Traditionally, cognitive neuroscientists have been probing the relation between brain states and mental states by manipulating the mental state of the participant through different conditions and then measuring the associated changes in neural activity, for example by means of Functional Magnetic Resonance Imaging (fMRI) or Electro-encephalogram (EEG). The results of these manipulations are usually taken to reflect a correlation between mental states and brain states, rather than a “genuine” causal relation. According to several neuroscientists, however, new brain stimulation techniques, such as Deep Brain Stimulation (DBS) and Transcranial Magnetic Stimulation (TMS), allow us to go beyond correlations and establish causal relations between mental states and brain states [for a review, see Ref. (1)]. This has important implications for other disciplines in which these techniques become increasingly popular. For example, in psychiatry, DBS has proven to be an effective treatment for patients with major depressive disorder (MDD) who do not respond to pharmacotherapy or psychotherapy (2–4).

In the current paper, we investigate whether and to what extent it is indeed justified to draw conclusions about causal relations between brain and mental states on the basis of cognitive neuroscience studies. In the next section, we start with a description of an interventionist account of causation, which is inspired by Woodward (5). We argue that this account is more or less in line with how causation is understood in scientific practice. The question is, however, whether it can be used to make causal claims about the interaction between mental states and brain states. In order to address this question, we introduce the notion of supervenience in Section “Mental States and Brain States: A Supervenience Relation.” This notion aims to capture the intuition that mental states are dependent on, but not identical with, brain states. In Section “Causation in Traditional Cognitive Neuroscience Studies,” we turn to Baumgartner’s “causal exclusion” argument. According to this argument, the assumption of a supervenience relation violates the criteria of what counts as a good intervention. As a result, we cannot draw conclusions about the causal relation between mental states and brain states. In his reply to Baumgartner, Woodward (6) proposes to adjust these intervention criteria in order to make room for supervenience relations and to secure causal claims on the basis of traditional cognitive neuroscience studies. In Section “Causation in Brain Stimulation Studies,” we discuss the consequences of both positions for causal claims on the basis of brain stimulation studies. Most importantly, we will show that Baumgartner’s causal exclusion argument does not apply to these studies. That is, we can make causal claims about brain stimulation studies even if we assume a supervenience relation and accept Woodward’s original intervention criteria. In Section “Articulating the Mind–Brain Relation,” we show what happens to these conclusions if we adopt a different view of the relation between mental states and brain states. Finally, in Section “Conclusion,” we briefly discuss the implications of our findings for psychiatry and the treatment of psychiatric disorders.

In most textbooks on experimental research two main requirements are described that an experiment must meet to be able to reveal a causal relation between X and Y. The first is that the levels of X must be systematically varied and the second is that all variables other than X and Y are to be controlled in order to eliminate other possible causes of Y. If these requirements are met and changes in X are accompanied by changes in Y, one is allowed to speak of a causal relation between X and Y (7, 8).

This notion of how to investigate causal relations in scientific practice is very much in line with a philosophical account of causation that has become quite popular recently: interventionism. One of the most established interventionist definitions of causation comes from Woodward (5):

(M) A necessary and sufficient condition for X to be a (type-level) direct cause of Y with respect to a variable set V is that there be a possible intervention on X that will change Y or the probability of Y when one holds fixed at some value all other variables Z_i in V. A necessary and sufficient condition for X to be a (type-level) contributing cause of Y with respect to variable set V is that (i) there be a directed path from X to Y such that each link in this path is a direct causal relationship… and that (ii) there be some intervention on X that will change Y when all other variables in V that are not on this path are held fixed [Ref. (5), pp. 59].

We mainly focus on the definition of a direct cause since this comes closest to the notion of causation as it is investigated in scientific practice (i.e., it explicitly involves the two requirements mentioned above). However, for the definition to make sense, we also need a clear notion of what an appropriate intervention is. Woodward (5) defines an intervention variable as follows:

(IV) I is an intervention variable for X with respect to Y if:

I causes X;

Finally, relative to the notion of an intervention variable an (actual) intervention can be straightforwardly understood in terms of an intervention variable I for X with respect to Y taking on some value z_i such that I = z_i causes X to take on some determinate value z_j [(5), pp. 98]. In terms of experimental design, an intervention can be seen as a manipulation that changes the variable X. In order for this manipulation to be able to reveal a causal relation, it has to meet the requirements in (IV).

