Editors note: Because of a major snowfall in Savannah, GA during the recent Eastern APA, the Blog of the APA will be hosting papers by people who missed the opportunity to present or who would like a larger audience for their work. If you were on the schedule for the Eastern and would like to take advantage of this opportunity, please fill out the submission form.
Many philosophers have thought that to act intentionally an agent must know what she is doing. This is one of the central theses in Anscombe’s seminal Intention (1958), where she writes that when an agent does not know what she is doing, “what happens does not come under the description—execution of intentions—whose characteristics we have been investigating” (1958, §48). We can put the Aristotle-Anscombe thesis as follows:
PRACTICAL KNOWLEDGE PRINCIPLE (PKP): If an agent is Φing (intentionally and under that description), she knows that she is Φing (intentionally and under that description).
Central to Anscombe’s understanding of (PKP) is the further contention that the knowledge that agents have of their own actions is a special kind of knowledge. This knowledge is non-observational (§8) and “practical”: unlike speculative knowledge, which ‘is derived from the objects known’, practical knowledge is ‘the cause of what it understands’ (§48). Although the phrase is widely agreed to capture an important truth, its exact meaning is a disputed question even among Anscombe’s supporters, and I won’t aim to settle it here. It is important for my purposes, however, that this is meant to be genuine knowledge, and, as such, must have a non-accidental relation to the truth, just like theoretical knowledge.
Although (PKP) seems initially plausible and has significant explanatory power, it was widely regarded as mistaken for many years, in light of counterexamples presented by Davidson (2001). The status of these counterexamples, however, has recently become a point of contention among philosophers of action, as a number of scholars have argued that attention to aspectual distinctions shows that they miss the mark (Small 2012; Stathopoulos 2016; Thompson 2011; Wolfson 2012). My main goal here is to present a new challenge to (PKP), one modelled after Williamson’s anti-luminosity argument in Knowledge and Its Limits (2000). Since this argument relies on general principles about the nature of knowledge rather than on intuitions about fringe cases, the recent responses that have been given to try to defuse the force of Davidson’s purported counterexamples to (PKP) are silent against it.
I’ll divide up the discussion in two parts. In the first part I explore the reasons to hold (PKP), and argue that the standard objections in terms of presumed counterexamples leave us at an impasse. In the second part I show how versions of the anti-luminosity argument show that (PKP) and related theses in the philosophy of action should be rejected.
(PKP) and Presumed Counterexamples
(PKP) expresses a purported necessary connection between intentional action and knowledge. Related theses about other practical phenomena, like intentions, have been defended in the literature, but for now I want to focus on the case of intentional action, since it has received the most attention and support. A few remarks about the principle are in order. First, as the parenthetical phrases indicate, the description under which one knows must match the description of the action under which it is intentional. Second, it is important (for reasons that will soon become clear) that (PKP) is aspect-sensitive: it is a thesis about the actions that an agent is doing (progressive aspect), rather than those she has done (perfective aspect). Third, since it is possible to distinguish between Φing intentionally and Φing simpliciter for many action types, we might ask whether simply knowing that one is Φing simpliciter is sufficient to satisfy the knowledge requirement in (PKP). I will assume (PKP) requires knowledge of an action as done intentionally, since it would be strange to think that I could express my practical knowledge of walking, say, by saying, ‘I’m walking, but I’m not sure whether I’m doing it intentionally or not’. More importantly, there are action types that, as Anscombe notes (1958, §47), it seems impossible to instantiate non-intentionally (her examples include calling, greeting, signalling, and marrying). (PKP) purports to be fully general in nature, so it must be able to cover at least some cases where one knows one is acting intentionally as such.
(PKP) seems attractive for a number of reasons. For instance, as Anscombe notes, if you asked a man why he was sawing a plank (as he moved a saw in the relevant way across a plank) it would be weird for him to reply that he didn’t know he was sawing a plank (1958, §6); such a reply would call into question whether he was acting intentionally. If the person acting doesn’t know, then who does? An agent seems to have a special sort of authority over what she is doing that ensures she is not epistemically alienated from her actions. Moreover, recent defenders of (PKP) like Small (2012) and Wolfson (2012) have argued for (PKP) on metaphysical grounds: one can be Φing at t (whether intentionally or not) only if there is a non-accidental connection between one’s mind and the possibility that one will have Φ’d at some future time t’. In the case of intentional action, it seems such a connection can be secured only by the agent’s knowledge that what she is doing at this time is such as to lead to her having Φ’d, unless interrupted. But this knowledge is sufficient for knowledge that one is Φing intentionally, so (PKP) must hold.
