for metaphysics.

The Heuristic Does Not Survive Metaphysics

The blame reflex assumes more than it can justify. It assumes that beneath the action there is a freely willing person who could have done otherwise in some deep sense. It assumes that the person is not merely the causal site at which behaviour occurred, but the ultimate author of that behaviour. It assumes that praise and blame track something more than consequences, risk, repair, deterrence, and future intervention. It assumes desert. That is where the heuristic breaks.

A person’s behaviour can reveal important things about them. It can tell us what they are likely to do, what risks they pose, what habits they have formed, what values they act from, what incentives move them, and what forms of intervention may be required. In that practical sense, character attribution remains useful.

But it does not follow that the person is metaphysically blameworthy in the strong retributive sense.

The person did not choose their genes. They did not choose their early environment. They did not choose the culture into which they were born. They did not choose the initial architecture of their nervous system. They did not choose many of the experiences that formed their fears, impulses, capacities, and blind spots.

This does not mean they are irrelevant to what happens next. It means they are part of the causal structure, not an exception to it.

The same is true of all of us. We deliberate. We choose. We inhibit impulses. We make commitments. We repair mistakes. We train ourselves. We become more or less trustworthy over time. All of this matters. But none of it requires the idea that we are uncaused agents standing outside nature.

A choice can be real without being metaphysically free.

A person can be accountable without being ultimately blameworthy.

A society can take consequences seriously without believing in retribution.

Free Will Without Magic

Free will is one of the most powerful human experiences. I decide to raise my hand. I feel that I could have left it down. I choose coffee instead of tea. I feel that I could have chosen otherwise. I resist an impulse and experience myself as the author of that resistance. The feeling of agency is immediate and persuasive.

But the experience of agency does not prove metaphysical freedom. A choice can be real without being uncaused. A decision can matter without being magic. The fact that I deliberate does not mean that deliberation floats outside biology, history, and circumstance. It means that a system capable of modelling futures, comparing options, inhibiting impulses, and updating behaviour is doing exactly that.

A more naturalistic picture does not need to deny agency. It only denies that agency requires a ghostly exception to the causal order. Human beings are embodied control systems. We build models of the world and models of ourselves inside the world. We anticipate consequences, regulate impulses, respond to feedback, and revise our behaviour over time. We are not puppets. But we are also not little sovereign spirits standing outside nature.

This matters because the traditional idea of moral responsibility often depends on that ghostly, special picture. If responsibility means being the ultimate, uncaused origin of an action, then responsibility becomes metaphysically suspect. But if responsibility means being part of the causal structure through which harm occurs, repair becomes possible, and future behaviour can change, then responsibility remains deeply meaningful.

The Self Is Not a supernatural Ghost

One way to make this clearer is to stop imagining the self as a metaphysical commander. The self is not a little executive sitting behind the eyes, issuing orders from outside the system. The self is a model: a living, social, embodied, memory-rich model that allows an organism to coordinate itself across time.

This self-model is not an illusion in the sense of being useless or unreal. It is one of the most important structures in human life. It allows me to say: I did this. I want that. I should not do this again. I owe someone an apology. I am becoming someone I do not want to become. These statements matter. They help organise behaviour, memory, identity, responsibility, and repair.

But none of this requires a supernatural will (in the sense of super-natural — above nature, beyond it). It requires attention, memory, emotion, self-representation, social feedback, and control. It requires a system capable of modelling itself as an agent and modifying its future behaviour in light of that model.

From this perspective, responsibility is not abolished. It is relocated. The question is no longer whether a person could have acted otherwise in some ultimate metaphysical sense. The question is whether the system can understand, update, regulate, repair, and be changed.

Letting Go of Blame Does Not Mean Letting Go of Consequences

People often worry that if we weaken belief in free will, ethics will collapse. The fear is simple: if nobody is ultimately free, nobody is responsible; if nobody is responsible, everything is permitted.

This fear rests on a confusion. The absence of metaphysical blame does not mean the absence of consequences. A bridge can collapse without free will; we still investigate the collapse. A virus can spread without free will; we still intervene. A dog can bite without free will; we still restrain, train, protect, and adapt. A company can cause harm without being conscious; we still regulate it. An AI system can produce damage without intending anything; we still audit, constrain, redesign, or remove it.

