The target article by Phillips et al. provides converging evidence for assuming that representations of knowledge are more basic than representations of belief. Although some findings that they take as evidence for representations of knowledge may perhaps be given deflationary interpretations, I am sympathetic to their broad descriptive conclusion about the primacy of knowledge representations. Similarly, I agree with their methodological conclusion that mindreading research should focus more on knowledge representation. Consequently, I will argue for three further points about knowledge-oriented mindreading research.

The first point is that it is fallacious to conclude that knowledge itself is primary from evidence that representations of knowledge are primary. Generally, it is fallacious to move from assumptions about the primacy of a mental representation to conclusions about the primacy of its referent. For example (from Gerken, Reference Gerken, A. Gordon and Carter2017b), it is a safe bet that representations of water are more basic than representations of hydrogen in terms of ontogenesis, phylogenesis, automatic processing, and so on. But this does not entail that the substance water is more basic or primary than hydrogen. Such a representation-representandum fallacy regarding representations of knowledge may occur in both epistemology and cognitive science. The fallacy is not committed in the target article, although its title – Knowledge before belief – might encourage it. Therefore, to ensure that the surveyed evidence is used responsibly, it is important to warn against the representation-representandum fallacy. For example, it would be fallacious to take the surveyed evidence to motivate any knowledge-first program concerning knowledge rather than the concept of knowledge, the word “knowledge,” and so on (Williamson, Reference Williamson2000).

The second point is that knowledge-oriented mindreading research should study misrepresentations and biased knowledge representations. The primacy of knowledge representations may be partly explained in terms of bounded rationality: Representations of knowledge are basic and prominent partly because they are cognitively “cheap” ways of representing complex epistemic matters in a manner that is accurate enough for many purposes. Insofar, as representations of non-factive representations, which require one to keep track of both the mental representation and what it represents, are more cognitively taxing than representations of factive representations, cognitive bounds may partly explain the primacy of knowledge representations. However, bounded cognition involves biases.

The target article provides evidence that knowledge representations are automatically processed (sect. 4.3). But it does not mention that automatic processing of primitive representations often exhibits signature biases (Apperly, Reference Apperly2011; Saxe, Reference Saxe2005). Similarly, misrepresentations of knowledge are not discussed. Given the aim of the target article, its focus on successful representations of knowledge is natural. However, this focus suggests a misleading picture of social cognition and overshadows the methodological upshot that knowledge-oriented mindreading research must also focus on misrepresentations and biases. Some research suggests that the patterns of knowledge representations exhibit egocentric bias (Nagel, Reference Nagel2008); focal bias (Gerken, Alexander, Gonnerman, & Waterman, Reference Gerken, Alexander, Gonnerman and Waterman2020; Gerken & Beebe, Reference Gerken and Beebe2016); source-content bias (Turri, Reference Turri2015); subadditivity effects (Dinges, Reference Dinges2018); and so on. Signature biases of knowledge representations are important to study empirically because they are very consequential. For example, they may result in discriminatory epistemic injustice which occurs when someone is wronged specifically in her capacity as an epistemic subject (Fricker, Reference Fricker2013, p. 1320; Gerken, Reference Gerken2019). Because representations of knowledge play central roles in navigating social life, cases in which someone is wronged specifically in her capacity as a knower are especially harmful. Generally, given that knowledge representations are central to social cognition, it is important to empirically study their signature biases and the social ramifications thereof (Gerken, Reference Gerken2017a; Spaulding, Reference Spaulding2018).

My third point is that knowledge-oriented mindreading research must confront a problem of the gold standard: Interpreting empirical data from tasks involving knowledge is often complicated because the gold standard response to the task is disputed. As noted, some researchers argue that particular patterns of knowledge ascriptions reveal a bias (op. cit.). But others reject this and argue that these patterns reflect correct responses that illuminate the concept of knowledge, the word “knowledge” or even knowledge itself (e.g., DeRose, Reference DeRose2009; Knobe & Schaffer, Reference Schaffer and Knobe2012; Stanley, Reference Stanley2005). Some even suggest that to explain these patterns of knowledge ascriptions in terms of cognitive bias is to “explain away” the relevant evidence (DeRose, Reference DeRose2009; Stanley, Reference Stanley2005).

Presumably, the false-belief test is a widely employed experimental paradigm partly because of agreement about the gold standard response (e.g., saying that the agent will seek an object where she last saw it rather than at its new location in verbal false-belief tests; Wellman, Cross, & Watson, Reference Wellman, Cross and Watson2001). In contrast, the gold standard response is part of what is investigated in many mindreading tasks involving knowledge. This is simply a methodological challenge for research on knowledge representation and not a reason to stick with established experimental paradigms. Moreover, the problem of the gold standard is far from unique to research on knowledge representation, although it is pertinent to it because many aspects of knowledge are disputed. Minimal properties of knowledge, such as the four considered by Phillips et al. (sect. 2), are good starting points. However, it is disputed how knowledge is related to luck, practical factors, actionability, competence, and so on. Knowledge-oriented mindreading research should study such relationships. But disputes over the gold standard response to tasks involving them constitute a methodological challenge in interpreting findings. Interestingly, the first two points mark specific methodological pitfalls. Given that some patterns of folk knowledge representation are biased (point 2), moving too swiftly from findings about participants' representations of knowledge to conclusions about whether they are correct would exemplify the representation-representandum fallacy (point 1).

In sum, here are my three main points:

1. (1) Evidence of primacy of knowledge representation is not evidence of primacy of knowledge.

2. (2) Knowledge-oriented mindreading research should also focus on misrepresentations and biased representations of knowledge.

3. (3) Knowledge-oriented mindreading research must confront the problem of the gold standard.

These three points are compatible with the surveyed evidence and the main conclusions that Phillips et al. draw from it. But the points are not included in Phillips et al.'s conclusions. Therefore, I wonder whether they agree with them or not.