2.1. Bradford’s ‘Competent causation’

Competent Causation is Bradford’s term for the kind of relationship between agent and goal that satisfies the anti-luck condition for achievement (Bradford Reference Bradford2015, ch.3). Starting from the idea that achievement requires that the agent knows what she is doing, she develops an account of competent causation in terms of justified true beliefs (JTBs) (which she treats as a proxy for knowledge).³

Bradford begins by observing that when we are consciously causing something, we have beliefs about the causal effects of our actions. For example, when I am driving from A to B, I believe that stepping on the gas will accelerate the car, that turning the steering wheel left will move it to the left, that taking a right at the second traffic light is the quickest route to B, and so on. According to Bradford, the more of my beliefs about the causal chain leading from my actions to the result are justified and true, the more competently I am driving my car. She adds that determining the degree of competence is not merely a matter of counting justified true beliefs. Beliefs are weightier if they concern more of the causal chain. So, my belief that moving down my foot will exert pressure on the gas pedal has less weight than my belief that moving down my foot will accelerate the car which, in turn, counts for less than my belief that my actions will lead to my car ending up at B. Finally, degree of competence depends on how much there is to know about the activity in question. Bradford suggests that the degree of competence equals the weighted sum of the JTBs that the agent has about the activity, divided by the weighted sum of the possible JTBs one could have about the causal chains involved in the activity. This allows her to capture the apparent fact that one needs fewer justified true beliefs for a given level of competence when engaging in a simple activity (such as tying shoes) than when doing something more complicated – such as building a complex nanomachine (Bradford Reference Bradford2015, 65–70).

There are two problems with this account. First, it falsely implies that every justified true belief about an activity an agent may (fail to) have increases (decreases) their degree of competence. But there are many facts about an activity, knowing which has no effect on how competently the activity is carried out. Consider Bradford’s example of Rudy:

Bradford plausibly contends that Rudy is competently moving his car from A to B, but that he is not competently causing the pistons in the engine of the car to move. She goes on to claim, however, that Rudy would be a more competent driver, if he knew how the engine works (Bradford Reference Bradford2015, 66). This is implausible. Rudy has much to learn about cars and doing so would presumably make him more competent in some respects (for example, he could be a more competent mechanic, or he could talk about cars more competently). But it would not make him drive from A to B more competently. All the knowledge that is relevant to how competent a driver he is, concerns what he has to do while driving. When giving out driver’s licenses we do not award extra points for knowing how the engine works. Nor should we. A driver’s license attests that its holder can competently drive; knowledge about engines is irrelevant to that.

The example of driving also serves to illustrate the second problem with Bradford’s view: no set of justified true beliefs is sufficient for competence. Rudy's friend Ben might know everything there is to know about cars, what one has to do when driving, how to react to unforeseen events on the road, and the route from A to B. If Ben has never been in a car before, no amount of such knowledge will make him a competent driver. As Gilbert Ryle puts it:

Competently reaching a goal is not a matter of knowing many truths about how to reach the goal. Of course, the two are strongly correlated; generally, somebody who knows much about an activity is more likely to carry it out competently than someone who knows little about it. But, as the examples of Rudy and Ben show, competence and knowledge can come apart.⁴

2.2. Hurka’s ‘Justified optimism’

Part of the problem with Bradford’s proposal is that knowing many of the details of a process is irrelevant to an agent’s competence. Hurka’s proposal for an anti-luck condition avoids this problem. Hurka suggests that, to fulfil the anti-luck condition, ‘[w]e must accompany each of our goals by a justified belief that we will achieve it.’ (Hurka Reference Hurka1993, 106). This account handles the cases of Rudy and Ben with ease. Rudy, but not Ben, is justified in believing that he will safely drive from A to B.