Consequences matter because causes matter. If a person is dangerous, that danger must be addressed. If a company’s incentives produce harm, those incentives must be changed. If a model’s deployment creates unacceptable risk, the system must be constrained. If a culture repeatedly rewards destructive behaviour, the culture must be examined.

What falls away is not accountability. What falls away is retribution as metaphysical payback.

That distinction is essential. Retribution asks what suffering someone deserves because of what they did. Causal accountability asks what happened, who was harmed, what caused the harm, what repair is owed, and what intervention would reduce future harm.

Retribution wants moral balance through punishment. Causal accountability wants causal repair.

Accountability Without Hatred

Causal accountability may be mistaken for moral leniency. It is not. In practice, it can be more demanding than blame because it refuses to stop at condemnation.

A person who has caused serious harm may need to be restrained, removed from a position of authority, required to make repair, monitored, treated, or imprisoned. A company that has caused harm may need to pay damages, change governance, expose internal records, redesign products, or face structural penalties. An AI system that has caused harm may need to be audited, constrained, retrained, disconnected, or banned from particular contexts.

These are consequences. They may be severe. But their justification is not that suffering itself is morally good when directed at the guilty. Their justification is protection, repair, deterrence, rehabilitation, truth, and systemic correction.

This is where the word “deserve” does a great deal of hidden work. Sometimes people use it loosely to mean that consequences are appropriate. But often it means something stronger: that the suffering of the guilty is itself a moral good. That is the part I suggest to reject strongly. Once suffering becomes the goal, ethics as a rational endeavour has been contaminated and corrupted.

Causal accountability does not ask how much pain restores the balance of the universe. It asks what intervention changes the future.

From Blame to Causal Accountability

The question, then, is not whether the blame reflex has a functional value. It does. It helped social groups move quickly, enforce norms, detect danger, reward cooperation, and coordinate expectations. It still plays a role today. We should not pretend to be creatures without moral emotions.

But we should also not let an evolved shortcut govern our deepest ethical thinking.

Blame is fast. Causal accountability is slower. Blame asks who the bad person is. Causal accountability asks what produced the harm. Blame looks for a target. Causal accountability looks for a causal architecture. Blame compresses behaviour into character. Causal accountability asks how character, situation, incentive, history, system design, and institutional pressure interacted.

This does not make causal accountability softer. In many cases, it is stricter. It may still require restraint, removal, punishment in the legal sense, compensation, redesign, public accountability, or institutional consequences. But the purpose changes.

We do not act because someone deserves suffering as metaphysical repayment. We act because harm is real, because victims matter, because risk must be reduced, because repair is owed, and because systems that produce harm must be changed.

AI Makes the Problem Obvious

AI makes the weakness of blame harder to ignore because it reveals how distributed agency often is.

Imagine a self-driving car kills someone. Who is to blame? The car? The passenger? The programmer? The company? The training data? The regulator? The road designer? The executive who pushed deployment? The team that ignored a warning? The sensor manufacturer? The culture that rewarded speed over safety?

The blame question wants a clean target. The system does not provide one. The event emerges from hardware, software, data, incentives, testing, regulation, user behaviour, road conditions, institutional pressure, and design decisions. Agency is distributed across the entire causal architecture.

This does not mean nobody is accountable. It means accountability must be traced through the system. Where did the failure enter? Where was risk created? Where was warning ignored? Who had control? Who benefited? Who was exposed to danger? Which intervention now reduces the probability of recurrence?

Maybe the software must be changed. Maybe the company must be liable. Maybe deployment was premature. Maybe regulation failed. Maybe the safety culture was broken. Maybe the passenger misused the system. Maybe the road environment made predictable failure more likely. Usually, several things are true at once.

Blame collapses complexity. Causal accountability preserves it.

The Algorithm Is Not Guilty

When an AI system causes harm, it is tempting to speak as if the algorithm is guilty. But that is moral theatre. The algorithm did not wake up and decide to be cruel. It did not hate the user. It did not enjoy discrimination. It did not intend humiliation.

