Epiphenomenalism and the Epistemic Argument

Abstract

The epistemic argument against epiphenomenalism aims to prove that even if epiphenomenalism is true, its adherents are not able to justify their inferential beliefs. This would mean that they cannot know that they are right which is a self-stultifying consequence. I elaborate on this problem and then present an updated version of epiphenomenalism based on property dualism. I argue that this position is capable of refuting the conclusion of the epistemic argument even in spite of accepting its essential assumptions. This was made possible by an upgraded property exemplification account of events. I also argue against a view which, if true, gives substantial support to the epistemic argument: that a belief justified by other beliefs is knowledge only if it is caused by those beliefs in virtue of their contents.

1 Introduction

One of the most widely accepted ideas in metaphysics of mind in the last century is the mental causation thesis, which states that our mental states, such as beliefs or conscious experiences, have causal powers and cause our behavior as well as other mental states. This thesis, when presented as above, seems to be very intuitive or even obvious, and it is endorsed by metaphysicians of nearly all sorts: physicalists and dualists, reductionists, and anti-reductionists. However, commitment to mental causation causes considerable trouble for many of them. Every major position, apart from reductive physicalism, is prone to charges that it cannot evade at least tacit commitment to epiphenomenalism, a theory according to which mental states are not causally effective. Epiphenomenalism is very rarely openly endorsed and is much more frequently attacked, so some could surely find defending it a rather pointless endeavor. On the other hand, one could also think that since epiphenomenalism is so often imputed to other theories, it is worth checking whether its critique is justified and epiphenomenalism, along with the positions which lead to it, should be avoided.

Led by this motivation, in this paper, I will pay attention to the epistemic argument, according to which even if epiphenomenalists are right, they are unable to know it, because beliefs about the verity of philosophical positions are generally obtained by inference and inferential knowledge is justified only if caused by the beliefs serving as its premises. I want to show that a certain brand of epiphenomenalism, epiphenomenalist property dualism, can accept the major premises of the epistemic argument yet still not accept its conclusion. This (not new itself) position will be formulated with the use of an upgraded property exemplification account of events, which is based on MacDonald and MacDonald’s (2006) proposal that events are property exemplifications and there may be multiple properties exemplified in a single event. I also consider the stronger condition that inferential knowledge must be caused by other beliefs in virtue of their contents, which is seemingly at odds with epiphenomenalism. My conclusion is that the arguments for this condition available in the literature are unsatisfying and epiphenomenalists’ potential knowledge that they are right has not been proven to be contradictory to their very position.

In Section 2, I will briefly present how epiphenomenalism can be understood and why—despite what some might think—it should occupy an important place in present debates in metaphysics of mind. Section 3 will be dedicated to reconstructing one of the most severe charges against epiphenomenalism: the epistemic argument, as presented by Dwayne Moore (2014). In Section 4, I will present epiphenomenalist property dualism, a position which is suitable to fight off the epistemic argument. This will be preceded by an explication of a general account of events which is well-suited for property dualists’ needs. Next, in Section 5, I will show how this position deals with this argument. I will also suggest how the epistemic argument might be strengthened by demanding that inferential knowledge is caused in virtue of its premises’ contents; Elizabeth Fricker’s (1993) reasoning in support of this condition will also be presented. From the perspective of epiphenomenalist property dualism I will argue against this reasoning. Section 6 sums up what has been achieved in this paper.

2 Epiphenomenalism and Its Role in the Metaphysics of Mind

Epiphenomenalism in its broadest formulation can be thought of as a thesis stating that the mental sphere of the world does not causally affect anything in the world. This, however, can be understood in more than one way. The most widely used distinction was made by Brian McLaughlin (1989), who discerned two main varieties: type epiphenomenalism and token epiphenomenalism. According to type epiphenomenalism, events can be causes in virtue¹ of falling under physical types but not in virtue of falling under mental types. In contrast, according to token epiphenomenalism, physical events can cause mental events, but mental events cannot cause anything (McLaughlin 1989: pp. 109–110). Most epiphenomenalists of the last half a century or so were committed only to type epiphenomenalism, the weaker of the two. These include Campbell (1970), early Jackson (1982) or late Kim (2005). Token epiphenomenalism was historically first and endorsed by early epiphenomenalists like Huxley (1874), but it can also be attributed to one of the most prominent contemporary epiphenomenalists, William Robinson (2019b).

Although explicit defenses of epiphenomenalism still appear in the debate (most notably Robinson 2019b, but also, e.g., Gadenne 2006, Lyons 2006, and Tammelleo 2008), it is much easier to find philosophers who are adamantly sure that epiphenomenalism is false (e.g., Burge 1993: p. 102; Tye 1995: p. 19; Fodor 1989: p. 59, Dretske 1988: p. 80). Showing the epiphenomenalist consequences of a theory tends to be treated as a successful reductio ad absurdum of this theory (Horowitz 1999: p. 423). Therefore, proving that epiphenomenalism is not as unacceptable as it is usually taken to be would not only support those who openly endorse it but also those whose views are accused of having epiphenomenalist consequences; instead of proving that they do not have to reject mental causation, the latter group could agree that some sort of epiphenomenalism is part of their theories and deny that it constitutes a problem (this suggestion seems parallel to what Umut Baysan cited as his motivation for undermining the validity of epiphobia; Baysan 2020).