Despite its apparent advantages, however, many reject (PKP). The reason is that the principle seems vulnerable to counterexamples. Here is a famous one by Davidson:
[I]n writing heavily on this page I may be intending to produce ten legible carbon copies. I do not know, or believe with any confidence, that I am succeeding. But if I am producing ten legible carbon copies, I am certainly doing it intentionally. (Davidson 2001, p.97)
Davidson is arguing here against a weaker thesis than (PKP), namely, the view that acting with an intention requires belief that one is doing what one intends. But since knowledge entails belief, the example, if successful, would show that (PKP) is false.
How should we respond to such cases? Many philosophers respond by rejecting (PKP). Among them, some, like Paul (2009a; 2009b), conclude that there is no interesting necessary connection between practical and cognitive attitudes. Many others follow Davidson in weakening the principle. For instance, Davidson held that “[a]ction does require that what the agent does is intentional under some description, and this in turn requires, I think, that what the agent does is known to him under some description [my emphasis]” (Davidson 2001, p.50). The third strategy, one that was neglected for many years but has received increasing support in recent times, is to deny that the presumed counterexamples actually tell against (PKP). Thus, a number of authors have recently argued, on the basis of renewed attention to aspectual differences, that the objection is unsuccessful (Small 2012; Stathopoulos 2016; Thompson 2011; Wolfson 2012). For instance, Michael Thompson argues that since the notion of intentional action at play in (PKP) is the notion of action in progress (as opposed to completed action) there are two possibilities to consider with respect to Davidson’s carbon copier. Either the man gets many tries to make 10 copies or only one. If he gets many tries, he can make 10 copies intentionally and knowingly despite not knowing whether he will succeed in his first try, since he can continue doing so. If he only gets one chance, as Davidson must be imagining, the case would be “like the buying of a lottery ticket”, and therefore not the phenomenon of interest to philosophy of action (p.210).
A parallel dispute in epistemology might help clarify the dialectic at this point. Many of us hold that knowledge entails belief, but there seem to be cases where we might be inclined to say that a person knows without believing something (Radford 1966; Myers-Schulz and Schwitzgebel 2013). However, these examples often seem somewhat recherche, and intuitions about them differ substantially. Hence, in light of how much there is to be said for the knowledge-belief entailment, we are entitled to regard them as fringe, and to develop a theory of knowledge that is not shaped to accommodate these intuitions. Likewise, the defender of (PKP) can say that cases like Davidson’s carbon-copier, where one only gets one chance, are fringe, and need not be accommodated by a theory of intentional action. For this reason, intuitions about fringe cases are going to be of limited help if our aim is to make progress in the debate about the nature of practical knowledge.
Practical Anti-Luminosity
At this point we could simply give up on the project of trying to reach an agreement on the status of (PKP), and accept that we have simply reached an impasse. I suggest a different strategy: instead of challenging the principle by appeal to intuitions about fringe cases, I shall argue that (PKP) conflicts with central epistemic principles. This is shown by the fact that (PKP) is called into question by considerations of the sort that Williamson (2000) has used to argue against the luminosity of mental states like appearances, meaning, and feelings of pain. (A condition is luminous just in case a subject can’t be in that condition without knowing that she is.) An additional advantage of such an argument is that, if sound, it shows that there is no interesting notion of intentional action for which (PKP) could hold, and even weaker principles like Davidson’s fail. It will thus show that the principle cannot be defended by holding that it applies to a different notion of intentional action than that subject to counterexamples.
Suppose, then, that Elizabeth decides to stage a protest at a ceremony awarding a prize to someone she knows has committed genocide. She goes to the ceremony, and loudly expresses her objections to the attendees. If anyone asked her, among the descriptions she would give of her action at this point, and under which it is intentional, would be ‘making herself heard to the attendees’. Soon after she has started protesting, however, the security guards step in and begin to draw her away from the ceremony. Eventually they lock her away in a cell several miles away from the ceremony’s location, and well beyond where she could hope to be heard by the attendees. Throughout the whole process, Elizabeth keeps protesting loudly. Let’s assume that Elizabeth has been focused on what she is doing all this time. Then, although she starts out knowing that she’s making herself heard to the audience, she gradually looses confidence that that’s the case as the guards drag her away. By the time she is in her cell, she knows pretty well that she is not making herself heard to the audience.
