Joshua Knobe makes three main claims in his paper. The first is that the influence of moral considerations on our judgments does not appear to be limited to the concept intentionally, nor even to closely related concepts such as intention and intending (sect. 3.2). Thus, it appears to affect our judgments about causation, knowledge, desire, and a number of other attitudes or processes. Knobe's second main claim is that the asymmetry found by Knobe and colleagues in people's judgments for such cases depends essentially on our normative evaluation with regard to counterfactual actions or situations; namely, on what should or could have been the case. Knobe's third claim and fairly radical conclusion, finally, is that we cannot make “a clear division whereby certain psychological processes are devoted to moral questions and others are devoted to purely scientific questions” (sect. 6, last para.).
In this commentary I would like to add further evidence in support of Knobe's first two claims, but express why I think we should be skeptical of the main conclusion he draws from them.
In agreement with Knobe's first claim, it may be pointed out that moral considerations influence at least two other general competences that would appear prima facie to be non-moral and that are not mentioned in Knobe's paper; namely, our qualitative evaluation of precise numerical probabilities and our qualitative evaluation of precise quantities (Egré Reference Egré2010). The evaluation of identical numerical probabilities is known to be subjectively influenced by how detrimental the outcome is perceived. The effect has been called the severity bias in the psychological literature (see Bonnefon & Villejoubert Reference Bonnefon and Villejoubert2006; Pighin et al. Reference Pighin, Bonnefon and Savadori2009; Weber & Hilton Reference Weber and Hilton1990).
For example, Pighin et al. (Reference Pighin, Bonnefon and Savadori2009) ran an experiment comparing the evaluations made by four groups of pregnant women of a scenario in which a gynecologist tells Elisa, a 30-year-old pregnant woman, that “there is a risk of [1 in 28; 1 in 307] that your child will be affected by [Down's syndrome; insomnia].” Subjects in each group were asked to rank the probability communicated for each disease on a 7-point scale ranging from “extremely low” to “extremely high.”
What Pighin et al.'s study found was that when the numerical risk for the two conditions was made the same, the women still ranked the probability of the child getting Down's syndrome as significantly higher than for insomnia. Even the probability of 1/307 for the child getting Down's syndrome was ranked higher than the probability of 1/28 for insomnia. Moreover, subjects were asked to rank each disease according to how severe they judged it to be. Their assessments of probabilities were found to correlate with those severity judgments.
Cova and Egré (Reference Cova and Egré2010) looked for the same effect regarding people's qualitative evaluation of identical quantities in terms of the word many. Subjects were given a scenario reporting that a fire had broken out in a school in which there were 10 children, 5 of whom died in the fire and 5 managed to escape. Each subject had to judge true or false the two sentences: “Many children perished in the fire. Many children survived from the fire.” Irrespective of the order in which the sentences were presented, the vast majority of subjects agreed that many children had perished; but they did not agree that many children had survived, despite the identical quantities and ratios involved.
Such cases comport with Knobe's model and main explanatory hypothesis in his paper (see also Pettit & Knobe Reference Pettit and Knobe2009); namely, they suggest that our subjective evaluation of probabilities or quantities, just like our evaluation of causation or intentional action, is sensitive not only to extensive magnitudes or processes, but also to normative expectations that are highly context-dependent and that vary with the kind of outcome under consideration.
For example, it is known from the semantics literature that our judgments concerning whether many As are Bs are not purely extensional (see Fara Reference Fara2000; Lappin Reference Lappin2000; Sapir Reference Sapir1944). That is, as the data with Cova confirm, those judgments do not merely depend on the cardinality of As and Bs and on the ratio of As to Bs; they intensionally depend on the kind of entities referred to by A and B, and on what is taken to be either normal or more desirable relative to context.
In agreement with Knobe's remark about the importance of counterfactual evaluations, presumably we judge that many children died because we reason that in a better and alternative course of events, fewer children would have died (and as a result, that more would have survived). Similarly, how high a probability value is considered for an outcome may depend on how much more probable or less probable we consider that outcome could be or should have been.
It would be quite doubtful, however, to infer from those considerations that we cannot distinguish between the moral processes that influence our qualitative evaluation of quantities or probabilities and the non-moral processes that underlie our scientific judgments based on numerical quantities or probabilities. Indeed, when it comes to having a scientific attitude towards relative or absolute quantities, our evaluation can safely rely on our nonmoral capacity to compare extensional magnitudes. (Contrast “Did many children die?,” which calls on our subjective and moral evaluation, with “How many children died?,” which can be given an exact and objective answer.)
More generally, I wish to make the qualification that whereas sensitivity to normative expectations is most likely directly encoded in the lexical semantics of most of our qualitative vocabulary (see Egré Reference Egré2010; Kennedy Reference Kennedy2007), including for vague concepts such as knowing, desiring, causing, and so on, this remains compatible with the hypothesis Knobe appears to reject in his article. That is, it is compatible with the view that our cognitive competence in each of those domains works in a modular way, based on the interaction of non-moral evaluative faculties and moral evaluative faculties.
From Knobe's interesting data and examples, it would be safer to conclude that our folk concepts of causation, knowledge, and desire are irreducibly norm-sensitive, without that impugning the division between moral and non-moral cognition.