One may worry about (the practical analogue of) Gettier-type cases. What if an archer with a justified belief that she will hit the target is startled by a strange noise when releasing the arrow which shoots up in the air at an unintended angle, but hits the target nevertheless, because of an unexpected gust of wind?⁵ This would satisfy Hurka’s anti-luck condition but is no more an achievement than Gettier’s classic examples instantiate knowledge (Gettier Reference Gettier1963). But we should not assign too much weight to such worries.⁶ It is fair to assume that Hurka, just like Bradford, simply uses JTB as a stand-in for knowledge. It is worth making this explicit though, because, once formulated in terms of knowledge, Hurka’s anti-luck condition seems a lot less intuitive. It sounds strange to say that reaching a goal is a lucky success (not an achievement), unless the agent knew in advance that she would succeed.

To bring this point home, consider that Hurka’s anti-luck condition rules out achievements by pessimists.

Karl clearly achieved his goal. But he fails Hurka’s anti-luck condition, because he did not know that he would succeed. That is not because he lacked the ability or the competence to succeed, but simply because he did not think he would. But whether I am optimistic or pessimistic about my chances of success should have no bearing on whether my success is an achievement or not.⁷

Hurka might reply that Karl would be justified in believing that he will succeed, if he were to form such a belief; and that is all that is required. There are two problems with this suggestion. First, consider Karl’s cousin Carla, who is just as able, enthusiastic, and wealthy as Karl, but not generally pessimistic. Unlike Karl she has excellent (though misleading) evidence that she is a poor philosopher, because she has consistently received negative feedback from senior philosophers. If Carla persists and gets her PhD, it would be an achievement, but she would not have been justified in believing that she would succeed.

The second problem is that we need to know what justifies (or would justify) Karl’s optimistic belief. Hurka has a coherence model of justification (Hurka Reference Hurka1993, 106), but it might well be inconsistent for Karl to think that he will have success in graduate school (not to mention Carla). The problem is not Hurka’s internalist picture of epistemic justification. No formal account of justification could answer to our purpose.⁸ What we need, instead, is a substantive answer to the question what justifies people in thinking they will be successful. If we wanted to stay in the framework of an epistemic anti-luck condition, it would be natural to look to things the agent knows about how to achieve the goal. Karl, while lacking the belief that he will succeed, knows a lot about how to get a PhD. So maybe we should just specify how much the agent has to know about how to reach the goal in order to count as achieving. But we have been down this road before – this was Bradford’s suggestion – and found it a dead end. It must be something else that justifies the belief that one will reach one’s goals.

The solution is simple: what usually justifies agents in believing they will reach a goal is evidence that they can reach it competently. Of course, what exactly that means is what we are trying to figure out. Nevertheless, we can already see the fundamental problem with epistemic accounts of the anti-luck condition: these accounts get things backwards. While it is correct that Rudy has a JTB that he will successfully drive from A to B, this does not explain why he acts competently when he does. Rather, that he can competently drive from A to B explains why his belief is justified. Similarly, Karl’s competence in obtaining a PhD is not constituted by the fact that ‘were he to form an optimistic belief, it would be justified’. Rather, such a belief would be justified, because he is competently pursuing his PhD.

Having a JTB that one will succeed is a somewhat reliable proxy for being competent because, ordinarily, being competent justifies such a belief. That is why Hurka’s account delivers the right verdict in many cases (such as Rudy’s and Ben’s). It fails to deliver the correct results when being competent does not lead the agent to form an optimistic belief (as in the cases of Karl and Carla), or where an agent has a justified belief despite being lucky in reaching the goal (as in practical Gettier cases). But even when the account gets the right result, it does so for the wrong reasons. It treats what is a somewhat reliable proxy for competence as a constitutive account of competence. I conclude that the anti-luck condition for achievements should not be spelled out in terms of JTBs or knowledge.

