I highly appreciate Elqayam & Evan's (E&E's) keen criticism of is-ought and ought-is fallacies in the psychological literature on reasoning. In this comment I shed critical light on E&E's account of normative systems of reasoning and of prescriptive normativism, which is understood by E&E as the view that human (rational) thinking ought to conform to a normative system of reasoning, S. Differing kinds of prescriptive normativism result from different normative systems S−but according to E&E (target article, sect. 1), S must always be a formal normative system, such as logic, probability theory, or decision theory.
My first objection concerns formality. While E&E are right that the normative systems that have dominated the psychology of reasoning have been formal systems, not every normative system of reasoning needs to be formal. For example, systems of spatio-temporal, mechanical, or chemical reasoning are not formal. Formality and universality are connected as follows: Formality implies universality, because formal systems abstract from the content of non-logical symbols (Schurz Reference Schurz2005). The other direction need not hold: A non-formal system of reasoning (e.g., reasoning in chemistry) can still be highly universal. For E&E, the reasoning systems of prescriptive normativism are typically universal, although they admit exceptions (sect. 1). In contrast, I want to suggest that the question of prescriptive normativism should be entirely separated from the questions of formality and universality. For example, Gigerenzer, who is classified by E&E as an anti-normativist (sect. 1), behaves like a prescriptive normativist when he recommends certain heuristic rules for prediction tasks, only that he regards these rules as neither formal nor universal (Gigerenzer et al. Reference Gigerenzer and Todd1999).
My second objection concerns E&E's overly narrow understanding of prescriptive normativism. This understanding rests on their distinction between (what they call) evaluative norms and instrumental (or directive) norms, and corresponding forms of rationality. While instrumental norms tell what one ought to do to achieve one's personal goals, evaluative norms tell what one ought to do simpliciter. For example, Gigerenzer's adaptive rationality is for E&E a kind of instrumental rationality, but Oaksford and Chater's Bayesian rational analysis (see Oaksford & Chater Reference Oaksford and Chater2007) is a kind of evaluative rationality. I think that E&E's distinction is misleading because both kinds of rationality, “adaptive” and “Bayesian,” have in turn to be justified in the light of the same superordinate epistemic norms.
E&E (sect. 2, para. 4) write that their distinction is close to my distinction between the “normative” and the “valuative” (Schurz Reference Schurz1997), but this is a misunderstanding. E&E's distinction is related, rather, to my distinction between fundamental versus derived norms, and corresponding values. According to the standard characterization in ethics (Frankena Reference Frankena1963), instrumental norms are norms that are derived from other−given and ultimately fundamental−norms, by the means-end principle that runs as follows: If (premise 1:) A is a (fundamental or given) norm, and (premise 2:) B is a necessary (or optimal) means for achieving A, then (conclusion:) B is a derived norm (Schurz Reference Schurz1997, Ch. 11.2, p. 8). While premise 1 is normative, premise 2 is descriptive-empirical in character. Therefore, the means-end principle explains how the findings of empirical scientists can become practically relevant without committing an is-ought fallacy: Empirical findings allow one to derive a multitude of derived norms from a small set of extra-scientifically given (intersubjectively accepted) fundamental norms (Schurz Reference Schurz and Pigden2010, sect. 6).
E&E think that “instrumental norms” are not value-laden but descriptive (sect. 2), because they assume that the fundamental norms from which they are derived are personal goals. But for many ethically important instrumental norms (such as “reduce carbon dioxide emissions”) the assumed fundamental norm has a high degree of intersubjective acceptance. This means that instrumental norms are also more-or-less value-laden: They inherit their value from the value of the fundamental norm or goal from which they are derived.
Concerning “evaluative” norms, E&E seem to assume that they are not in need of further justification by higher-ranking norms. This view is also mistaken. For example, E&E's evaluative norm “Poverty should not exist” (sect. 2, para. 3) is instrumental for the higher-ranking norm that people should not suffer. Just the same is true for normative systems of reasoning. E&E, or at least the psychological tradition which they criticize, seem to assume that normative reasoning systems are accepted on the basis of mere intuition or authority. This view has its historical roots in the Kantian doctrine that principles of reasoning are a priori and necessary. This view is no longer considered tenable in contemporary epistemology.