The cluster of positions potentially threatened by the menace of epiphenomenalism is numerous. Jaegwon Kim’s classic causal exclusion argument (Kim 1989, 1993) is designed to prove that mental causation cannot be upheld in any version of non-reductive physicalism; it also similarly affects the vast majority of property dualists (Kim 1998: p. 58). Furthermore, Kim has an argument which aims to show that substance dualists also cannot secure mental causation (Kim 2005: pp. 78–85). Since every kind of dualism and non-reductive physicalism leads to epiphenomenalism, the only position able to uphold mental causation is reductive physicalism. Kim draws a thesis which he calls conditional reductionism: “If mentality is to have a causal influence in the physical domain—in fact, if it is to have any causal efficacy at all—it must be physically reducible” (Kim 2005: p. 161). After adding that epiphenomenalism is incompatible with reductionism, the following dilemma emerges: either reductive physicalism is true (and epiphenomenalism is false) or epiphenomenalism is true (and reductive physicalism is false).²

At this point, it should already be clear why epiphenomenalism is an important issue: if Kim is right and conditional reductionism is true, then accepting mental causation leaves us with only one position available in metaphysics of mind. This would be a huge blow for the majority of philosophers in this field, many of whom are property dualists or non-reductive physicalists of some sort. This is why the charges against epiphenomenalism should be carefully examined in order to determine whether everyone must accept reductionism, or whether the proponents of non-reductive positions can still hold them after they have been weakened by the rejection of mental causation.

In this paper, I delve into the objection against epiphenomenalism that Robinson (2019a) takes to be the most powerful: the epistemic argument, according to which epiphenomenalism is incompatible with some crucial types of knowledge that virtually nobody would ever like to abandon. I dedicate the next section to describing this argument, as constructed by Dwayne Moore (2014).

3 The Epistemic Argument

Moore’s epistemic argument against epiphenomenalism is part of his analysis of Kim’s conditional reductionism thesis, which makes us face the following dilemma: either we accept epiphenomenalism, or we embrace mental causation along with reductionism. Moore argues that both ways lead to trouble with grounding inferential knowledge, thus weakening the plausibility of conditional reductionism and bolstering the attractiveness of non-reductive physicalism. Only the first part of Moore’s argument (the one concerning epiphenomenalism) is of interest for me now as it is supposed to prove that even if epiphenomenalism is true, its supporters cannot have knowledge about it.

Moore starts by introducing a justification for his endorsement of mental causation. He considers how one can ground inferential knowledge: if an agent A knows that p, then the following statements are true: A believes that p, this belief is justified, and p is true. For inferential knowledge, this justification should also be inferential: p is inferentially justified for A iff A previously possessed some beliefs q₁,…,q_n from which one can infer p. However, such conditions are not sufficient to determine that A knows that p through inference because they do not exclude the possibility of so-called epistemic luck, namely, a situation in which an agent possesses a true belief and this very fact is accidental or lucky in some way. Moore proposes a thought experiment intended at showing that when a belief is true due to lucky circumstances, it should not be deemed knowledge. He describes a situation in which a scientist hooks up some human named Ruppert to his machine. The machine has a database in which each natural number from 1 to 10 million has a unique proposition assigned to it; the machine randomly draws three numbers in this range and then forms a syllogism from the propositions assigned to the drawn numbers; finally, the machine sends a signal to Ruppert’s brain which forces him to repeat the syllogism and to form beliefs with contents identical to previously mentioned propositions.³ After many cases of nonsensical and non-valid reasonings, the machine happens to draw numbers that correspond with propositions that form a valid syllogism and Ruppert says “All humans are mortal and Socrates is a human, therefore, Socrates is mortal”; he also forms beliefs corresponding to these propositions. In this case, Ruppert has a true belief that Socrates is mortal and this belief is inferentially justified in virtue of Ruppert’s beliefs that all humans are mortal and that Socrates is a human. However, one would not like to state that Ruppert knows that Socrates is mortal as he formed this belief independently of his beliefs, which serve as its justification. What follows is that in order to talk about knowledge it is not sufficient to show a justification based on belief types (that is, propositions which are contents of particular beliefs). What is still needed is some relation between tokens: beliefs themselves.

Moore finds such a relation by taking inspiration from Davidson (1963: pp. 3–4), who introduced the notion of a primary reason: a pair consisting of a) some positive attitude (e.g., a desire or urge, but also a moral view or social convention) towards actions of certain kind; and b) a belief that some specific action is of that kind. The contents of a primary reason’s constituents stand in a logical relation (forming a sort of practical syllogism) with the action, and a positive attitude-belief pair constitutes a primary reason only if this relation between them and some action is present. For example, Kelly might have a desire to satisfy her hunger and a belief that eating lasagna is pleasurable, but there is no appropriate logical relation here as the belief is not connected to the desire. However, Kelly’s desire to satisfy her hunger and her belief that eating lasagna satisfies hunger together constitute a primary reason for eating lasagna (but not for eating, e.g., shakshuka because the desired logical relation does not obtain between the contents of the above propositional attitudes and eating shakshuka).

