Testing Sripada’s Deep Self Model Florian Cova, Hichem Naar* * This research was supported in part by a Grant from the French Agence Nationale de la Recherche (ANR) (ANR Blanche: SoCoDev). We also thank Chandra Sripada, Edouard Machery and two anonymous referees of Philosophical Psychology for their helpful comments. Correspondence: Florian Cova, Institut Jean Nicod, Ecole Normale Supérieure, 29 rue d’Ulm, 75005, Paris, FRANCE Email: florian.cova@gmail.com Florian Cova Institut Jean Nicod, Ecole Normale Supérieure, 29 rue d’Ulm, 75005, Paris, FRANCE florian.cova@gmail.com Hichem Naar Department of Philosophy, University of Manchester, Manchester, United Kingdom hm.naar@gmail.com Abstract Sripada has recently advanced a new account for asymmetries that have been uncovered in folk judgments of intentionality: the “Deep Self model”, according to which an action is more likely to be judged as intentional if it matches the agent’s central and stable attitudes and values (i.e., the agent’s Deep Self). In this paper, we present new experiments that challenge this model in two ways: first, we show that the Deep Self model makes predictions that are falsified, then we present cases that it cannot account for. Finally, we discuss how the Deep Self model could be modified to accommodate these new data. Word count: 5407 1. Introduction Recent studies, pioneered by Joshua Knobe (2003a, 2003b, 2006), have suggested that commonsense judgments about whether an action is intentional do not depend only on causal and psychological considerations but are also shaped by evaluative (and, more precisely, moral) evaluations. This claim is supported by the following puzzling phenomenon that has since been dubbed the “Knobe Effect”: people seem to be more likely to judge a side effect as intentional if they consider it as morally wrong than if they see it as morally right. Consider the following scenario: Harm Case: The vice-president of a company went to the chairman of the board and said, “We are thinking of starting a new program. It will help us increase profits, but it will also harm the environment.” The chairman of the board answered, “I don’t care at all about harming the environment. I just want to make as much profit as I can. Let’s start the new program.” They started the new program. Sure enough, the environment was harmed. In this case, 82% of participants considered that the chairman intentionally harmed the environment (Knobe, 2003a). Now, consider the following scenario: Help Case: The vice-president of a company went to the chairman of the board and said, “We are thinking of starting a new program. It will help us increase profits, and it will also help the environment.” The chairman of the board answered, “I don’t care at all about helping the environment. I just want to make as much profit as I can. Let’s start the new program.” They started the new program. Sure enough, the environment was helped. In this case, only 23% of participants answered that the chairman intentionally helped the environment. But there seems to be no difference between these two scenarios in causal features or the chairman’s psychological attitudes. The only difference seems to be evaluative: “harming the environment” is a bad outcome while “helping the environment” is a good outcome and most people tend to say that the chairman in the Harm case deserves blame whereas very few consider that the chairman in the Help case deserves praise. This led some researchers, including Knobe (2006) and Nadelhoffer (2004a, 2004b, 2006), to argue that this asymmetry in ascriptions of intentionality is driven by these evaluative considerations (the bad / good distinction for Knobe and the blame / no praise asymmetry for Nadelhoffer). In a recent article, Sripada (forthcoming; see also Sripda and Konrath, forthcoming) argues against the view that evaluative judgments exert an influence on our ascriptions of intentionality. Rather, he offers an alternative explanation for the asymmetry we described. In this paper, after presenting his account, we argue that it fails on two points: (i) it generates predictions that do not hold and (ii) there are cases that it cannot accommodate. 2. Sripada’s Deep Self model 2.1. Unidirectional and Bidirectional accounts of the Knobe Effect In his paper, Sripada distinguishes two kinds of competing accounts for the Knobe Effect: the accounts based on what he calls the “Unidirectional Thesis”, and those based on what he calls the “Bidirectional Thesis”. According to the Unidirectional Thesis, “intentionality judgments are completed prior to and/or independently of normative judgments” (see for example Machery, 2008, Wright and Bengson, 2009, Nanay, 2010). On this picture, contra Knobe and Nadelhoffer, evaluative considerations do not play any role in people’s judgments of intentionality so that it is always possible to tell if an agent acted intentionality regardless of the moral status of the action in question. According to the Bidirectional Thesis, by contrast, “intentionality judgments and normative judgments are interdependent – intentionality judgments both influence and are influenced by normative evaluations”. From the asymmetry found above, we might be tempted, like Knobe, to accept the Bidirectional Thesis. Indeed, if the only difference between the scenarios is a difference in (the side effect or the agent’s) moral status, it seems quite plausible that evaluative considerations can fully account for the difference in judgments of intentionality across the two conditions. Sripada is skeptical, and puts forward a new theory that would account not only for the asymmetry found above, but also for the variety of cases that are found in the literature for which Knobe’s account may not be fully satisfying. Inspired by Hume, Sripada puts forward the Deep Self Model, according to which an agent’s action is intentional to the extent that it originates in features of the agent’s psychology that are stable and constant, as opposed to fleeting and determined by factors beyond the agent’s psychology. For instance, an agent acting in accordance with her strong-held beliefs, desires, and values will clearly act intentionally, while an agent acting ‘out of character’ is less likely to be considered as acting intentionally. 2.2. Sripada’s own account Sripada distinguishes between two aspects of an agent’s psychology, the “Acting Self” and the “Deep Self”, defined in the following way: “we can understand the distinction roughly along Humean lines, i.e. the Acting Self contains the relatively ‘temporary and perishing’ beliefs and desires that are the immediate causal source of the action, while the Deep Self contains the agent’s relatively stable values, principles and other more fundamental attitudes.” The more an agent acts in accordance with her Deep Self, the more likely her action is to be seen as having been performed intentionally. To capture this observation, Sripada proposes the following criterion: Concordance criterion: Does the outcome concord with the psychological attitudes of the agent’s Deep Self? As the vignettes used in the experiments are underdetermined when it comes to the psychological features of the agent, Sripada argues that it is highly likely that people do not just stick with the description provided, but rather make additional inferences, which he calls ‘deep attitude attributions’, about the agent’s deeper psychological attitudes. Applied to the Harm case, Sripada’s analysis is telling: the Chairman states ‘I don’t care at all about harming the environment. I just want to make as much profit as I can.’ Based on this statement, it’s plausible that subjects make attributions that the Chairman possesses deep attitudes with contents such as ‘profit is more important than the environment’, ‘the environment is not worth helping or preserving’, and even perhaps ‘helping myself is more important than preventing harms to others’, and other attitudes as well. These inferred deep attitudes then interact by means of the Concordance Criterion described above to yield judgments about intentionality (…). Alternatively, in the Help case, the chairman is less likely to be considered as having intentionally helped the environment for the simple reason that he doesn’t care about helping the environment, and therefore doesn’t appear to have any enduring attitude in favor of the well being of the environment. To test this hypothesis, Sripada presented to subjects the Harm or the Help case and asked them to “rate the chairman’s values and attitudes with regard to the environment” on a 7 point scale (1 = Anti-environment and 7 = Pro-environment). The results showed that, for both cases, subjects rated the chairman as anti-environment (mean 1.9 for the Harm case and 2.7 for the Help case). These results seem to confirm Sripada’s hypothesis: subject consider an action as intentional when the outcome “concords” with the agent’s values and attitudes (the Harm case) and as non-intentional when the outcome goes against these same values and attitudes (the Help case) Sripada and Konrath (forthcoming) have also tried to argue for the Deep Self model using Structural Equation Modeling techniques. We won’t discuss these here but a critique can be found in Rose et al. (forthcoming).. 2.3. Testing Sripada’s hypothesis Sripada’s aim is not just to explain the particular asymmetry between the Harm and the Help cases. His goal is to present a general account of our ascriptions of intentionality. Can his Deep Self model fulfill this role? In the two next sections, we propose to challenge this model in two different ways: First, we argue that the Deep Self model makes predictions that can be falsified. Second, we propose cases that the Deep Self model cannot accommodate. 3. Testing the Deep Self model’s predictions The Harm case is a case in which an anti-environment agent harms the environment while the Help case is a case in which an anti-environment agent helps the environment. There are two more combinations left (see Table 1): A case in which a pro-environment agent helps the environment. In this case, the agent’s Deep Self and the outcome concord and the Deep Self model should predict that this agent intentionally helps the environment. A case in which the pro-environment agent harms the environment. In this case, the agent’s Deep Self and the outcome do not concord and the Deep Self model should predict that this agent doesn’t intentionally help the environment. We designed two scenarios to test these two predictions. 