3.1. Achievement in the epistemology literature

The notion that a success is an achievement only if it is properly attributable to the agent’s abilities has been developed in the context of an epistemological research program that attempts to conceptualize knowledge as a form of achievement. This program has been variously termed ‘robust virtue epistemology’, ‘performance epistemology’, or simply ‘AA’ (for ‘achievement account of knowledge’), and the seminal figure in this context is Ernest Sosa. Sosa laid out the basic model of the knowledge as achievement program in his 2007 monograph. In the ensuing decade, him and many others have proposed more or less friendly amendments, but the general model remains the same (see e.g. Turri Reference Turri2011; McKinnon Reference McKinnon2013; Littlejohn Reference Littlejohn2014; McKinnon Reference McKinnon2014; Carter, Jarvis, and Rubin Reference Carter, Jarvis and Rubin2015; Sosa Reference Sosa2015; Turri Reference Turri and Fernandez2016). According to Sosa, achievements have what he calls a ‘AAA structure’. For a performance to constitute an achievement it must be accurate (i.e. achieve its goal), adroit (i.e. carried out competently), and apt (i.e. the success must be properly attributable to the fact that the performance was carried out competently) (Sosa Reference Sosa2007, 22). Sosa understands full competence (or adroitness), in turn, as having a ‘SSS structure’ consisting of seat, shape, and situation. Thus, a performance is competent (or adroit), if the agent is generally constituted in such a way as to be disposed to succeed if they try under the right circumstances (seat), the agent is in the proper condition to perform, i.e. sober, awake etc. (shape), and the circumstances in which the agent attempts to exercise their ability are generally favourable (situation) (Sosa Reference Sosa2015, 95-6).

Among the many refinements proposed of this general view, John Greco’s work contains one of the most developed accounts (Greco Reference Greco2010) as well as extensive criticism of many others (Greco Reference Greco2012). I will focus my discussion on Greco here partly because of that, and partly because his exposition lacks two features of Sosa’s most recent contribution that are potentially confusing in the current context.¹⁰ First, like many virtue-epistemologists (most prominently Turri Reference Turri2011), Sosa has taken to making extensive use of the notion of manifestation as a primitive (i.e. unanalyzed) metaphysical concept (e.g. Sosa Reference Sosa2010, 466). To unfriendly observers, this may appear as a case of renaming the problem posed by certain types of Gettier cases instead of solving it. The original problem is that competent performance and success need to be related ‘in the right way’, where it turns out to be difficult to define what that comes to. Since ‘manifestation’ is unanalyzed, the manifestation solution appears to simply replace the mysterious ‘related in the right way’ with the equally mysterious ‘manifests competence’. While I have no doubt that friends of manifestation have answers to this worry, I find it preferable to avoid the issue by focussing on Greco’s account that does not use the notion of manifestation. Second, when Sosa offers a theory of competence in chapter 4 of his most recent book, he is concerned with something slightly different from my current interest. Sosa’s theory of competence is a theory about what it takes for someone to be competent at X (e.g. be a competent tennis player). By contrast, my interest is only in whether a particular performance has been competent in the sense that it could potentially count as an achievement. These two evaluations (of an agent, and of an action) are likely to be related, but they are not identical. Thus, Sosa’s ‘Theory of Competence’ (Sosa Reference Sosa2015, 95–106) is not a direct competitor to my own account of succeeding competently. Focusing on Greco’s account will thus avoid terminological confusion.¹¹

3.2. Greco’s ‘Success because of ability’

For Greco, what distinguishes achievements from lucky successes is that an achievement is a success because of ability.¹² This requires some explanation. We need to know, first, what Greco means by ‘ability’ and, second, what he means by ‘because of’.

Take the second point first. When he says that achievements are successes because of ability, Greco says that ‘[t]he term “because” is here intended to mark a causal explanation.’ (Greco Reference Greco2010, 71) In other words, a success is an achievement, if and only if, the agent having an ability explains why she was successful (Greco Reference Greco2010, 74). This seems to handle cases like Able Pessimist and Competent Driver. Karl’s philosophical abilities explain his success in pursuing a PhD, and Rudy’s ability to drive explains why he reaches B safely, and so both are correctly classified as achievements. The account also looks promising when considering lottery winners, as their abilities are not what explains their success. Of course, these assessments are preliminary until we know more about abilities. In addition, however, we need to know what counts as an explanation.