One can have many such reasons for an action: Kelly might also have a desire to please her wife and a belief that she can please her wife by eating lasagna. This is why a primary reason ought to be somehow singled out, and Davidson states that this can be done by adding causality to the mix: agents perform actions because they have reasons for them, and the only way to explain this “because,” according to Davidson, requires not only a logical relation but also a causal relation between a reason and an action (Davidson 1963: pp. 9–11). Kelly’s desire to satisfy her hunger and her belief that lasagna will do the job constitute Kelly’s primary reason for eating lasagna when they cause this to happen.

Moore notices an analogy between the above situation and justifying inferential knowledge. In the latter case, reasons are present too: if Ruppert is to know that Socrates is mortal, then his beliefs about human mortality and Socrates’ humanity should be reasons for his belief about Socrates’ mortality. He would have the last belief because he would have the previous ones. When transferring the reasoning from action to inference, this “because” should also point not only to a logical relation but also to a causal relation between the reasons and the conclusion. In order for agent A’s true belief p to be her inferential knowledge it not only needs to be inferable from beliefs q₁,…,q_n, which A already had, but also q₁,…,q_n have to be causes for p. Ruppert’s machine-induced beliefs do not appear together in any single causal chain, so they do not fulfill the updated conditions for inferential knowledge, and this is the desired outcome.

Since the conditions for inferential knowledge are now established, the epistemic argument against epiphenomenalism can be put forward. The essential epiphenomenalist thesis about non-obtainment of mental causation should surely be reachable via reasoning from certain premises. Moore chooses an example of a Kim-style reasoning that leads to epiphenomenalism: if [P1] no property or event can have more than one sufficient cause (this is the principle of causal exclusion), and [P2] every property and event has a sufficient physical cause (this is the principle of causal closure of the physical world), and [P3] mental properties and events cannot be identical with physical properties of events (the principle of psychophysical irreducibility), then [P4] mental properties and events do not cause anything (and this is epiphenomenalism).

We can treat P1, P2, P3, and P4 as propositions expressed by the above statements. The epistemic argument goes like this: let us imagine that these propositions are contents of Charlotte’s (an epiphenomenalist convinced by the above reasoning) beliefs, m₁, m₂, m₃, and m₄. Suppose that propositions P1–P4 are true and epiphenomenalist Charlotte has true beliefs m₁–m₄. Her belief about the truth of epiphenomenalism (m₄) is inferentially justified because she also possesses beliefs m₁, m₂ and m₃, from which m₄ is inferable. However, this is not enough for m₄ to constitute a case of knowledge because m₁, m₂, and m₃ also need to be causes of m₄. They are beliefs, and beliefs are paradigmatic cases of mental states. If epiphenomenalism is true, then mental states cannot be causes of anything, so beliefs m₁, m₂, and m₃ cannot be causes of m₄. Conditions of inferential knowledge are not fulfilled, so Charlotte’s belief that epiphenomenalism is true is a case of epistemic luck. Generalizing, epiphenomenalists never know and are never in a position to know that the position they endorse is true. This does not render epiphenomenalism blatantly false, but it drastically reduces its attractiveness: belief in something about which knowledge is unreachable might be acceptable in the field of, e.g., religious beliefs, but it seems that one should expect a bit more from philosophers and their theories.

Moore does not try to justify the principle that one should not endorse a statement which one can never know is true: he just considers self-stultifying every position whose supporters cannot know about its verity. One might still wonder whether this assumption is true. For example, Bart Streumer argues for an error theory about all normative judgements, yet he still says that not only can we not know that such a theory is true, but we cannot even have such a belief (Streumer 2017: pp. 129–138). He also cites other philosophical positions which imply the impossibility of believing them with no effect on their integrity and verity, like eliminative materialism (Streumer 2017: pp. 193–195). I do not want to consider whether epiphenomenalists can in a similar fashion agree with the epistemic argument and consider its conclusion to be harmless for them. In this paper I want to show how this argument can be refuted; therefore, for my current reasons I will accept Moore’s assumption that the impossibility of possessing knowledge that some theory is true renders this theory self-stultifying.

The epistemic argument and other similar charges against epiphenomenalism which try to prove its self-stultification have usually attracted responses which put in doubt the causal theory of knowledge by showing that there are other some relations between mental states⁴ which are not directly causal but nevertheless strong enough to secure knowledge. Such voices have come from the likes of Chalmers (1996), Nagasawa (2010), or Robinson (2006), the last of whom engaged in a discussion with Moore on this topic. He proposed that an earlier mental state m can make a later belief b⁵ a case of knowledge if they have a common underlying cause c, which necessitates both m and b even in the absence of a causal relation between them. Since c can be a physical state, this option is open for epiphenomenalists. This prompted a response from Moore (2012) who claimed that if epiphenomenalism is true then c necessitates m merely nomologically, while only metaphysical necessitation would turn b into knowledge. Robinson’s idea that such dependencies (in general—psychophysical laws) hold in our world but do not hold in many possible worlds makes Moore wonder how are we to know that they actually hold. If there is no causation from m to b, there should at least be a strong, metaphysically necessary dependency between them mediated by some other factor, like a common cause c; otherwise even if we correctly identify that psychophysical laws hold in our world, it will be a case of epistemic luck and not a case of knowledge. Robinson (2013) responded by pointing at the special status on psychophysical laws which are not discovered and validated in the same way as physical laws because, in contrast to physical states, we do not observe our mental states; we just have them. While his point is important for all epiphenomenalists projects which postulate the nomological necessity of psychophysical laws or psychophysical supervenience, I am not going to pursue his argumentation and develop a token epiphenomenalist position like his. Instead, in the rest of this paper I will show how epiphenomenalists can accept the causal theory of knowledge and retain some sort of mental causation while keeping the epiphenomenal character of mental properties. I will do so by presenting a type of epiphenomenalist position which, I believe, is fit to deal with Moore’s reasoning by using the distinction between type epiphenomenalism and token epiphenomenalism that was overlooked by Moore.