Yet the system may still be ethically significant. A hiring model that filters out qualified applicants is ethically significant. A credit model that reproduces historical injustice is ethically significant. A recommender system that amplifies self-harm content is ethically significant. A chatbot that manipulates vulnerable users is ethically significant. A military targeting system that makes lethal errors is ethically significant.

The question is not whether the model deserves blame. The question is what causal role it plays. What data shaped it? What objective function guided it? What evaluation failed? What incentives pushed deployment? What oversight was missing? What human judgment was displaced? What institution benefited? What harm followed? What should now be changed?

This is not a weaker form of accountability. It is the only form that understands the object. The algorithm is not guilty in the human sense, but it can still be the relevant causal node. Changing it may change the future. That is enough to bring it within the scope of accountability.

Humans Are Also Systems

It is easy to apply causal thinking to machines. Of course the model behaved this way because of training data, objective functions, architecture, deployment, and feedback. But the uncomfortable symmetry is that humans are systems too.

A violent person was not born outside causality. A corrupt executive was not born outside causality. A negligent engineer was not born outside causality. A manipulative leader was not born outside causality. Their actions emerge from biology, development, incentives, personality, trauma, culture, opportunity, ideology, stress, attention, habit, and choice.

Choice is part of the system, not outside it.

This does not excuse harm. It explains how harm becomes possible. Explanation is not exoneration. Understanding is not forgiveness. Identifying causes does not remove consequences. It makes consequences more intelligent.

This is one of the hardest points to hold because our moral psychology often treats explanation as a threat. If we explain why someone became violent, it can sound as if we are excusing the violence. If we explain why a leader became corrupt, it can sound as if we are softening the corruption. But explanation and excuse are different operations. An explanation tells us where intervention may be possible. An excuse tells us not to intervene. Causal accountability requires the first and rejects the second.

The Moral Use of Anger

None of this means anger has no place. Anger is information. It tells us that a boundary has been crossed, that harm has occurred, that a value has been violated, that someone or something may need protection.

But anger is a signal, not a philosophy. It should start inquiry, not replace it. The problem with blame-based ethics is that it often lets anger decide what truth is, what justice requires, and when investigation may stop.

Causal accountability uses anger differently. It takes the signal seriously, then asks for analysis. What exactly happened? Who was harmed? What interests were violated? What pattern does this reveal? What needs protection? What must change?

Anger can wake us up. But it should not run the whole system.

Leadership Without Scapegoating

This distinction matters especially in leadership. When something goes wrong inside an organisation, blame is tempting because it is fast, simple, and psychologically protective. It creates the appearance of action. Someone failed. Fire them. Move on.

Sometimes a person really does need to be removed. But often the deeper failure belongs to the system. A team misses warnings because the culture punishes bad news. A product ships too early because growth is rewarded more than safety. Employees burn out because leadership praises resilience while designing overload. A scandal happens because everyone knew the incentive structure was broken, but nobody wanted to name it. An AI system causes harm because nobody had ownership of the full causal chain.

Blame finds the nearest visible person. Causal accountability asks what made the outcome more likely.

A serious leader does not only ask who failed. A serious leader asks what about the system they built made this failure more probable. That is a harder question, which is why it is often avoided.

Blame Is Low-Resolution Ethics

Blame is low-resolution ethics. It takes a complex causal field and compresses it into a moral label: good person, bad person, responsible, irresponsible, guilty, innocent, hero, villain.

This can be useful in simple cases. But it breaks down in complex ones. A society is not a village. A corporation is not a single person. A platform is not a hammer. An AI system is not a moral subject in the ordinary sense. A human mind is not a free-floating will. A crisis is rarely caused by one decision alone.

We need higher-resolution ethics. Ethics that can see systems, incentives, feedback loops, institutional pressures, psychological conditioning, and technological mediation. Ethics that can see minds as embodied, self-modelling control systems. Ethics that can see AI systems as causal agents inside larger institutional machinery. Ethics that can still protect victims, constrain harm, and demand repair without pretending that blame explains the world.