Greco says that an account of (good) causal explanations would provide criteria for singling out one of the partial causes leading to an event as salient. While not providing a full account of explanatory salience, he considers two elements of any such account uncontroversial: salience is sensitive to (a) our purposes and interests and (b) what is normal or usual (Greco Reference Greco2010, 74-5). However, while this kind of context-sensitivity is probably part of the best account of causal explanations, we should be wary of introducing it into an account of achievements. Greco’s resulting account of knowledge (as cognitive achievement) is a version of attributor contextualism, i.e. the view that whether a knowledge ascription is true depends partly on the circumstances of the person who makes it (Pritchard Reference Pritchard2002). For example, I might be asserting truly that Charlemagne knew such and such, while a historian making the same assertion at a conference would be speaking falsely.

Whatever the merits of attributor contextualism as a theory about the propriety of knowledge ascriptions, this kind of context-sensitivity has no place in an account of what achievements are. Whether Charlemagne’s conquest of Saxony was an achievement, for example, does not depend on the context in which we talk about the question. This should be particularly clear, if we think (as Greco does) that achievements have intrinsic value (or, as he would put it, ‘final value’ see end note 2). Since events can be discussed in different contexts, but an event cannot both have and lack intrinsic value, whether an event is intrinsically valuable cannot depend on the context in which it is discussed.¹³ Thus, if Greco is right that causal explanations are context-sensitive, we do best not to wed our account of achievements to them.¹⁴

Maybe the spirit of Greco’s account could be retained by focusing on how much of a causal role the agent’s ability played in bringing about her success (rather than on whether that role ensured explanatory salience). But now we encounter a second problem with Greco’s account: it is doubtful whether what he calls abilities have any causal power. Greco understands abilities to be a dispositional property.¹⁵ He thinks that someone has an ability to do X, if and only if were this person to try to do X, they would be likely to succeed.¹⁶ By that Greco means that having the ability to do X means succeeding when trying to do X in many or most of a set of a relevantly close possible worlds (Greco Reference Greco2010, 77).

In deciding which possible worlds are relevantly close, Greco employs a two-pronged strategy. First, he relies on an intuitive notion of closeness. Second, the nature of an ability helps to fix the set of relevant possible worlds. Greco argues that abilities are always tied to some set of environments and conditions (this closely mirrors Sosa’s discussion of competence in terms of the SSS account). When we say that Derek Jeter had the ability to hit fastballs, for example, we understand that this ability was tied to a set of favourable environments and conditions. We do not mean that he could hit fastballs with sand in his eyes, or in an active war zone (Greco Reference Greco2010, 76-7). No matter how close these possible worlds might be, they are not part of the relevant set, because the ability is implicitly indexed to exclude such worlds. Greco concludes that abilities have the following structure.

But how could abilities, thus understood, cause (or causally explain) anything? Without getting into questions about the nature of causation, let us assume that causing an event is either constituted or closely accompanied by making that event likely. Clearly, that I am likely to succeed cannot be what makes me likely to succeed. But this is what saying that my ability (in Greco’s sense) causes my success would come to.

What seems right is that, for my success to be an achievement, I need to do things that make it likely that I will succeed. This contrasts with Greco’s picture on which I have to be such that I am likely to succeed. This leads to another point. To say that I have an ability (in Greco’s sense) is akin to saying that I am likely to act in ways that make my success likely. To have an ability, then, is to have a somewhat stable disposition to be successful (cf. McKinnon Reference McKinnon2013). But why should it be part of the anti-luck condition for achievement that I have such a stable disposition? Consider

Ricky’s success is an achievement.¹⁷ But he does not succeed in most nearby possible worlds. In most of them, he cannot find the strike zone, serves up a base hit, or uncorks a wild pitch. Thus, (according to Greco’s notion) he could not have succeeded from ability.¹⁸

4.1. Making success likely

While Greco’s model of achievement as ‘success because of ability’ is flawed, he is right that the anti-luck condition for achievements has to do with the relationship between the agent and the likelihood of success.¹⁹ He goes wrong when he locates the required source of likelihood in the agent’s dispositional makeup. The agent does not need to be such that success is likely, she has to act such that success is likely. Thus, a complete conception of the anti-luck condition requires a theory of action. As defending such a theory cannot be done here, I will help myself to the notion of action and describe the role to be played by it.