4 Epiphenomenalist Property Dualism

Of course, Moore is aware that epiphenomenalism can be understood in two ways. In a footnote, he notices that the principle of mental causation can be read in two ways (Moore 2014: p. 150) which are parallel to the token/type epiphenomenalism distinction. He states that his argument works well against both types of epiphenomenalism; however, I am not so sure about this and I would like to present a position from the type epiphenomenalism family that may deal with the epistemic argument. “Epiphenomenalist property dualism,” as I will call it, is not a new term (e.g. Ravenscroft (2018) uses it to describe Jackson’s position), and the theory it refers to is not groundbreaking in any way. Apart from early Jackson (1982) other prominent authors who could be given this label are Chalmers (1996), late Kim (2005) or Campbell (1970), with the last one being the closest to what I will present. The intuitions for epiphenomenalist property dualism have been present in the debate for over half a century now; however, the term has not been used in a systematic fashion and I hope to make the position clearer by explicating the commitments necessary for any its adherents.

The metaphysical basis of my position is property dualism, which can be thought of as an intermediary theory between Cartesian substance dualism and non-reductive physicalism: it differs from the former because it denies the existence of any sort of non-physical substance (so it is a substance monism), and it differs from the latter because it denies that psychophysical supervenience obtains with metaphysical or stronger necessity. This choice is motivated by the assumption that in order to rebut the epistemic argument it is helpful (if not necessary) to posit unity of substance but without equating the mental and physical realms—and this is what property dualism openly does. A good starting point for defining property dualism is David Chalmers’ statement that “conscious experience involves properties of an individual that are not entailed by the physical properties of that individual, although they may depend lawfully on those properties. (…) This is not to say it is a separate «substance».”⁶ (Chalmers 1996: pp. 110–111) This means that property dualism does not have to get rid of the widely accepted psychophysical supervenience. Supervenience is a transitive, antisymmetric covariance relation, but what is relevant here is a more particular psychophysical supervenience that is usually understood along the lines laid out by Kim (1998): p. 39: mental properties/events supervene on physical properties/events iff for any X which instantiates any mental property m at t, there is a physical base property p such that X has p at t and necessarily anything with p at a time has m at that time, where this necessity is at least nomological, namely it results from the laws of nature which obtain in the actual world. Simply speaking, the mental states of an individual depend on her physical states and cannot change unless these change. Chalmersian property dualism denies that supervenience can obtain with logical or metaphysical necessity, but it provides space for supervenience with nomological necessity.

This stance also makes it possible to identify mental events with physical events because of unity of substance. Property dualists can, as Davidson does,⁷ claim that “psychological events simply are (in the sense of are identical with) physical events”. (Davidson 1973: p. 248) For example, when I suddenly remember that it is my friend’s birthday today, for a property dualist this remembering is not only a mental event but also a brain event in which the prefrontal cortex and the hippocampus are activated. It is not a case of relating two distinct events with a causal link or their co-occurrence: there are not two events but one; remembering my friend’s birthday is literally a certain activation of the prefrontal cortex and the hippocampus.

I should also determine the scope of events that would contain non-physical properties: in order to provide a non-trivial answer to Moore, I shall assume that not only qualitative states (as is usually accepted by epiphenomenalists like early Jackson (1982) or Robinson (2019b)) but also other states that are widely accepted to be mental, including intentional states, contain non-reducible mental properties (such an approach to intentional states is present e.g. in Campbell’s (1970) epiphenomenalism). This is necessary because Moore’s argument is based on the assumption that epiphenomenalists take beliefs to be as epiphenomenal as qualia, but I want to challenge this argument while accepting as many of its assumptions as possible.

One might ask whether it is really a dualism and not a sort of non-reductive physicalism, like Davidson’s anomalous monism, which also combines substance monism with token identity of physical and mental events. I classify this view as a dualist position by following Chalmers who defined materialism (which for present purposes is synonymous to physicalism) as the doctrine that “every positive fact is entailed by the physical facts” (Chalmers 1996: p. 110). In dualism, on the other hand, although there might be some dependency between mental and physical facts, it must be weaker than entailment: if there are any psychophysical laws, they are contingent. Token identity of physical and mental events does not lead to physicalism if this identity is contingent—in other words, if there could possibly be a world where such identity does not obtain. Property dualists, by claiming that supervenience obtains with at most nomological necessity, subscribe to the metaphysical possibility of a world with other dependencies between physical and mental facts. Furthermore, they explain this contingent token identity of events by the appeal to properties, which cannot be both mental and physical. Property dualism entails realism about properties, while physicalism does not have to be committed to the existence of properties. For example, Davidson (1970, 1993) formulated his anomalous monism, a brand of non-reductive physicalism, using an ontology consisting only of particular objects and events.