3.1. First prediction: pro-environment agent helping the environment According to the Deep Self model, the reason why most people consider helping the environment as unintentional in the Help case is that the chairman’s Deep Self doesn’t match the side effect: the chairman doesn’t care about the environment. If the chairman genuinely cared about the environment, then we would consider his helping the environment as intentional. Wible (2009) has run an experiment using a case similar to the Help case, but in which the chairman was described as caring about the environment. He gave to 36 participants the following case: Nice Chairman: The vice-president of a company went to the chairman of the board and said, ‘We are thinking of starting a new program. It will increase profits, and it will also help the environment.’ The chairman of the board answered, ‘Great! I care about helping the environment. I am happy that we can help the environment and make a profit at the same time. Let’s start the new program.’ They started the new program. Sure enough, the environment was helped. In this case, 55% of participants thought that the chairman intentionally helped the environment, which is higher than in the original Help case. Nevertheless, 55% is not that high. One possibility is that, in this case, most participants don’t really believe that the chairman genuinely cares about the environment but is happy because he thinks that helping the environment will improve his reputation. To address this concern, we designed the following scenario: Very Nice Chairman: The vice-president of a company goes to the chairman of the board and says: ‘You asked us to imagine new programs that would enable us to make more money. We can propose two programs. Program A will enable us to make a lot of money for a very small cost. Program B will generate as much money as Program A for the same cost, but will have the supplementary effect to help the environment. Nevertheless, it will be impossible to prove that it is our action that helped the environment, and that won’t help our reputation. What program do you want us to start?’ The chairman of the board answers: ‘Let’s start Program B.’ Program B is started and the environment is helped. After the scenario, two questions were asked. The first question was: ‘Did the chairman intentionally help the environment? Answer YES or NO and justify’. The second question was: ‘Helping the environment was: 1) the chairman’s goal, 2) a means to achieve his goal OR 3) a side effect of his action?’ The second question was designed to make sure people considered helping the environment as a side effect and not as a means or, more probably, his goal (since he seems to genuinely care about helping the environment). The experiment was run in French. Among 25 participants, 20 (80%) answered “YES” to the intentionality question. To the second question, 20 (80%) answered that helping the environment was a side effect and 5 (20%) answered that it was a means. Among the 20 participants who answered that helping the environment was a side effect, 16 (80%) answered that the chairman intentionally helped the environment. These results confirm the first prediction of the Deep Self model: a side effect (and even a morally good side effect) is judged as intentional when the agent genuinely cares about bringing out this effect. 3.2. Second prediction: pro-environment agent harming the environment The Deep Self also predicts that a side effect will be considered as unintentional if it doesn’t match the agent’s “deep self”. According to Sripada, this is what happens in the Help case. If this is true, then a modified Harm case in which the agent doesn’t want to harm the environment and even cares about the environment would lead people to judge the chairman harming the environment as unintentional. To test this prediction, we designed the following scenario: Reluctant Chairman: The vice-president of a company went to the chairman of the board and said, “We are thinking of starting a new program. It will help us increase profits, but it will also harm the environment.” The chairman of the board answered, “That’s annoying. I’d rather not harm the environment, if it is possible. I heard that you had a back-up program that wouldn’t harm the environment though it would generate much less profits. I think it would be better to start this program instead.” The vice-president answered, “I know your concern for the environment, and I know you have fought for its defense. However, starting this back-up program would not reduce our profits by two but by four.” The chairman answered, “This is indeed a considerable loss. That’s sad but start the first program.” They