Norms of reasoning have to be justified by higher-ranking epistemic goals. The most important fundamental epistemic goal−not mentioned in E&E's article−is truth-conduciveness: Rules of reasoning should be a reliable means to increase true beliefs and avoid false beliefs (Bishop & Trout Reference Bishop and Trout2005; Goldman Reference Goldman1986; Reference Goldman1999; Leplin Reference Leplin2009, Ch. 2; Schurz Reference Schurz2009). This gold-standard of epistemology is, for example, used to justify the inference rules of classical logic by the provable fact that they lead in all possible worlds from true premises to true conclusions. Given recent findings that the conditionals in human reasoning correspond more to high conditional probability assertions than to exceptionless implications of classical logic (Evans et al. Reference Evans, Handley and Over2003), the same gold-standard has been used to argue that the system of (conditional) reasoning that maximizes truth-conduciveness is not classical logic, but system P or the stronger system Z of probabilistic default reasoning (Adams Reference Adams1998; Goldszmidt & Pearl Reference Goldszmidt and Pearl1996; Leitgeb Reference Leitgeb2004; Schurz Reference Schurz1998; Reference Schurz2005; Reference Schurz, Vosniadou, Kayser and Protopapas2007). Incidentally, this fact contradicts E&E's claim (sect. 3, para. 3 and Table 1) that conditional elimination inferences constitute a single-norm paradigm. Truth-conduciveness has also been applied to demonstrate the optimality of meta-induction in prediction tasks (Schurz Reference Schurz2008). Other generally accepted epistemic goals (besides truth-conduciveness) are simplicity and efficiency.
In conclusion, while I agree with E&E's criticism of certain dogmatic views on normative systems of reasoning, I don't agree with their rejection of normativism per se. A non-dogmatic and weak version of prescriptive normativism, in the form of normative recommendations regarding which typical human thinking processes are more and which are less truth-conducive and reliable, is still an important part of practically applied psychology.
I highly appreciate Elqayam & Evan's (E&E's) keen criticism of is-ought and ought-is fallacies in the psychological literature on reasoning. In this comment I shed critical light on E&E's account of normative systems of reasoning and of prescriptive normativism, which is understood by E&E as the view that human (rational) thinking ought to conform to a normative system of reasoning, S. Differing kinds of prescriptive normativism result from different normative systems S−but according to E&E (target article, sect. 1), S must always be a formal normative system, such as logic, probability theory, or decision theory.
My first objection concerns formality. While E&E are right that the normative systems that have dominated the psychology of reasoning have been formal systems, not every normative system of reasoning needs to be formal. For example, systems of spatio-temporal, mechanical, or chemical reasoning are not formal. Formality and universality are connected as follows: Formality implies universality, because formal systems abstract from the content of non-logical symbols (Schurz Reference Schurz2005). The other direction need not hold: A non-formal system of reasoning (e.g., reasoning in chemistry) can still be highly universal. For E&E, the reasoning systems of prescriptive normativism are typically universal, although they admit exceptions (sect. 1). In contrast, I want to suggest that the question of prescriptive normativism should be entirely separated from the questions of formality and universality. For example, Gigerenzer, who is classified by E&E as an anti-normativist (sect. 1), behaves like a prescriptive normativist when he recommends certain heuristic rules for prediction tasks, only that he regards these rules as neither formal nor universal (Gigerenzer et al. Reference Gigerenzer and Todd1999).
My second objection concerns E&E's overly narrow understanding of prescriptive normativism. This understanding rests on their distinction between (what they call) evaluative norms and instrumental (or directive) norms, and corresponding forms of rationality. While instrumental norms tell what one ought to do to achieve one's personal goals, evaluative norms tell what one ought to do simpliciter. For example, Gigerenzer's adaptive rationality is for E&E a kind of instrumental rationality, but Oaksford and Chater's Bayesian rational analysis (see Oaksford & Chater Reference Oaksford and Chater2007) is a kind of evaluative rationality. I think that E&E's distinction is misleading because both kinds of rationality, “adaptive” and “Bayesian,” have in turn to be justified in the light of the same superordinate epistemic norms.