The above talk of events and properties calls for a specification of at least the relation between these two types of entities. I have already committed myself to the view that events are property exemplifications and now I am going to elaborate on what I mean by it and propose an account of events, which might suit property dualists better than any existing account, while still being unproblematic for non-dualists. The reason why I think it is important to work out an updated definition of event is that standard theories tend to define events in a physicalist framework, via spatiotemporal regions (like Lewis 1986 or Quine 1985) or their causes and effects (Davidson 1969). Kimian events (Kim 1976) might be acceptable for substance dualists, but not for property dualists, for reasons which I will provide in a moment. In general, proponents of any kind of double aspect theory have troubles seeking a suiting event theory on the market. This includes Chalmers (1996) or Campbell (1970), both of whom could be called epiphenomenalist property dualists. However, I am not interested in a theory of events tailored specifically to the needs of property dualists. While we do not have to expect that theories of events will be neutral with respect to other metaphysical issues, I think that an ecumenical account could be considered an improvement and is what should be sought after. Unless an universal theory of events is on the table, it is hard to compare different brands of physicalism, dualism and others with respect to what they say about events. Below is my attempt at defining events in a way in principle acceptable for property dualists and all other metaphysicians of mind, which will help me articulate epiphenomenalist property dualism.

The proposal will be based on Kim’s account of events, according to which every event is constituted by a unique triple: a substance (the event’s object), a constitutive property (exemplified by the object) and a time (Kim 1976: p. 160). So, for Kim, an event e = ⟨o, p, t⟩ is an event constituted by an object o having a property p at the time t. However, this definition of events seems like it cannot accommodate the idea that mental events are physical events but mental properties are distinct from physical properties. According to property dualism a single mental event should be an exemplifying⁸ of at least two constitutive properties (a physical one, which would make it a physical event, and a mental one, which would make it a mental event), whereas Kim says that each event is individuated by a single constitutive property—the one which makes a causal difference (Kim 1976: p. 162). However, as Lombard (1986: pp. 54–55) points out, one can accept a property exemplification account of events and agree that events are individuated by its constitutive properties without claiming that there is only one property exemplified in a single event. Such a multi-property account has been championed by MacDonald and MacDonald (2006). In their case it was tailored specifically to the needs of non-reductive physicalism and accompanied with views on causation to which I do not subscribe; however, I believe that it can be presented in a way that is acceptable and useful for all parties of the debate. What I will now propose is an account of events which draws on the MacDonald and MacDonald model but unlike them I do not consider it to support non-reductive physicalism—I want to show it as a neutral account which suits well property dualists, among others.

An event is, to expand the original Kimian model, a tuple e = <o, p₁,…, p_n, t> where o is an object, which at the time t has properties p₁,…,p_n. The set of properties which constitute the event cannot be all or any arbitrary properties exemplified at the same time by the object. We do not want a multitude of spurious events such as an event of Gary feeling anxious and having two legs or an event of a rock being of volcanic origin and moving with a constant speed. All properties exemplified in a single event must be somehow related and relevant to each other. We should also avoid the possibility of different events exemplifying the same properties. A particular rock moving with a speed of 5 meters per second is not a different event from this rock moving with a constant speed and certainly not different from and event of this rock moving with a constant speed of 5 meters per second. These are different descriptions of the same event of a rock moving. The same with mental events in a property dualist framework – me believing that epiphenomenalism is true and me being in a certain underlying physical state at the same time are not different events; they are the same event which is an exemplifying of multiple properties.

Generally, the properties which we will want to be constitutive of one event are those between which there holds some necessitation relation, be it nomological or stronger supervenience, determination, or conceptual entailment. If (and only if) an object exemplifies p₁ along with p₂ at the same time and exemplifying p₁ necessitates exemplifying p₂, then it is sensible to talk about one event of exemplifying p₁ and p₂. This condition can be formulated by stating that for each event e=<o, p₁,…, p_n, t> there is a set S whose members are all and only constitutive properties of e, such that 1) S is not empty; 2) a property p belongs to S only if o has p at t; 3) p belongs to S iff for any p_n belonging to S, having p at t necessitates or is necessitated by having p_n at t.⁹

Let’s focus on mental events. For those property dualists who postulate nomological psychophysical supervenience, each time an object exemplifies a mental property, it must also exemplify a subvenient physical property at the same time. This means that an event which is an exemplifying of a mental property is also an exemplifying of a physical property because of the necessitation relation—and vice versa for those physical properties which are supervenience bases for mental properties. Every mental event is also a physical event. For example, when I speak of my headache I refer to a particular event which is an exemplifying of my quale, the subjective way in which I feel the headache, and also an exemplifying of my physical state, which necessitates the occurrence of this quale. My headache is not only this or only that; it is both. It is also an exemplifying of endlessly many other properties, such as my having a mild headache (if it is actually mild), my having some ache, and even my being conscious.¹⁰ Non-reductive physicalists may adopt the same picture—they will just posit a stronger psychophysical supervenience relation. Reductive physicalists will claim that any mental property I have is just a physical property, so my headache is just me having a property of having a headache (and some other entailed properties like having a body). For substance dualists when I have a headache there is a mental event which is an exemplifying of my quale and a distinct physical event which is an exemplifying of my physical property, of being in a certain brain state. These properties do not necessitate each other so they cannot be a part of a single event. This account of events also covers all possible events, for example, physicalists may describe events in possible (or impossible) dualist worlds without changing the way they describe events.