started the new program. Sure enough, the environment was harmed. We gave this case to 25 participants (recruited by the Laboratoire de Sciences Cognitives and Psycholinguistique in Paris). After reading the scenario, they were asked to answer the three following questions: Did the chairman intentionally harm the environment? (“YES” or “NO”) What is the chairman’s attitude towards the environment? Is he rather anti-environment or pro-environment? (on a scale from -3 = “anti-environment” to 3 = “pro-environment”) Did the chairman reluctantly harm the environment? (on a scale from 0 = “not at all” to 7 = “absolutely”) Questions (ii) and (iii) allow us to measure the concordance between the outcome and the agent’s Deep Self. The mean answer to question (ii) was 0.64, a score that differs significantly from the 0, the midpoint (t(24) = 3.2, p<.01**) (16 subjects out of 25 gave an answer superior to 0 and only 3 gave an answer inferior to 0). The mean answer to question (iii) was 4.04, a score that doesn’t differ significantly from 3.5, the midpoint (t(24) = 1.3, p=.20) (16 subjects out of 25 gave an answer superior or equal to 4). As a result, participants seemed to situate the chairman somewhere between indifference and benevolence towards the environment. Since both these attitudes do not concord with the outcome (harming the environment), the Deep Self model should predict that most subjects would judge that the chairman did not intentionally harm the environment. However, regarding question (i), 76% of subjects answered that the chairman intentionally harmed the environment. If we focus on the 16 subjects who gave an answer higher to 0 in question (ii) (and thus judged the chairman to be pro-environment), we found that 14 (88%) answered that the chairman intentionally harmed the environment. These results spell trouble for Sripada’s theory, because they do not match its predictions (see Table 1). Therefore, it seems that the Deep Self model needs at least some modifications. Case Agent’s Deep Self Outcome Concordance Intentionality judgments predicted by the Deep Self model Intentionality judgments obtained in experiments Harm Anti-environment Harming the environment YES Intentional Intentional Help Anti-environment Helping the environment NO Non-intentional Non-intentional Very Nice Chairman Pro-environment Helping the environment YES Intentional Intentional Reluctant Chairman Pro-environment / Indifferent Harming the environment NO Non-intentional Intentional Table 1. Deep Self model’s predictions and actual results for the different Chairman cases. 4. Cases that the Deep Self model cannot accommodate Our second criticism of the Deep Self model will be that there are cases of asymmetry similar to the one found between the Harm and the Help cases that it cannot accommodate. Consider the following pair of scenarios, inspired from Nanay (2010): Apple Tree, Version 1: A company has decided to expand its building. The vice-president of the company has been assigned the task to prepare the building’s new plans. Once the plans finished, the vice-president goes to submit them to the chairman of the board. On his way, he thinks, “The chairman will surely be happy. Expanding our building will increase our profits. However, to start the expansion, it will be necessary to cut down the apple tree in front of the chairman’s window. The chairman has always enjoyed its view from his office’s window, mainly because, as a child, he spent a lot of time playing on that tree. That’s what he told me many times. The vice-president submits the plans to the chairman: “Expanding the building will help us increase profits. However, to expand our building, we will need to cut down the apple tree that is in front of your office.” The chairman answers, “So what? Although, as a child, I spent a lot of time playing in that apple tree and I enjoyed its view from my office, I don’t care at all about its being cut down. All I care about is making profits. Let’s start the expansion.” They started the expansion and the apple tree was removed. Apple Tree, Version 2: A company has decided to expand its building. The vice-president of the company has been assigned the task to prepare the building’s new plans. Once the plans finished, the vice-president goes to submit them to the chairman of the board. On his way, he thinks, “The chairman will surely be happy. Expanding our building will increase our profits. Moreover, to start the expansion, it will be necessary to cut down the apple tree in front of the chairman’s window. The chairman has always hated this tree that has annoyed ever since he moved into this office. That’s what he told me many times. The vice-president submits the plans to the chairman: “Expanding the building will help us increase profits. Moreover, to expand our building, we will need to cut down the apple tree that is in front of your office.” The chairman answers, “So what? Although that apple tree has annoyed me ever since I moved into this office, I don’t care at all about its being cut down. All I care about is making profits. Start