(Please note that nothing hangs on the choice of act-description here: a similar case with a mundane description like ‘cleaning’ (atelic) or ‘cleaning the floors’ (telic) would generate anti-luminosity arguments equally easily. A simpler case is considered below.)
Let t0, t1, . . . , tn be a series of times at one millisecond intervals from the time she starts yelling to a time when she is in the cell, and let αi be the case at time ti in this series. By the description of the case at α0 Elizabeth is making herself heard, and knows it, while at αn she isn’t making herself heard, and knows it. We can suppose that Elizabeth can’t discriminate between cases αi and αi+1 with respect to what she is doing. Hence, since knowledge requires a margin for error that ensures that the knower could not easily have been wrong, it seems the following principle applies.
Margin of Error for Action (MARA): If S knows that she is Φing at αi, she is Φing at αi+1, for all times in the series t0, t1, . . . , tn. :
From (MARA), it follows that Elizabeth is making herself heard at α1, given that she knows she is making herself heard at α0. Now suppose (PKP) is true. In that case, Elizabeth knows that she is making herself heard at α1, and again, by (MARA), that she is making herself heard at α2, and so on. Evidently, sufficient applications of this mode of reasoning (n-many) will yield the conclusion that Elizabeth knows she is making herself heard at αn, which contradicts the initial assumption that she isn’t doing that at the time. (PKP) leads to a contradiction, so it must be rejected.
(MARA) clearly plays a crucial role in the foregoing argument, so it might be worth giving at least a brief defense of the principle, while acknowledging that there is a large literature on this topic that I cannot even begin to address on this post (see Srinivasan 2013 for a proper defense.). As noted, (MARA) can be seen as an instance of the more general principle that to constitute knowledge, a belief must not easily be false: a knowledge state must be non-accidentally true. This modal property becomes salient when we consider Carl Ginet’s famous fake-barn cases (first discussed in Goldman 1976). Suppose I find myself in a field full of objects, most of which are fake barns, and only one of which is a real barn. I point to one of them and say, ‘That’s a barn’. It seems that even if I am pointing to the real barn I could not know that what’s in front of me is a barn. Here is an attractive explanation of what goes wrong: given the presence of all those fake barns nearby, in cases that are too nearby, the belief is false. A parallel explanation can be given for why (MARA) holds of Elizabeth’s case.
Now, we saw how recent defences of (PKP) rely on the fact that it is possible to be doing something, even if one has not yet completed it or even will never get to complete it. Thus, it might be claimed that in the example given, Elizabeth is still making herself heard to the attendees, even when they are many kilometers away and can’t hear her; it’s just that she’s not succeeding at achieving her goal. More generally, it might be claimed that so long as someone is trying to accomplish something, they are, in the sense of (PKP), doing it (progressive). Now, that one can be engaged in an action that one never completes is obviously true: I can be crossing the road even if I never cross it because a truck runs me over. What is utterly unbelievable, but is crucial for this response, is that one can be engaged in an action regardless of the material conditions of one’s environment: I can’t be crossing a road in a desert (regardless of my intentions) if the closest road is many kilometers away, and, similarly, Elizabeth can’t be making herself heard to her audience (regardless of her intentions) if she is so far away that it is impossible for them to hear her. Hence, even if appeal to the progressive can help account for Davidson’s carbon copier, it won’t salvage (PKP) from the problems raised by the anti-luminosity argument.
We can focus on simpler actions to not only call (PKP) into question, but even weaker principles like Davidson’s discussed above, namely:
Weak Practical Knowledge Principle (WPKP): If an agent is Φing she knows that she is Φing under at least one description of what she is doing.