Having defined events, I can characterize property dualism using the following three theses:

[PD1] All events are physical events and they are exemplifyings of physical properties.

[PD2] Some physical events are also mental events, and they are exemplifyings of physical properties as well as mental properties.

[PD3] Mental properties are not reducible to physical properties, nor do they supervene on physical properties with necessity stronger than nomological necessity.

This gives a view of the world with no non-physical events, so all mental states are physical events—probably brain states¹¹—and their place in the physical world (their physical properties, causes and effects) is in principle describable by appropriately developed natural sciences, possibly neuroscience. However, there is one aspect of mental states which neuroscience cannot describe, namely mental properties, which are exemplified in mental states. Mental properties are what distinguishes our actual world, which is filled with conscious creatures from a hypothetical “zombie world” which is physically identical to our world but lacks any conscious experiences (Chalmers 1996: p. 84). While mental events are identical to some physical events, mental properties are always distinct from physical properties and their mutual dependency and correlation is contingent and could obtain differently.

Such constructed property dualism is now ready to be joined with type epiphenomenalism. Let’s recall that, according to McLaughlin, type epiphenomenalism states that “events can be causes in virtue of falling under physical types, but events cannot be causes in virtue of falling under mental types”. (McLaughlin 1989: p. 109). For the property dualist, falling under physical or mental types is equivalent to being exemplifyings of physical or mental properties, so the type epiphenomenalism that will suit me can be formulated in the following way:

[TE] Events can be causes only in virtue of being exemplifyings of physical properties.

Mental properties are not causally relevant, but this does not stop mental events from being causes as their causal powers are provided by their physical properties.

Conjunct theses [PD1], [PD2], [PD3], and [TE] define a position which I will call epiphenomenalist property dualism (EPD). According to EPD, all mental events are physical events and are thus part of the physical world. This means they have causal powers but not in virtue of their mental properties; this allows us to call mentality an epiphenomenon. To make the last point more plausible, let us take notice of what follows: if all the events from our world could be stripped of their mental properties, then no causal relations would be lost in the process; on the other hand, if we got rid of all physical properties and left only mental properties, then the remaining world would be a hardly imaginable creation devoid not only of physicality but also of causality. Now that it is clear that EPD indeed belongs to the epiphenomenalist family, I can proceed to confronting it with the epistemic argument.

5 Response to the Epistemic Argument

Let us recall the story about Charlotte, an epiphenomenalist who reached her view via inference. Charlotte inferred her belief that epiphenomenalism is true (m₄) from her other beliefs m₁, m₂, and m₃, which logically entail m₄. According to Moore, if epiphenomenalism is true, then mental states, including beliefs, cannot be causes, so m₁, m₂ and m₃ are not causes of m₄. Due to this, m₄ does not fulfill the causal condition for inferential knowledge, and Charlotte does not possess the knowledge that epiphenomenalism is true. But what if Charlotte endorses EPD (and if EPD is true)? According to EPD, all events are physical events; in particular, mental states are physical events—brain states in this case. Charlotte has the belief m₄, and she is also in some physical brain state p₄. These are both the exact same event: they have the same object (namely Charlotte), they are exemplifyings of the same properties (the physical property of being in a brain state p₄ and the nomologically connected mental property of having the belief m₄) and they occur at the same time. Brain state p₄ has brain states p₁, p₂, and p₃ among its causes, which are identical to beliefs m₁, m₂, and m₃. Just like every other event in the world, they are all exemplifyings of physical properties, but they also contain certain mental properties, in virtue of which they fall under some mental types. Since p₁, p₂, and p₃ are causes for p₄, and states p₁–p₄ are states m₁–m₄, then m₁, m₂ and m₃ are causes of m₄. They are causes in virtue of their physical properties, namely those we talk about when we talk about brain states. States m₁–m₄, as intentional states, also have contents which can be taken to be propositions; in virtue of the logical relations between these contents, m₁, m₂ and m₃ inferentially justify m₄. Charlotte’s belief that epiphenomenalism is true is inferentially justified by her other beliefs and is caused by these beliefs. So, if epiphenomenalism (in the form of EPD) is true, then Charlotte’s belief fulfills the conditions for inferential knowledge and Charlotte knows that epiphenomenalism is true. The epistemic argument is not a threat for EPD.

Or maybe this conclusion is too hasty? In a further part of his paper, Moore adds one more condition for inferential knowledge: for m₄ to be a case of knowledge, not only the beliefs m₁, m₂, and m₃, which are reasons for m_4, must be its causes, but they also must cause it in virtue of their intentional and logical properties (we can call them “content” without any loss in meaning). To support this requirement, he once again uses an analogy to action: inspired by Dretske’s classic opera singer example, Moore presents a slightly modified version in which an opera singer confesses a crime onstage through a high-pitched shriek. In the front row, there is a policeman who scowls after the confession (Moore 2014: p. 163). The singer’s shriek is surely a cause of the policeman’s scowl, and in order to also count as the reason for the scowl, the shriek needs to cause it in a specific way, namely in virtue of being a confession, so, more generally, in virtue of its content. If the policeman scowled instinctively because he involuntarily scowls every time he hears a high-pitched sound, the scowl would not count as a rational action because it would be caused in virtue of some physical, non-intentional and non-logical property of the shriek.