E&E (sect. 2, para. 4) write that their distinction is close to my distinction between the “normative” and the “valuative” (Schurz Reference Schurz1997), but this is a misunderstanding. E&E's distinction is related, rather, to my distinction between fundamental versus derived norms, and corresponding values. According to the standard characterization in ethics (Frankena Reference Frankena1963), instrumental norms are norms that are derived from other−given and ultimately fundamental−norms, by the means-end principle that runs as follows: If (premise 1:) A is a (fundamental or given) norm, and (premise 2:) B is a necessary (or optimal) means for achieving A, then (conclusion:) B is a derived norm (Schurz Reference Schurz1997, Ch. 11.2, p. 8). While premise 1 is normative, premise 2 is descriptive-empirical in character. Therefore, the means-end principle explains how the findings of empirical scientists can become practically relevant without committing an is-ought fallacy: Empirical findings allow one to derive a multitude of derived norms from a small set of extra-scientifically given (intersubjectively accepted) fundamental norms (Schurz Reference Schurz and Pigden2010, sect. 6).
E&E think that “instrumental norms” are not value-laden but descriptive (sect. 2), because they assume that the fundamental norms from which they are derived are personal goals. But for many ethically important instrumental norms (such as “reduce carbon dioxide emissions”) the assumed fundamental norm has a high degree of intersubjective acceptance. This means that instrumental norms are also more-or-less value-laden: They inherit their value from the value of the fundamental norm or goal from which they are derived.
Concerning “evaluative” norms, E&E seem to assume that they are not in need of further justification by higher-ranking norms. This view is also mistaken. For example, E&E's evaluative norm “Poverty should not exist” (sect. 2, para. 3) is instrumental for the higher-ranking norm that people should not suffer. Just the same is true for normative systems of reasoning. E&E, or at least the psychological tradition which they criticize, seem to assume that normative reasoning systems are accepted on the basis of mere intuition or authority. This view has its historical roots in the Kantian doctrine that principles of reasoning are a priori and necessary. This view is no longer considered tenable in contemporary epistemology.
Norms of reasoning have to be justified by higher-ranking epistemic goals. The most important fundamental epistemic goal−not mentioned in E&E's article−is truth-conduciveness: Rules of reasoning should be a reliable means to increase true beliefs and avoid false beliefs (Bishop & Trout Reference Bishop and Trout2005; Goldman Reference Goldman1986; Reference Goldman1999; Leplin Reference Leplin2009, Ch. 2; Schurz Reference Schurz2009). This gold-standard of epistemology is, for example, used to justify the inference rules of classical logic by the provable fact that they lead in all possible worlds from true premises to true conclusions. Given recent findings that the conditionals in human reasoning correspond more to high conditional probability assertions than to exceptionless implications of classical logic (Evans et al. Reference Evans, Handley and Over2003), the same gold-standard has been used to argue that the system of (conditional) reasoning that maximizes truth-conduciveness is not classical logic, but system P or the stronger system Z of probabilistic default reasoning (Adams Reference Adams1998; Goldszmidt & Pearl Reference Goldszmidt and Pearl1996; Leitgeb Reference Leitgeb2004; Schurz Reference Schurz1998; Reference Schurz2005; Reference Schurz, Vosniadou, Kayser and Protopapas2007). Incidentally, this fact contradicts E&E's claim (sect. 3, para. 3 and Table 1) that conditional elimination inferences constitute a single-norm paradigm. Truth-conduciveness has also been applied to demonstrate the optimality of meta-induction in prediction tasks (Schurz Reference Schurz2008). Other generally accepted epistemic goals (besides truth-conduciveness) are simplicity and efficiency.
In conclusion, while I agree with E&E's criticism of certain dogmatic views on normative systems of reasoning, I don't agree with their rejection of normativism per se. A non-dogmatic and weak version of prescriptive normativism, in the form of normative recommendations regarding which typical human thinking processes are more and which are less truth-conducive and reliable, is still an important part of practically applied psychology.