To see this, imagine that a group of (trustworthy) scientists give you a drug at noon which, as they tell you, will slowly block the relevant parts of your motor system so that around three hours later you will be unable to move your fingers at all, and it will take you a day to recover mobility. Suppose you believe them. For the present study, they blindfold you and instruct you to continually move your finger as you usually would. Since the drug has not affected you at noon, when you set out to move your finger, you move it intentionally and know this, but by 5pm you don’t move it, and know this (on the basis of the scientists’ testimony), even though you seem to yourself exactly as though you were. The case so far has a similar structure to that of Elizabeth’s and it should be easy to see how the argument would go from here to show that you do not know that you’re moving your finger in this case. However, the description moving a finger seems like the most basic kind of description under which you might know what you are doing for a case like this. Understanding the principle in such a way that even more basic descriptions will count as fulfilling it risks triviality. After all, a sufficiently weak principle will apply to not just actions, but any happening whatsoever: the principle that necessarily, if x is happening a person knows that x is happening under at least one description will come out true if we allow the following description to fulfill it: something is happening.
It might be replied that there is a more basic and non-trivial description under which the person might know what she is doing in a case like this, perhaps trying to move a finger, or intending to move a finger. But anti-luminosity arguments also show that such defences fail. Suppose at noon I form the intention to go to my friend’s party. If the party were to start right there and then, I would go: I am firmly committed, and I know it. Unfortunately, the party is not until 9pm, and I am at home, with nothing to do. Thus, having formed the intention, I sit on my couch through the afternoon thinking about the party and trying to keep track of what I am intending. I am the kind of the person who, the more I reflect on something, the less I want to do it. So as I think about whether I intend to go to the party, my resolve begins to fade very slowly. By the time 9pm comes, I have lost all resolve: at that point it is clear (and clear to me) that I don’t intend to go. Let t0, t1, . . . , tn be a series of times at one millisecond intervals from noon to 9pm, such that I intend to go to the party at t0 and don’t intend to go at tn. Finally, let αi be the case at time ti in this series. By the description of the case at α0 I intend to go to the party, while at αn I don’t intend to go. Since it seems clear that I couldn’t discriminate between cases αi and αi+1 with respect to whether I intend to go, the following modified version of (MAR) seems to apply:
Margin of Error for Intention (MARI): If I know that I am intending to go to the party at αi, I intend to go at αi+1, for all times in the series t0, t1, t2, . . . , tn.
Now suppose that intention was luminous (as e.g. Fleming (1964) held): if I intend to Φ, then I know that I intend to Φ. It follows that I know that I intend to go to the party at α0, and by (MARI), that I intend to go to the party at α1; by the luminosity of intention, it would follow that I know this, so it would follow that I intend to go to the party at α2. Evidently, n applications of this mode of reasoning will yield the conclusion that I intend to go to the party at αn, which contradicts the initial assumption that I don’t. The assumption that intention is luminous leads to a contradiction, so it must be rejected.
The foregoing argument relies on the gradability of commitment. Because commitment is a feature that intentions share with the lot of practical entities that are the subject of study of philosophy of action, entities like decisions, attempts, and basic actions, it should be straightforward how to run an anti-luminosity argument for them too on the basis of the one just offered. Hence, I submit, there are no interesting necessary connections between practical entities and knowledge.
Conclusion
I want to end by noting that the arguments in this paper should not be taken to suggest that the connection between intentional action and practical knowledge is unimportant or uninteresting. After all, there are many important and interesting connections that do not require a necessary connection between entities, like the connection between perception and knowledge: there is clearly an important connection between perceptual and knowledge states, even if a person can perceive something without acquiring the relevant knowledge (in a fake-barns case, for instance, we perceive a barn in front of us without knowing that there is a barn in front of us). One way to capture the connection is by appeal to the idea that the function of perception is to acquire knowledge. Or, following Reid, we could hold that unlike other processes, perceiving can on its own (and without the need of an inference or any further reflection whatsoever) give us direct knowledge of the external world. Perhaps ideas analogous to these might hold with respect to our will, understood as the faculty in virtue of which we intend and act intentionally. I find both these ideas attractive, though, of course, much clarification and defense would be needed to support them.
Juan S. Piñeros Glasscock
Juan S. Piñeros Glasscock is Ph.D. Candidate in philosophy at Yale University. His main research interests lie in epistemology, philosophy of action, and moral psychology, although he has broader interests in other areas, like ancient philosophy. You can learn more about his work at his website, https://juanpineros0.wixsite.com/action.