If this is true and reasons for action need to be causally efficacious in virtue of their reasonableness, it might be tempting to treat all reasons this way, including the reasons for inferential beliefs—and this is what Moore wants to do. The updated causal condition takes the following general form:

[CC] In order for agent A’s true belief p to be her inferential knowledge, p needs to be inferable from beliefs q₁,…,q_n, which A already had, and q₁,…,q_n have to cause p in virtue of their contents.

This is a more serious threat for epiphenomenalism than the sole causal requirement, which I have already shown to be harmless for EPD. Remember that, according to [TE], events can be causes only in virtue of being exemplifyings of physical properties. This stands in tension with [CC]: for a belief to be a case of inferential knowledge, some other beliefs must cause it in virtue of their content. EPD could agree with this only if it provided a physical, naturalized account of content, but then it would not be an epiphenomenalism (at least not with respect to intentional states). The only way out for the epiphenomenalist is to reject [CC].

What are the arguments for [CC]? One has already been given above: the analogy to reasons for actions. Since an action can be deemed reasonable only if it is caused in virtue of the very properties which act as reasons for this action, then it would seem natural to accept the same condition for reasoning. However, a mere analogy to a similar yet distinct phenomenon does not seem to constitute a convincing argument, especially as there is a significant difference between actions guided by reasons and knowledge gained by reasoning. In the case of the opera singer and the policeman, it can easily be seen how the shriek could cause the same reaction from the policeman, but it would not do so in virtue of its content. If the policeman scowls every time he hears a high-pitched sound, then it is quite obvious that his action was caused by some physical properties of the shriek which are fully irrelevant to its content and independent from it. The shriek could have easily had the same physical properties but no content (e.g., if the opera singer intended to sing gibberish and neither she nor the policeman knew the language in which the sounds she made did actually mean something). However, if we look at inferential reasoning, where both reasons and their effects are beliefs, it is not that easy to find a similar example. Physical properties of mental states and their content are not as independent of and irrelevant to each other as the pitch of a shriek and this shriek’s content. How could we make an inference to the conclusion that epiphenomenalism is true if our premises bore no content? There are undeniable correlations between physical and mental properties of mental states, so it seems highly implausible that states with the physical properties of a certain belief have no content, at least in the actual world. Epiphenomenalist property dualism is allowed to claim that these correlations are a product of psychophysical supervenience as long as it obtains with only nomological necessity, while no such relation of dependency obtains between the physical and intentional properties of a shriek. This is where the analogy fails.

But there should be plenty of other arguments for [CC], right? Surprisingly, there aren’t many. Most proponents of [CC], e.g., Taylor (2000: p. 209) or Martin (1993: p. 72), just take it for granted. I managed to find only one argument for this condition which does not make use of an analogy to rational action and it came from Elizabeth Fricker (1993). Since, for the reasons outlined above, I don’t consider said analogy to constitute a good rationale for [CC], I will now turn my attention to Fricker’s argument. This seems to be an important step as she is the only author who clearly and systematically argued for a condition which looked intuitively right for many others. Fricker’s (1993: pp. 276–277) argument comprises three premises:

[Pr1] The main aim of an agent engaged in reasoning is finding what is true.

[Pr2] Attempts to find what is true by means of entertaining a series of thoughts with different contents must be predicated on there being appropriate systematic relations between these contents.

[Pr3] These appropriate systematic relations must be causal relations.

These premises lead to a conclusion:

[C] An agent engaged in reasoning can reach her aim only if there are causal relations between contents of thoughts.

If we take “causal relations between contents” to mean that causal relations between thoughts obtain in virtue of these thoughts’ contents, then [C] is very closely related to [CC]. Reasoning generally seems to work most of the time, so if causation in virtue of content is necessary for reasoning to succeed, then causation in virtue of content should be included in our account of gaining inferential knowledge. An argument for [C] can thus be considered to validate [CC].

How can an epiphenomenalist, or more specifically a proponent of EPD, reject Fricker’s argument? [Pr1] is rather uncontroversial and [Pr2] also seems to be well motivated: since content is the truth-evaluable element of thoughts and the systematic success of finding what is true needs to be explained somehow, postulating some systematic relations between contents sounds about right. The most vulnerable part of the argument is [Pr3], as no reason is provided for saying that causal relations are the only type of relation which can adequately and systematically link contents.¹² As I mentioned previously, EPD can accept nomological psychophysical supervenience¹³, and it is a relation which should fulfill the condition presented in [Pr2]. It does not relate contents directly, but the connection between them is stable and systematic. When a belief p causes belief q, the situation looks like this: p is an exemplifying of various physical properties which can be grouped into one physical property p_p, and also some mental properties, including content which can be grouped into a mental property m_p. The same can be said of belief q: it contains some physical property p_q and some mental property m_q, the latter of which includes the content of q. All mental properties supervene on physical properties, so m_p supervenes on p_p and m_q supervenes on p_q. There is the causation relation between p and q which obtains in virtue of p_p and p_q¹⁴, so there is a link of nomological dependencies between the contents of p and q. The content of p supervenes on p’s physical properties; a causal relation obtains between p and q in virtue of their respective physical properties, and q’s physical properties are a supervenience base for q’s content. This establishes a systematic relation between contents without granting them causal powers.