Can we use interventionism to make causal claims about the interaction between mental states and brain states? To answer this question, we will (initially) assume a very minimal relation between mental states and brain states – one that captures the intuition that mental states are dependent on brain states. In the philosophy of mind, this relation is known as “supervenience.”

A schematic representation of a supervenience relation between mental states M1 and M2 and brain states P1 and P2 is depicted in Figure 1. Although the notion of supervenience has been much discussed, there are two features that are common in most definitions:

(S1) ¬(M causes P) ∧ ¬(P causes M);

This means that (i) supervenience is a non-causal relation such that neither M causes P nor vice versa¹ and (ii) any change in mental state is necessarily accompanied by a change in brain state. Furthermore, with regard to Figure 1, we will assume that:

(S3) P1 causes P2.

The end result is a schematic representation of two types of relations: one between properties (M1 and P1, M2 and P2), which is captured by a supervenience relation, and one between events (M1/P1 and M2/P2), which is captured by a causal relation (i.e., event 1 causes event 2).

With the interventionist account of causation and the notion of supervenience in place, let us now take a closer look at traditional (non-invasive) cognitive science studies.

In most of these studies, the relation between mental states and brain states is investigated by observing the effect of changes in mental state M1 on brain state P2 (see Figure 2). This is done by manipulating the mental state of the subjects by letting them participate in separate conditions that differ on some stimulus characteristic or task that is meant to induce changes in M1. To investigate the effect of these manipulations on brain states, the subjects’ brain activity P2 is measured in all conditions. Then, if the researcher has made sure that the conditions only differ on the manipulated mental variable (using all kinds of controls like randomization of subjects), and a (significant) difference in brain activity between the conditions is found, the researcher concludes that the manipulated mental variable M1 has had an effect on the measured brain state P2.

However, is it valid to conclude that the change in mental state M1 caused the change in brain state P2? According to the causal exclusion argument put forward by Baumgartner (9), it is not.

Since the introduction of brain stimulation techniques such as TMS and DBS, it has become possible for scientists to directly manipulate (intervene on) the electrical activity in the brain. Many neuroscientists have been using these techniques to draw conclusions about the causal relations between brain states and mental states. The following are quotes from TMS and DBS studies published in high-impact journals:

“Making the causal link: frontal cortex activity and repetition priming” (10).

“Causal implication by rhythmic TMS of alpha frequency in feature-based vs. global attention” (11).

“DBS of the subthalamic nucleus markedly improves the motor symptom’s of Parkinson’s disease, but causes cognitive side effects such as impulsivity” (12).

“Stimulation of a restricted site in the upper midbrain can cause major acute depression” (13).

Are these claims justified? In the present section, we will explore this question in light of Baumgartner’s and Woodward’s arguments. Before we continue, we should mention that a large part of the brain stimulation studies only focuses on the effects of changes in brain states on other brain states [e.g., Ref. (14–16)]. This is the relation between P1 and P2. As described in (S3), we assume that there exists a causal relation between these variables. Therefore, in these types of brain stimulation studies, it is perfectly justified to talk about causal effects.

In the brain stimulation studies in which the relation between brain states and mental states is investigated, this seems more complicated. In these studies, the relation between P1 and M2 as depicted in Figure 4 is investigated. P1 is manipulated by stimulating a certain brain area using TMS or DBS in one condition and not stimulating it in another condition, while measuring some mental variable M2 in both conditions. The researcher tries to make sure that the two conditions only differ on P and not on other variables, for example by applying sham stimulation in the control condition. If then a (significant) difference in M2 between the two conditions is found, the researcher concludes that there was an effect of the change in brain activity on the mental state. However, is he or she justified in saying that P1 has caused M2?