the expansion.” They started the expansion and the apple tree was removed. We gave these cases to 70 subjects recruited on the streets of the Quartier Latin (Paris). Each subject received only one of the two scenarios, read it then had to answer the five following questions: Did the chairman intentionally have the apple tree cut down? (on a 7-point scale ranging from -3 = “NO” to 3 = “YES”) How much did the chairman want to have the apple tree cut down? (on a 7-point scale ranging from -3 = “he didn’t want to” to 3 = “he really wanted to”, 0 being “he didn’t care”) Do you think the chairman reluctantly had the apple tree cut down? (on a 7-point scale ranging from 1 = “not at all” to 7 = “absolutely”) Do you think the chairman gave a great importance to the apple tree? (on a 7-point scale ranging from 1 = “not at all” to 7 = “absolutely”) Is the chairman responsible for the apple tree’s removal? (on a 7-point scale ranging from 1 = “not at all” to 7 = “absolutely”) Results for question (v) are discussed in the Appendix. There is an asymmetry between these two scenarios. To question (i), the mean answer was 1.28 for the first version (28 subjects out of 36, that is 78%, gave an answer superior to 0) while was -0.62 for the second version (only 10 subjects out of 34, that is 29%, gave an answer superior to 0). So, people consider that the chairman intentionally had the apple tree cut down in the first version but not in the second. Can the Deep Self model account for this asymmetry? Question Version 1 Version 2 Comparison (Student t-test) (i) Intentionality 1.28 -0.62 p<.001*** (ii) Desire -0.86 0.30 p=.33 (iii) Reluctantly 2.83 1.65 p<.01** (iv) Importance 3.61 1.79 p<.001*** (v) Responsibility 5.47 4.94 p=0.24 Table 2. Mean answers to the different questions and their comparisons for the Apple Tree cases In this case, the Deep Self model predicts that the chairman’s Deep Self concords with the outcome in the first version but not in the second version. Our questions (ii), (iii) and (iv) constitute three different ways to measure this concordance. As we can see in Table 2, their results point towards the two following facts: In both versions, the outcome concords with the chairman’s deep self (he seems to be indifferent to the apple tree being cut down, doesn’t have it cut down reluctantly and doesn’t give it much importance). The outcome concords more with the chairman’s deep self in the second than in the first version. While, according to the Deep Self model, we should expect the following opposite results: In the second version, the outcome shouldn’t concord with the chairman’s deep self. Because the outcome is considered more intentional in the first version, the outcome should concord more with the chairman’s deep self in the first than in the second version As a result, it seems that the Deep Self model cannot account for the asymmetry between the two versions of the Apple Tree case. Now, one might object that this last experiment is unconvincing because, in these two cases, the Apple Tree being cut out is not a side effect, but a means: the chairman has the Apple Tree cut down in order to expand the building. But how is it problem if we focused on a means rather than on a side effect? There might be two ways in which it could be a problem. A first potential problem would be that, by studying a means, we went out of the scope of the Deep Self model. The Knobe effect being about side effects, it is natural to think that the Deep Self model is concerned only with side effects, since Sripada’s goal is to account for the Knobe effect. So, showing that the Deep Self model doesn’t work for means would be irrelevant, since the Deep Self model doesn’t make prediction on means. Nevertheless, this problem is not a real problem, since Sripada model also claims that his model can account for ascriptions of intentionality in the case of means. Indeed, Sripada claims that his model can account for judgments of intentionality in the following cases: Jack calmly and deliberately gave the homeless man his only jacket even though it was freezing outside. Did Jack intentionally give the homeless man his jacket? Jake desperately wants to win the rifle contest. He knows that he will only win the contest if he hits the bulls-eye. He raises the rifle, gets the bull’s eye in the sights, and presses the trigger. But Jake isn’t very good at using his rifle. His hand slips on the barrel of the gun, and the shot goes wild...Nonetheless, the bullet lands directly on the bull’s-eye. Jake wins the contest. Did Jake intentionally hit the bull’s eye? In both cases, we are not dealing with side effects, but with means (giving his jacket to help the homeless man, hitting the bull’s eye to win the contest). Sripada’s Deep Self model claims to account for ascriptions of intentionality in cases of means, and therefore it is fair to test it using questions bearing on means. A second problem would be that the fact that it is a means and not a side effect that we are dealing with might explain participants’ answers. Indeed, by default, means