To be fair, Fricker anticipated a response along these lines and added in a footnote that although one might say that her argument shows “only that all the systematic nomological dependencies must hold”, she thinks that “such «virtual causation» is causation” (Fricker 1993: p. 277). I do not think that I need to agree with this, although I can see how one might reach such a belief. It might be related to Kim’s notion of “supervenient causation”, which, in most general terms, obtains between two supervenient entities only when an appropriate causal relation holds between the supervenience bases of those supervenient entities (Kim 1984: p. 53). This supervenient causation, which is a type of what Fricker calls “virtual causation”, is supposed to be as good a causal relation as any other. The situation I presented above, namely one belief causing another in virtue of both their physical properties but also with mental properties supervening on these physical properties, might look similar enough to supervenient causation to think that here a genuine causal relation obtains between beliefs in virtue of their supervenient mental properties. However, it must be noted that not every causal relation between two supervenience bases (or in virtue of their occurrences) establishes a causal relation between respective supervenient entities (or in virtue of their occurrences). Several reasons could be brought up here, of which I will present two.

First, Kim himself wrote in his later work that “causality requires a domain with a space-like structure (…) and, as far as we know, the domain of physical objects is the only domain with a structure of that kind”. (Kim 2005: p. 151) Supervenient causation might work well in the case of supervenient physical properties, but irreducible mental properties are not physical, and causality obtaining in virtue of non-physical properties is not so compelling. Second, the relation between supervenient properties should be insensitive to realization in order to be considered a genuine causal relation. This means that it must hold regardless of the actual supervenience bases for these properties. In our situation, the relation between contents m_p and m_q would be insensitive to realization if every physical property which can realize m_p caused some physical property which realizes m_q. No matter how a belief p is realized physically, its content m_p should ensure that it would cause some other belief q with content m_q. Is this the case? I am not convinced. It seems plausible that the causal powers of various physical realizers of the same belief type could vary quite significantly. Some realizers of this belief type will ensure that some other beliefs will follow and some will not. In such a case, the relation between supervenient properties such as contents is too unstable to be a genuine causal relation.¹⁵ Therefore, I do not have to concede that a nomological dependency between contents of beliefs makes it the case that the causal relation between these beliefs obtains in virtue of their contents. [Pr3] should be denied, so the whole argument for the causal efficacy of content can be rejected by epiphenomenalist property dualists. Fricker does not provide a convincing case for [C] and thus neither for [CC].

This strategy of arguing against [CC] faces one more potential objection which deserves mention. In another paper, Moore (2012) claims that just nomological connection between a belief and some prior state which serves as a justification for this belief does not make for a reliable epistemic mechanism. His point, made in response to Robinson (2006), pertains to knowledge about one’s own qualitative states, but one could try and extend it to inferential knowledge. It would go as following: imagine a nomologically distinct world where Charlotte, the epiphenomenalist who reached her view by inferring it from three prior premises, exemplifies all the physical properties of beliefs p₁–p₄, which in the actual world constitute a valid reasoning for epiphenomenalism. However, in this other world the psychophysical dependencies differ from those from the actual world, so these beliefs are exemplifyings of different mental properties, i.e., have different contents. The premises are, say, beliefs that sky is blue, that Reagan is dead and that nuts are high in fats—and let the conclusion that epiphenomenalism is true remain. Physically these beliefs are just as the valid premises in the actual world, including the causal relations between them, so the basic causal condition of inferential knowledge is met. However, it seems plainly wrong to claim that the product of this reasoning is knowledge. Why epiphenomenalists consider Charlotte’s belief a case of knowledge in our world although it wouldn’t be knowledge in a world with the same physical and causal story? Charlotte’s belief seems to be just lucky.

All an epiphenomenalist needs to answer this objection is to remind that meeting the (basic) causal condition might be necessary to secure inferential knowledge, but (possibly in contrast to knowledge about one’s own qualitative states) it is not sufficient. The inferential justification is also necessary: in order for a belief to be a case of inferential knowledge, there have to be the right logical/intentional relations between this belief and its premises. In a world where the beliefs that sky is blue, that Reagan is dead and that nuts are high in fats are exemplifyings of the same physical properties as some beliefs which are premises in a valid reasoning in the actual world, the contents of these beliefs do not stand in appropriate relations between themselves and the conclusion that epiphenomenalism is true. This is the difference between this possible world and the actual world and this is what makes Charlotte’s reasoning in the actual world not only valid, but also not lucky. [CC] is still not necessary to secure inferential knowledge.

6 Conclusions

To sum up, Moore’s argument gets off ground only when supplemented by a theory of events which does not allow for a single event to be an exemplifying of both a physical and a mental property. Epiphenomenalists need not subscribe to this: if they accept a multi-property exemplification account of events and endorse EPD or a similar theory, they are not affected by Moore’s version of the epistemic argument. Further worries might arise if one accepts a stronger causal condition for knowledge which says that in order for some belief to be a case of inferential knowledge, not only this belief must be caused by its premises but this causation must obtain in virtue of these premises’ contents. However, reasons to adopt this stronger condition are scarce and I have shown them to be insufficient. An epiphenomenalist property dualist who accepts nomological psychophysical supervenience can reject the stronger condition and claim that the fact that causation between beliefs obtains only in virtue of their physical properties and not in virtue of their mental properties does not negate the possibility of inferential knowledge. Until some novel charge against EPD or a good argument for the stronger causal condition is presented, it should be assumed that type epiphenomenalism conjoined with property dualism does not have to worry about the epistemic argument.