The conclusions drawn in the previous sections rely heavily on the assumption of a supervenience relation between brain states and mental states. We believe that most neuroscientists would agree with this assumption. Quoting one of the key textbooks in cognitive neuroscience programs: “Cognitive neuroscience is an academic field concerned with the scientific study of biological substrates underlying cognition, with a specific focus on the neural substrates of mental processes” [(17), p. 12]. This definition suggests that what lies at the heart of cognitive neuroscience is a dependency relation between mental states (“cognition”) and brain states (the “neural substrate”).

Supervenience is not a “deep” explanatory relation; however, it only indicates the presence of a dependence relation without telling us what it is (18). A common way to further explain this dependency is by appealing to the notion of emergence. Central to emergentism is the idea that supervenient properties are “novel” properties over and above the properties upon, which they supervene. In the context of the mental causation debate, emergentism can be understood as the more specific claim that mental states are the emergent properties of a complex physical system, which have their own causal power and cannot be reduced to the basic physical properties of this system.² It is also possible to explain the dependence relation between mental and physical properties in terms of reduction. In contrast to emergentism, reductionism claims that supervenient properties are reducible to their base properties, and hence that mental properties are reducible to physical properties.

Thus far we have investigated whether interventionism allows us to make causal claims about the relation between mental states and brain states, given a notion of supervenience that in principle allows mental states to be causally efficacious qua mental. In other words, we have assumed a dependence relation between mental states and brain states that is non-reductive, in principle compatible with the notion of emergence, and less strong than an identity relation. However, some people might not agree with this characterization of the relation between mind and brain. What happens to the conclusions drawn in this paper when one wishes to assume a reductive relation between mental states and brain states instead? In what follows we will briefly explore this option and also consider the possibility of a causal relation.

The aim of this paper was to investigate whether we can draw conclusions about causal relations between brain states and mental states from traditional cognitive neuroscience studies and brain stimulation studies, given an interventionist account of causation. We have argued that, if one follows Woodward in embracing the notion of supervenience and revising the criteria for what counts as an intervention, both types of studies can be used to establish causal claims. If, by contrast, one follows Baumgartner and his causal exclusion argument, traditional cognitive neuroscience studies cannot be used to establish causal claims but brain stimulation studies can.

Brain stimulation is being used more and more as a form of treatment for psychiatric disorders. We have shown that from an interventionist point of view, it is valid to say that these brain stimulation treatments cause changes in mental states. Now this is not necessarily an argument in favor of these treatments. However, if Baumgartner is right, then it seems reasonable to conclude that brain stimulation treatments will become increasingly attractive. Unlike traditional cognitive neuroscience studies, they actually have the potential to elucidate the causal structure of certain psychiatric disorders (i.e., the underlying causal relations between mental states and brain states). It is a safe bet that this will appeal to many psychiatrists.

At the same time, this conclusion is based on a “conservative” interpretation of interventionism, and a rejection of non-causal metaphysical relations between mental states and brain states such as supervenience. In this respect, Woodward’s position is much more “liberal,” insofar as it proposes adjusted intervention criteria and allows for the inclusion of non-causal relations between mental states and brain states. One advantage of Woodward’s position is that it allows psychiatrists to draw conclusions about the causal structure of mental disorders from traditional cognitive neuroscience (and not just brain stimulation studies). Another and perhaps even more important advantage is that it legitimates the claim that cognitive and behavioral therapy, aiming at influencing the mental state of a patient, can cause changes in the patient’s brain state.

A note of caution is required when applying our conclusions to psychiatric practice. When defining mental states and brain states as separate targets of intervention, we assume an ideal situation in which such a separation can be easily obtained, and Woodward’s intervention criteria are met. In practice, however, such a situation might be difficult to achieve. For example, in their paper on degeneracy, Price and Friston (26) have argued that different neural configurations can lead to similar mental states. This means that a disruption of one of these configurations by brain stimulation might not necessarily lead to a change in mental state. The experimenter who uses interventionism in the context of a single study would then be forced to conclude that there is no causal relation between the brain state and mental state in question. Now this scenario could be avoided by making sure that conclusions about causal relations between mental states and brain states are supported by multiple studies (controlling for both inter- and intra-individual variation). However, there might be a larger worry here, not just about the fact that the application of interventionism to single studies in practice might sometimes result in misguided causal claims, but also about the very possibility of applying interventionism to cognitive neuroscience studies.

