Dodging the magic bullets 

Denny Borsboom



In one of my favorite movie scenes from The Matrix, protagonist Neo finds out how to see through the appearances and dodge the bullets fired at him. The moment he understands how appearances are related to reality, Neo gains the power to predict and manipulate that reality. Understanding, prediction, and control form the classical trinity of scientific goals, and that’s why I love that scene. For me, it’s a metaphor for science at its best: when a new understanding of reality suddenly allows for a radical improvement of prediction and control.

Although scientific work tends to be quite a bit less romantic than the mystical revelations of Neo, I do think scientific progress often involves a shift of perspective that suddenly opens the door to new ways of thinking about old problems. Complex systems approaches tend to shift a lot of perspectives.


As an example, take the phenomenon of panic attacks. Panic attacks are sudden bouts of intense anxiety, often accompanied by hyperventilation and a strong urge to escape a situation. During a workshop at the Institute for Advanced Study in Amsterdam, working with a diverse group of scientists from fields ranging from psychiatry to ecology, we constructed a computational model for this phenomenon. We based the model on Clarke’s (1986) theory, which states that panic attacks arise out of feedback between bodily arousal and perceived threat. In his theory, panic attacks can arise when bodily arousal is interpreted as signifying an impending threat. This perceived threat serves to increase arousal further, which in turn increases the level of perceived threat. This means the system enters a runaway feedback loop that propels it into a full-fledged panic attack. Such panic attacks will arise if the feedback loop between arousal and perceived threat is strong, and this is thought to depend on the degree to which a person is prone to interpret bodily arousal as a sign that something’s wrong (“I think I am having a heart attack”). This is called anxiety sensitivity.


My colleagues Don Robinaugh and Jonas Haslbeck constructed a visualization of this process in the below movie. You can see how, for low levels of anxiety sensitivity, random fluctuations do not affect the system much. However, for high levels of anxiety sensitivity, the system produces behavior that clearly resembles the phenomenology of panic attacks. In my view, this simple model offers a quite interesting explanation for how the phenomenon of panic attacks arises. 

The feedback loop of panic attacks creates a macroscopic state of the human system that is very unpleasant. People often get these attacks in social situations – e.g. a supermarket or a movie theatre – which is not just frightening but also embarrassing. In the model that we built, these panic attacks support a second, slower feedback loop between the fast feedback loop above and avoidance behavior. That’s when things start to get really nasty. By avoiding situations in which a panic attack arises, the person precludes the opportunity to learn that they need not in fact be harmful; they cancel the opportunity to obtain information that falsifies beliefs that fuel the interpretation of arousal as signifying threat. As a result, they may increasingly withdraw from the social environment – panic disorder is often comorbid with Agoraphobia. 

This slow feedback loop represents a learning process that is common in anxiety disorders. A therapist once told me the story of a boy who frantically avoided the lines separating tiles, because he was convinced that he would be eaten by crocodile if he stepped on the lines. When his therapist asked “Have you ever tried stepping on a line, to see what happens?”, the boy answered “Of course not! That’s why I’m still alive!”. For these reasons, it’s a bad sign when a person starts engaging in what therapists call “safety behaviors” – e.g., when a person books a chair closer to the exit of a theatre, just in case. Safety behaviors can culminate in avoidance and if that happens there are bound to be problems.


In psychological research, I have often noted that researchers are inclined to search for the cause of a phenomenon, and typically they think of this cause as being a singular entity. In our group, we call this the magic bullet fallacy: the idea that there is some singular level of explanation that should be prioritized: a simple, one-line explanation of a phenomenon that identifies the cause of X. It’s the brain. It’s the genome. It’s the environment. It’s personality. It’s the gut. It’s the trauma. It’s the repressed memory. It’s the p-factor. The list runs on. But suppose, for the sake of argument, that something like our panic model is true. Then what is the cause of panic disorder? Is it the effect of arousal on perceived threat? The effect of perceived threat on arousal? The degree of anxiety sensitivity? The cognitive content of the representation of arousal as a sign of danger? The neural implementation of that representation? The random fluctuations in arousal? The external and internal factors that cause the arousal levels? The slow feedback loop? Avoidance? Technically, each of these factors may be taken to qualify as a cause, because interventions on them will change the probability of panic attacks. That’s because each of the factors in the panic model is needed to generate the phenomenon of panic disorder.


Yet if God descended from heaven and gave you this list of causes along with the assurance that the list is complete, you still wouldn’t be able to explain panic attacks just by looking at these factors. That’s because the secret in this kind of explanation is not which factors are involved, but how they connect to each other in the system[1]. So sometimes you can hold all the causes in your hand, but you may still not understand the phenomenon because you’re not able to put the pieces together.


The insidious coordination between slow and fast feedback loops explains how panic attacks can lead to panic disorder. Such feedback loops arise throughout psychopathology. Patients with Anorexia Nervosa have been shown to judge their body as being fatter than it is, which affirms their conviction that they are too fat, and this cognitive state is likely to bias their perception again. Patients with agoraphobia overestimate the size of spiders, which likely strengthens their belief that spiders are dangerous. Alcoholics may use drinking as a coping mechanism, and may drink to forget the problems that they have because they drink. I had a friend with a gambling addiction, who was genuinely convinced that the proper way to deal with losing money at the slot machine was to try to win the money back.


When we ask the question how environmental factors, such as those that arise in an urban environment, contribute to mental disorders, perhaps we should not seek causes of mental disorders. Perhaps we need to think how urban factors get entangled in feedback loops. The presence of bars and coffeeshops, the density of the population, the loneliness that can arise in the urban environment, the pace of life, the structure of the social network: from microscopic to macroscopic and from slow to fast, many of these factors are possible candidates for getting entangled in feedback loops.


Of course, systems theory is no panacea. Human beings are an order of magnitude harder than anything else we study, because they imbue their world with meaning, and that meaning matters. When the system under study starts directing it’s behavior based on it's own subjective interpretation of what that behavior means, you get interactions that aren’t just complex in the complexity science sense, but that feature a form of reflexivity that, as far as I can tell, exists nowhere else in the known universe. Since we don’t have an inkling about how that works, detailed mechanistic explanations (e.g., “down to the level of the brain”) of mental disorders won’t be cracked as long as I remain on the good side of the grass. But that’s ok. As Skinner said: Human behavior is the most difficult topic ever subjected to scientific analysis. You can’t expect to crack a topic like that in a few centuries.


But at least we can start thinking in systems, to dodge the magic bullets.




Clark, D. M. (1986). A cognitive approach to panic. Behaviour research and therapy, 24(4), 461-470. doi:10.1016/0005-7967(86)90011-2

Meadows, D. H. (2008). Thinking in systems: A primer. Vermont: Chelsea Green Publishing.

Robinaugh, D., Haslbeck, J. M. B., Waldorp, L., Kossakowski, J. J., Fried, E. I., Millner, A., … Borsboom, D. (2019, May 29). Advancing the Network Theory of Mental Disorders: A Computational Model of Panic Disorder.

[1] Donella Meadows provides perhaps the most compelling account of this line of reasoning in her book Thinking in systems (if you don’t know this book, please do yourself a favor and buy it today). 

Denny Borsboom (1973) is a Dutch psychologist and psychometrician. Since 2013, he has worked as a professor of psychology at the University of Amsterdam. His work focuses on applying network theory to the study of mental disorders and their symptoms. In 2018 he presented the Paul B. Baltes Lecture at the Berlin-Brandenburg Academy of Sciences and Humanities.

Previous blog posts