A useful shift is to move from courtroom imagination to control-room imagination. The courtroom asks who is guilty. The control room asks where the signal is coming from, which feedback loop is unstable, which constraint failed, which incentive is misaligned, and which intervention changes the trajectory.

Both models have their place. But in a technological society, we need far more of the second.

Causal Accountability Is More Demanding

Causal accountability is more demanding than blame because it refuses easy moral closure. Blame can satisfy itself with condemnation. It finds someone to denounce and then relaxes. Causal accountability keeps asking what produced the outcome.

It asks for the full chain: the person who acted, the environment that shaped them, the incentives that rewarded them, the warnings that were ignored, the institution that benefited, the culture that normalised it, the design that enabled it, the victims who were harmed, the repair that is owed, and the future that must be prevented.

This is harder because there may be no clean villain. There may be several responsible nodes. There may be uncomfortable complicity. There may be design failures rather than dramatic evil. There may be cultural patterns we participate in. There may be incentives we quietly benefit from.

That is precisely why the shift matters. Causal accountability does not let us purify ourselves by hating the guilty. It asks us to change the conditions.

The Same Applies to Ourselves

This is not only about AI, law, corporations, or institutions. It is also personal.

When I react badly, I can blame myself. I can say: I am terrible, weak, broken, guilty. Sometimes guilt contains useful information. It can point to a violated value. But blame often becomes paralysis. It turns attention inward in a way that produces shame rather than change.

Causal accountability asks a better set of questions. What happened in me? What was triggered? What was I protecting? What pattern repeated? What condition made this more likely? What repair is needed? What practice would change the future? What conversation must happen? What boundary must be set? What support do I need? What must I train?

This does not remove responsibility. It makes responsibility actionable. You are not a ghost guilty of failing to transcend causality. You are a system that can learn. That is more useful and more honest.

Responsibility After Free Will

So what remains of responsibility after free will?

A lot. But it must be rebuilt.

Responsibility no longer means being the uncaused origin of an action. It means being part of the causal structure through which action occurs, harm is produced, repair is possible, and future behaviour can change.

I am responsible when my behaviour affects others and when I am one of the places where understanding, repair, constraint, or transformation can occur. A company is responsible when its decisions, incentives, and products affect others and when it is one of the places where intervention must occur. An AI system is causally accountable when its operation contributes to ethically significant outcomes and when changing the system changes the future. A society is responsible when it builds conditions that reliably produce harm and then pretends to be surprised by the result.

This is responsibility without metaphysical theatre.

No ghost. No ultimate desert. No cosmic scoreboard. But consequences, repair, protection, redesign, and learning.

What We Should Ask Instead

When something goes wrong, we need better questions.

Not: who can we hate? But: who was harmed?
Not: who deserves pain? But: what repair is owed?
Not: who is metaphysically guilty? But: what causal structure produced this?
Not: who can we sacrifice so the system feels clean again? But: what must change so this becomes less likely?

These questions are less satisfying in the short term because they do not give us the immediate emotional clarity of blame. But they are more useful because they preserve contact with reality. They help us protect victims, repair damage, constrain dangerous systems, redesign incentives, and prevent recurrence.

Blame will likely not disappear. It is too deeply human and we are a long way from transcending our cognitive intuitions and biases. Nor does it need to disappear entirely for the time being. There are situations where direct condemnation is appropriate, especially when harm is ongoing, denied, or defended. It can act as a suitable cognitive programming or adjustment tool. But blame should not be the foundation. It should not be where ethical inquiry ends.

In a world of AI systems, corporations, platforms, institutions, incentives, psychological conditioning, and complex human minds, blame is too crude.

The point is not to make ethics colder. It is to make it more precise. More honest. More capable of repair. More capable of prevention. More capable of dealing with the kinds of systems we now live inside.

The question is not who deserves suffering in return. The question is what happened, what caused it, who was harmed, what must be repaired, and where in the causal architecture of the situation we can intervene so that the future becomes less harmful than the past.

That is accountability without metaphysical theatre.

That is responsibility after free will.