For example, one might argue that various factors such as degeneracy, redundancy, path-dependency, non-linearity and complex feedback loops make it (theoretically) impossible to establish linear causal chains.³ In the light of this, several theorists have proposed a concept of “circular causation” (27–29). Circular causation, which is taken to be typical of a self-organizing system, is realized by the cooperation of the individual parts of the system, yet it also governs or constrains the behavior of these individual parts. A good illustration of circular causation is given by McGilchrist (30) in his account of the brain as a complex system: “Events anywhere in the brain are connected to, and potentially have consequences for, other regions, which may respond to, propagate, enhance or develop that initial event, or alternatively redress it in some way, inhibit it, or strive to re-establish equilibrium. There are no bits, only networks, an almost infinite array of pathways” (2010, p. 34).

Circular causation is attractive, but also slightly misleading – at least when it is articulated in opposition to linear causation. As Von Bertalanffy (31, 32) already pointed out, to make sense of circular causation we still require a notion of linear and “unidirectional” causation. The kind of feedback regulation that is implied by circular causation is obviously not unidirectional in spatial terms: it moves back and forth or circles around the various components of a system (33). However, despite circling in space, feedback still proceeds forward in linear time, one component being separated from the next in time. Circular causation, thus understood, is compatible with the assumption of a supervenience relation between mental states and brain states. The question is whether it is also compatible with interventionism. Let us say we propose a modified version of Figure 1 that involves feedback loops, for example one in which P1 causes P2 which in turns causes P1 (Figure 5A). Now, at first glance, such a feedback loop seems to violate the second requirement of (IV), in the sense that one might think that P1 not only depends on the value taken by I but also on the value of P2. However, the problem is that such a depiction of circular causation fails to take into account the fact that the relations between P1 and P2, and P2 and P1 are temporal relations between different events, and therefore they cannot be circular. That is, the P1 that causes P2 is different (not spatially, but temporally) from the P1 that is caused by P2 as the result of the feedback loop. The correct (linear) way to represent circular causation is shown in Figure 5B: P1 causes P2, and P2 causes P1* (which is temporally different, but spatially identical with P1). And this seems to be compatible with interventionism, to the extent that an intervention on P1 does not violate the second requirement of (IV).

It is important to note, at this point, that interventionism as such is relatively free of metaphysical commitments, in the sense that it does not make claims about how exactly one should spell out the relation between mental states and brain states. It only tells us what needs to be in place and which conditions need to be met for a given relation between variables to be described as “causal.” Furthermore, as we have shown in Section “The Interventionist Account of Causation,” one of the main attractions of interventionism is that it seems to correspond to how causal relations are investigated in scientific practice. Therefore, even if interventionism turns out to be incompatible with certain assumptions about brain functioning, such as circular causation and feedback loops, then this indicates a larger problem with mainstream scientific method and textbook accounts of experimental research. Obviously, these are issues that deserve critical attention. For the purpose of this paper, however, we have taken the mainstream scientific method as our starting point.

In the end, how psychiatrists approach these issues will probably depend on their intuitions about interventionist causation and the relation between mind and brain. However, one thing is certain. What they conclude will have important implications for the way they communicate the effect of different treatments to their patients.

ND wrote the core sections of this paper (“The Interventionist Account of Causation,” “Mental States and Brain States: a Supervenience Relation,” “Causation in Traditional Cognitive Neuroscience Studies,” and “Causation in Brain Stimulation Studies”) and the main argument; LB wrote the other Sections “Introduction,” “Articulating the Mind–Brain Relation,” and “Conclusion,” revised/rewrote Sections “The Interventionist Account of Causation,” “Mental States and Brain States: a Supervenience Relation,” “Causation in Traditional Cognitive Neuroscience Studies,” and “Causation in Brain Stimulation Studies and supervised the project.”

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.