are judged intentional (see Cova and Naar, forthcoming) and that would explain why people consider that the chairman intentionally had the tree cut in the first case, even if the chairman doesn’t want the tree to be cut. Nevertheless, such an explanation would leave us with no clue as to why the same action is judged unintentional in the second case. Why do people judge a means to be unintentional in this case? Is it because it doesn’t concord with the chairman’s Deep Self? If so, then the same action should also be judged unintentional in the first case. Thus, even if we grant that the status of means can influence an action’s perceived intentionality, our results still leave the Deep Self model with a troubling puzzle Note that certain existing accounts do have an answer to propose to this puzzle. Thus Machery’s “trade-off hypothesis” would explain this asymmetry by saying that cutting the apple tree is a cost for the chairman in the first version but not in the second version (see Machery, 2008).. 5. Conclusion: Enhancing the Deep Self model In the light of these results, should we give up the Deep Self model? A possible line of defense against our counter-examples would be to claim that the Deep Self model did not aim at accounting for all our intentionality judgments, but only for our intentionality judgments in cases similar to the original Knobe Effect. We find this (hypothetical) line of defense unconvincing. First, our case of the Reluctant Chairman is very close to the original Harm case. How could it be possible that the Deep Self model successfully account for the Harm and the Help cases, but fail to account for the Reluctant Chairman? Second, we feel that such a defense would betray the spirit of the original Deep Self model: one of Sripada’s aims was to show that our intentionality judgments are not affected by moral considerations (the “Unidirectional Thesis”). How could the Deep Self model support such a claim if it was not intended as a general theory of how we attribute intentionality? Another line of defense would be to claim that the “Concordance criterion” is not the whole story but is still part of it: even if other factors can influence ascriptions of intentionality, we have not ruled out the possibility that the concordance between the agent’s Deep Self and the outcome sometimes increases the perceived intentionality of the action. Though also betraying Sripada’s original intent, such a defense would be right: we have not shown that such a concordance never influences our intentionality judgments. In fact, our aim is not to make such a claim, because we are more than willing to grant that this kind of concordance does have an impact on our ascriptions of intentionality. Indeed, we think there is a way to modify the Deep Self model that would allow it to survive our counter-examples and in which the concordance between the agent’s Deep Self and the outcome does play a role. To obtain this modified Deep Self model, we just have to exchange the Concordance criterion for the following: Expectation criterion: The outcome is intentional only if the agent’s Deep Self concords more with this outcome than we expected it to. This criterion can be analyzed as a conjunction of the two following sub-principles: If we expect the agent’s Deep Self to concord with the outcome, then this outcome is intentional only if the agent’s Deep Self concord with the outcome at least as much as we expected. If we expect the agent’s Deep Self to be opposed to the outcome, then this outcome is intentional only if the agent’s Deep Self is less opposed to the outcome than we expected it to be. Where “expectations” can be understood in two different ways: there are normative and descriptive expectations. When we expect A to x, we have normative expectations: we mean that A ought to x. We have descriptive expectations when we attribute a high probability to A’s x-ing. In the following, when we use the word “expectation”, we mean both types of expectations. Note that we do not claim that normative expectations influence people’s deep attitude attributions (a possibility suggested but rejected by Sripada). What we claim is that normative and descriptive expectations influences the degree of concordance (or discordance) with the agent’s Deep Self that is needed for an outcome to become intentional. For example, in the Reluctant Chairman case, we expect the chairman to be strongly opposed to harming the environment in order to make profits. We have such normative expectations because we think that he should prefer not to harm the environment to making profits. And we have such descriptive expectations because the case starts by describing the chairman as someone who has fought for the defense of the environment. So, although the chairman’s Deep Self seems to be opposed to the outcome (he seems to be pro-environment), it isn’t as much in opposition to it as we had expected it to be (the chairman is not pro-environment enough, and is more anti-environment than expected), and this is why we consider the outcome as intentional. In