Consider Pritchard’s necessary condition for an event to count as lucky.

Combined with the claim that achievement excludes luck, (L1) correctly classifies many cases as non-achievements. Consider, for example, young Jerry’s win of the track-race in the Seinfeld episode The Race. Jerry wins, because nobody (but George) realizes when he accidentally starts 2 seconds early. In most nearby possible worlds somebody would have noticed Jerry’s early start. Thus, his winning the race is a lucky event according to (L1).

Pritchard argues that the vagueness of the locution ‘relevant initial conditions’ in (L1) is unproblematic. This is partly, because he thinks that we have a good enough intuitive grasp of what the relevant initial conditions are (Pritchard Reference Pritchard2005, 131-2). Moreover, he thinks that the vagueness captures a corresponding vagueness in our everyday concept of luck (Pritchard Reference Pritchard2005, 143 n.12). In the current context, however, we should at least specify the following: the relevant initial conditions include the actions taken by the agent. Here, then, is a first pass at an anti-luck condition.

Anti-Luck 1 will not do, for I could make one of my goals likely through actions that I am performing for unrelated reasons. Consider

Anti-Luck 1 implausibly implies that Tony achieved his goal of keeping Kyle’s business solvent. We need some restriction on which of an agent’s actions count as part of the relevant initial conditions. The natural thing to say is that one can pursue a goal competently, only if one pursues it at all.²⁰ And while Tony has the goal of keeping Kyle’s business afloat, he is not pursuing that goal when placing his order. However, we should not use too strong a notion of pursuit here. Consider, for example, the suggestion that only those actions count that are performed with the primary purpose of reaching the goal. This would be too strong. Suppose I want to lose some weight; but I also often oversleep and – having recently moved further from my office – find myself having to either run to work or take a cab to avoid being late. I choose to run, realizing that running would also help with my goal of weight loss. If I lose weight as a result, it seems that I have achieved my goal. But my primary goal in running was to arrive in time. It is not necessary, then, that my actions are performed primarily in order to reach the goal. What is necessary (and what distinguishes this case from Fresh Money) is that I have to be aware that my actions are making my goal more likely.²¹ This leads to

4.2. Practical Gettier cases

Even if Anti-Luck 2 passes muster as an account of competent pursuit, it is not enough that an agent competently pursues and reaches her goal. She must reach it through her competent pursuit. The former does not guarantee the latter. Cases that lack the right connection between competent pursuit and success are the practical analogue of Gettier cases. Here is one.

In Young Millionaire Oliver competently pursues his goal (as correctly diagnosed by Anti-Luck 2) and reaches it, but he does not reach it through his competent pursuit. There is a deviant causal chain from his actions to his success.²³ At this point, we might be tempted to return to an epistemic view like Bradford’s requiring that an agent needs to know (most of) the causal chain leading from her actions to success. This could handle Young Millionaire, as Oliver’s beliefs about how his actions are going to lead to his goal are false. But, as we have seen, Bradford’s view excludes too much. People can achieve their goals without much knowledge about the causal chains leading from their actions to success.

Alternatively, we can try to fine-tune the modal modelling to exclude deviant causal chains. Consider

This also handles Young Millionaire. While Oliver’s success is likely at the outset, it becomes unlikely once his business is being outlawed. One might worry that Anti-Luck 3 is too restrictive. In many cases, some time passes between someone’s competent action and the resulting success. What if, during that interval, it became very likely (even though it was not so at the time of the action) that something interfered with the normal causal chain but then (luckily) this danger passes? In such cases of ‘bad-luck-almost’ we should still want to ascribe achievement. This problem is merely apparent, however, for the ‘relevant initial conditions’ clause takes care of such cases. There is nothing in this clause (or the motivation for it) that requires us to interpret ‘initial’ exclusively temporally. So, if, between action and success, nuclear holocaust becomes overwhelmingly likely but is then narrowly avoided, we might say that success remained likely throughout, because it obtained in most possible worlds with the same relevant initial conditions, the last clause excluding all worlds featuring nuclear holocaust.