Joshua Knobe makes three main claims in his paper. The first is that the influence of moral considerations on our judgments does not appear to be limited to the concept intentionally, nor even to closely related concepts such as intention and intending (sect. 3.2). Thus, it appears to affect our judgments about causation, knowledge, desire, and a number of other attitudes or processes. Knobe's second main claim is that the asymmetry found by Knobe and colleagues in people's judgments for such cases depends essentially on our normative evaluation with regard to counterfactual actions or situations; namely, on what should or could have been the case. Knobe's third claim and fairly radical conclusion, finally, is that we cannot make “a clear division whereby certain psychological processes are devoted to moral questions and others are devoted to purely scientific questions” (sect. 6, last para.).
In this commentary I would like to add further evidence in support of Knobe's first two claims, but express why I think we should be skeptical of the main conclusion he draws from them.
In agreement with Knobe's first claim, it may be pointed out that moral considerations influence at least two other general competences that would appear prima facie to be non-moral and that are not mentioned in Knobe's paper; namely, our qualitative evaluation of precise numerical probabilities and our qualitative evaluation of precise quantities (Egré Reference Egré2010). The evaluation of identical numerical probabilities is known to be subjectively influenced by how detrimental the outcome is perceived. The effect has been called the severity bias in the psychological literature (see Bonnefon & Villejoubert Reference Bonnefon and Villejoubert2006; Pighin et al. Reference Pighin, Bonnefon and Savadori2009; Weber & Hilton Reference Weber and Hilton1990).
For example, Pighin et al. (Reference Pighin, Bonnefon and Savadori2009) ran an experiment comparing the evaluations made by four groups of pregnant women of a scenario in which a gynecologist tells Elisa, a 30-year-old pregnant woman, that “there is a risk of [1 in 28; 1 in 307] that your child will be affected by [Down's syndrome; insomnia].” Subjects in each group were asked to rank the probability communicated for each disease on a 7-point scale ranging from “extremely low” to “extremely high.”
What Pighin et al.'s study found was that when the numerical risk for the two conditions was made the same, the women still ranked the probability of the child getting Down's syndrome as significantly higher than for insomnia. Even the probability of 1/307 for the child getting Down's syndrome was ranked higher than the probability of 1/28 for insomnia. Moreover, subjects were asked to rank each disease according to how severe they judged it to be. Their assessments of probabilities were found to correlate with those severity judgments.
Cova and Egré (Reference Cova and Egré2010) looked for the same effect regarding people's qualitative evaluation of identical quantities in terms of the word many. Subjects were given a scenario reporting that a fire had broken out in a school in which there were 10 children, 5 of whom died in the fire and 5 managed to escape. Each subject had to judge true or false the two sentences: “Many children perished in the fire. Many children survived from the fire.” Irrespective of the order in which the sentences were presented, the vast majority of subjects agreed that many children had perished; but they did not agree that many children had survived, despite the identical quantities and ratios involved.
Such cases comport with Knobe's model and main explanatory hypothesis in his paper (see also Pettit & Knobe Reference Pettit and Knobe2009); namely, they suggest that our subjective evaluation of probabilities or quantities, just like our evaluation of causation or intentional action, is sensitive not only to extensive magnitudes or processes, but also to normative expectations that are highly context-dependent and that vary with the kind of outcome under consideration.
For example, it is known from the semantics literature that our judgments concerning whether many As are Bs are not purely extensional (see Fara Reference Fara2000; Lappin Reference Lappin2000; Sapir Reference Sapir1944). That is, as the data with Cova confirm, those judgments do not merely depend on the cardinality of As and Bs and on the ratio of As to Bs; they intensionally depend on the kind of entities referred to by A and B, and on what is taken to be either normal or more desirable relative to context.
In agreement with Knobe's remark about the importance of counterfactual evaluations, presumably we judge that many children died because we reason that in a better and alternative course of events, fewer children would have died (and as a result, that more would have survived). Similarly, how high a probability value is considered for an outcome may depend on how much more probable or less probable we consider that outcome could be or should have been.
It would be quite doubtful, however, to infer from those considerations that we cannot distinguish between the moral processes that influence our qualitative evaluation of quantities or probabilities and the non-moral processes that underlie our scientific judgments based on numerical quantities or probabilities. Indeed, when it comes to having a scientific attitude towards relative or absolute quantities, our evaluation can safely rely on our nonmoral capacity to compare extensional magnitudes. (Contrast “Did many children die?,” which calls on our subjective and moral evaluation, with “How many children died?,” which can be given an exact and objective answer.)
More generally, I wish to make the qualification that whereas sensitivity to normative expectations is most likely directly encoded in the lexical semantics of most of our qualitative vocabulary (see Egré Reference Egré2010; Kennedy Reference Kennedy2007), including for vague concepts such as knowing, desiring, causing, and so on, this remains compatible with the hypothesis Knobe appears to reject in his article. That is, it is compatible with the view that our cognitive competence in each of those domains works in a modular way, based on the interaction of non-moral evaluative faculties and moral evaluative faculties.
From Knobe's interesting data and examples, it would be safer to conclude that our folk concepts of causation, knowledge, and desire are irreducibly norm-sensitive, without that impugning the division between moral and non-moral cognition.
ACKNOWLEDGMENT
Research supported by the Agence Nationale de la Recherche (grant ANR-07-JCJC-0070). Thanks to F. Cova, S. Pighin, D. Ripley, and P. Schlenker for exchanges related to this commentary.