the Apple Tree cases, there are no strong normative expectations. For most subjects, cutting down the Apple Tree is neither good nor bad. But there are descriptive expectations, varying from one version to another. In the first version, subjects are told that the chairman has a strong past history with the apple tree. Through the text, the subjects came to expect that the chairman will be opposed to having the apple tree cut down, and to expect his Deep Self to be clearly anti-cutting-down. But the chairman doesn’t seem to care. As a consequence, its Deep Self is less opposed to the outcome than subjects expected it to be (it is indifferent while subjects expected strong discordance); the outcome is therefore judged intentional. On the other hand, in the second version, the scenario led subjects to predict that the chairman will be pleased to have the apple tree cut down. But, once again, he’s indifferent. So, his Deep Self concord less with the outcome than expected (it is indifferent while subjects expected a strong concordance) and the outcome is judged not intentional. We believe that these modifications would allow the Deep Self model to survive our attacks. But making this change will have an important consequence: it will make the Deep Self model an instance of the Bidirectional Thesis, since it will allow normative expectations to have an impact on our ascriptions of intentionality. Thus, although the Deep Self model was proposed as an alternative to the bidirectional thesis, we believe the best version of the model is one that in fact recognizes bidirectional influences of normative factors on intentionality judgments. Appendix: Testing Wright and Bengson’s account of asymmetries in judgments of intentionality Wright and Bengson (2009) have developed a competing account for the asymmetry between the Harm and the Help case. According to Wright and Bengson, this asymmetry is due to two factors: (i) an asymmetry in judgments of responsibility between the Harm and Help cases and (ii) the fact that people sometimes use their judgments of responsibility to infer judgments of intentionality (because they think that intentionality is most of the time a condition for responsibility). So, in the Harm case, we judge the chairman responsible for harming the environment and infer that he intentionally harmed the environment. In the Help case, by contrast, we consider that the chairman isn’t responsible for helping the environment and thus did not intentionally help the environment.s The asymmetry we have already mentioned between the two versions of the Apple Tree case also seems to constitute a counter-example to this hypothesis. According to Wright and Bengson, we should expect to observe an asymmetry in judgments of responsibility between the two versions. But the answers to question (v) that are described in Table 2 suggest that no such asymmetry was found: there are no significant differences for judgments of responsibility between the two versions and, in both cases, people tend to consider that the chairman was responsible for the removal of the apple tree. As a result, Wright and Bengson’s account is unable to accommodate and account for this asymmetry. References Cova, F. & Naar, H.(forthcoming) Side-effect effect without side effects: The pervasive impact of moral considerations on judgments of intentionality. Philosophical Psychology. Knobe, J. (2003a) Intentional action and side-effects in ordinary language. Analysis, 63, 190-193. Knobe, J. (2003b) Intentional action in folk psychology: an experimental investigation. Philosophical Psychology, 16, 309-324. Knobe, J. (2006) The concept of intentional action: a case study in the uses of folk psychology. Philosophical Studies, 130, 2, 203-231. Machery, E. (2008) The folk concept of intentional action: Philosophical and experimental issues. Mind & Language, 23, 165-189. Nadelhoffer, T. (2004a) Praise, side effects and intentional action. Journal of Theoretical and Philosophical Psychology, 24, 196-213. Nadelhoffer, T. (2004b) Blame, badness and intentional action: a reply to Knobe and Mendlow. Journal of Theoretical and Philosophical Psychology, 24, 259-269. Nadelhoffer, T. (2006) Bad acts, blameworthy agents and intentional actions: some problems for jury impartiality. Philosophical Explorations, 9, 2, 203-220. Nanay, B. (2010) Morality of modality? What does the attribution of intentionality depend on? Canadian Journal of Philosophy, 40, 28-40. Rose, D., Livengood, J., Sytsma, J. & Machery, E. (forthcoming) Deep trouble for the deep self. Philosophical Psychology. Sripada, C. (forthcoming) The ‘Deep Self’ model and asymmetries in folk judgments about intentionality and responsibility. Philosophical Studies. Sripada, C. & Konrath, S. (forthcoming) Telling more that we can know about intentional action. Mind & Language. Wible, A. (2009) Knobe, side effects, and the morally good business. Journal of Business Ethics, 85, 173-178. Wright, J. & Bengson, J. 2009: Asymmetries in folk judgments of responsibility and intentional action. Mind & Language, 24, 237-251.