But Anti-Luck 3 is still too permissive. In Young Millionaire, we could stipulate that, when the new government is elected, this has two simultaneous effects: it makes the success of his business unlikely and it makes the sudden death of his uncle very likely. In this case, Anti-Luck 3 would be satisfied, but Oliver’s success would not plausibly be an achievement. It is tempting to think that Anti-Luck 3 is in trouble whenever there is an event that reduces the likelihood of the most likely causal chain, while simultaneously increasing the likelihood of a different causal chain leading to success. But this generalization is too quick, as the case of Lloyd shows.

Both Lloyd and Oliver reach their goals as the result of a causal chain that was not likely to occur at the outset. But Lloyd’s profits are an achievement whereas Oliver’s riches are not. Thus, Oliver’s claim to achievement cannot be undermined simply by the prior unlikelihood of the actual causal chain. Instead, it seems the problem is that the actual causal chain was not accounted for in Oliver’s plans. To see this more clearly consider

Felix’s actions made success likely, success remained likely, and it was likely to come about via the actual causal chain right from the start, even though Felix had a different causal chain in mind. If we do not want to credit Felix with an achievement, we need to revise Anti-Luck 3. But since we do not want to exclude cases like Correlation Trade it is not obvious how to do this.

4.3. An epistemic enabling condition

A third response to the Young Millionaire scenarios is to take another look at epistemic anti-luck conditions. While such proposals by themselves are inadequate, they might prove useful when combined with a condition based around modal safety. In section 2.1, I argued (1) that no set of JTBs about the causal chain from action to success suffices to rule out luck, and (2) that some JTBs do not contribute to the elimination of luck. The lack of sufficiency is unproblematic in the current context as we are not looking for an epistemic condition that rules out luck by itself. But the second point is still pertinent. The epistemic part of the anti-luck condition must not assign significance to (the absence of) irrelevant JTBs. Recall Competent Driver. Rudy’s false beliefs about how cars work, do not undermine his driving competence. Thus, we must be careful to formulate our epistemic condition such that Rudy is not subject to the same criticism as Oliver and Felix: that, even though his actions made success likely, he did not bring about success in the way he intended.

I submit that the crucial difference between Rudy on the one hand, and Felix and Oliver on the other is that correcting Rudy’s beliefs would not provide him with a reason to change his plans, but the same cannot be said of Felix and Oliver. In Young Millionaire, there is an event that Oliver does not foresee (the prohibition of porn), and had he foreseen it, he should not have acted as he did. Similarly, had Felix foreseen the military coup, he should have changed plans. Of course, had he foreseen both the military coup and the later government handout, he might have acted as he did; but in that case, his situation would be akin to Correlation Trade and we should have no qualms calling his success an achievement (the same goes for Oliver foreseeing Tobi’s death). The problem with Bradford’s proposal is that it does not discriminate between (the absence of) JTBs that make a difference to the way the agent should act in pursuing his goal and those that do not. But only the former can plausibly play a role in an anti-luck condition for achievements. I propose

Anti-Luck 4 combines a modal safety condition with an epistemic enabling condition that assigns significance only to false beliefs whose correction would make a difference to the quality of the agent’s plans.²⁷

Anti-Luck 4 has been developed in response to rather convoluted scenarios, but it is not just a gerrymandered proposal designed to handle these cases. Instead, it expresses two simple and compelling ideas about how to distinguish achievement from lucky success. First, the actions taken must make success likely (clause 1); second, that (clause (1)) is why the agents performs these actions (clause 2).