The Idealized Mind (2025) examines how idealized models are used to interpret the nature and function of the mind and brain, whilst defending a version of scientific realism.
Drayson presents several challenges to this project. She says that The Idealized Mind claims that “some scientific theories are neither true nor false” and that this makes me “one of the very few philosophers since the logical empiricists to deny semantic realism.” That’s a misreading of the book’s position. My worry about semantic realism – the view that scientific discourse is truth-conditional – only concerns idealization. I defend the view that most scientific models consist of mixed components: (i) idealized parts, serving no representational function across the model-world relation; and (ii) nonidealized parts, which may represent (in various ways) aspects of the target system (see e.g., p. 61, pp. 68-69). There is no logical empiricism here. To suggest otherwise, would be to examine a straw version of the book’s main arguments.
Drayson worries about my nonliteral view of idealization (chapters 3-4): “how could we use the model to reason about the roles played by aspects of the target system unless the model also performed some sort of representational function with respect to the target system?” There’s no need to worry. Consider the Hardy-Weinberg model. It shows that in the absence of evolutionary forces, allele frequencies remain constant over generations. It is based on five idealizations: (a) zero mutation; (b) zero gene flow; (c) no natural selection; and (d) random mating. These idealizations make it the case that allele frequencies remain constant over generations in virtue of a crucial fifth idealization: (e) a biological population must be infinitely large. The idealizations are useful because they define a baseline ideal world in which allele frequencies do not change. The infinity function, I argue in the book, does not misrepresent biological populations as infinite. It is not useful, as many think, in virtue of being false. The infinity idealization is a mathematical tool that does one thing: it eliminates stochastic effects that occur in finite populations. The useful implication of this is: a clean baseline null model, i.e., any deviation from the deterministic mapping ‘allele frequency ® genotypic frequencies’, guaranteed by the infinity function, must be due to non-random mating, selection, population structure, etc. I argue that idealizations must be treated nonliterally, i.e., as lacking truth-values over the model-world relation. Yet, Drayson questions, how can one reason with such a model? I’ve just given you one example. Here’s a second one. One can make use of idealizations as the antecedent structures of a conditional to make inferences to measurable consequences about allele frequencies. This is what I argue in chapter 9 of The Idealized Mind, although in a discussion of the free energy principle in theoretical neuroscience.
Finally, Drayson worries that a nonliteral treatment of idealization makes scientific discourse meaningless, jeopardizing any defence of scientific realism. It is entirely possible to relax the semantic thesis of scientific realism without weakening a defence of scientific realism. Moreover, the insistence on scientific representation is unnecessary. Idealized models often look nothing like their targets.
References
Kirchhoff, Michael. 2025. The Idealized Mind: From Model-based Science to Cognitive Science. Cambridge, MA: The MIT Press.
Thanks, Michael. To be clear, I don’t claim that you’re a logical empiricist, but only that you share their rejection of semantic realism. (I take it there is more to logical empiricism than denying semantic realism, just as there is more to your position than denying semantic realism.) And it’s your specific denial that idealized scientific claims lack truth conditions that worries me: for these claims to be epistemically valuable, they’d need to be meaningful. The example you give of idealizations as antecendent structures of conditional reasoning is based on them being meaningful but not true. How do we understand their meaning? It can’t be from grasping their truth conditions, because your rejection of semantic realism for idealizations entails that idealizations lack truth conditions. So you can’t have a truth-conditional theory of meaning in mind – so I’d I’d like to hear more about your theory of meaning for idealized scientific claims.
I think some of the confusion here stems from the distinction between truth values and truth conditions. If all you want to claim is that idealized scientific claims lack truth values, then you don’t need to reject the idea that they have truth conditions – in which case you could keep semantic realism.
Hi Zoe
Many thanks for allowing me to add some additional detail to our discussion about semantic realism. Generally speaking, I understand ‘semantic realism’ as it is usually defined in the context of scientific realism, i.e., semantic realism takes scientific theories at face-value, seeing them as truth-conditioned descriptions of their intended domain, both observable and unobservable. Hence, they are capable of being true or false. Thus, semantic realism can be characterized as the idea that scientific theories (and models) are truth-bearers and that they are true or false in virtue of the world.
In the book, I don’t claim to reject semantic realism. I am perfectly fine with some elements of scientific models being consistent with semantic realism. The more interesting thing is: why are we so hung up on semantic realism? It’s the truth-conditional status of scientific theories that secures their meaning. Yet, should we understand scientific theories and models as inherently meaningful? This would imply that a scientific model, say, independent of ever being used would be meaningful. Or that a scientific model in a hypothetical vacuum would be meaningful and representational. I take this to be confused. Idealization, on my account, is a tool. Why would a tool need to be semantic? Idealizations can be given a rigorous mathematical definition and put to use unambiguously to derive certain results. However, idealizations have no objective truth-makers in and of themselves. All that said, idealized models can be used by scientists to say something true, accurate, meaningful and so on, where ‘semantics’ is contingent on practical use.
Inside a scientific model, idealization can have different uses. For example, as I mention in my response to your comment, the infinity function in the Hardy-Weinberg model removes stochasticity. And yet, the infinity function, if semantic realism must be accepted, will always be false across the model-world relation. I want to resist this standard view of idealization in the literature. I argue that one should relax semantic realism in such a way that idealization has no inherent truth-conditions from a model to a target phenomenon. Even if one is OK with this, one can still work in a meaningful way with idealizations: either by probing internal model consistency or by using idealizations in explaining aspects of the world (without this implying that idealizations must somehow be false of the world). The representational value of an idealization rests entirely on its use by, say, a scientist. By itself, an idealization is just a sophisticated modelling tool. It’s a ‘hammer’ but with significant mathematical properties.
In sum, my view of idealization is: (I) idealizations can be used to say something true or approximately true about the world even if idealizations themselves have no representational function; and (ii) idealizations can be used to say something meaningful about the world, even if they have no objective truth-bearers. I hope this goes someway towards answering your queries.
Michael
This is super helpful, Michael – thanks! I’m thinking of semantic realism as the universally quantified claim, so when I say that you reject it, I’m just claiming that you think that some idealizations are exceptions to semantic realism. I fully agree with you that idealizations could be used to say something meaningful about the world while lacking truthmakers, because I’m assuming that there’s a difference between having truth conditions and having truthmakers. And I’m fine with the idea that uninterpreted models are just tools. But I take standard scientific realism to involve semantic realism about scientific theories, which is why I raised the question in my piece about the relationship between models and theories. How are you thinking of this relationship? (I proposed that a theory can be true even if most of its models misrepresent.) Would you have time to expand upon this a little?
Hi Zoe
You’re right that standard scientific realism involves semantic realism about theories, of course. I think this assumption is misguided, as I argue in chapter 4 of the book. One way to answer the question you raise at the end is to distinguished between theory-embedded models and autonomous models (following Reutlinger et al. in BJPS). A theory embedded model might be a model of two planets embedded in Newtonian mechanics. The latter, assuming we’ve got our model calculations correct, ensures the truth of the two-planet model. I don’t really think we have anything like that in the sciences of the mind and brain. So my focus has mainly been on the model-target relation. In chapter 4, I argue (along the lines of Godfrey-Smith) that a good way to understand how a model can deliver accurate results or close to accurate results don’t imply that the model is true or close to true (or accurate). I’d want to say that a model doesn’t really contain any commitments about what the target phenomenon is like, which to me means that a model and its application are two different things. Hence, on my account, the application of the model might lead to true or approximately true or misleading accounts of target phenomena, even if parts of the model is neither true nor false of the target phenomena being investigated.
Thanks for engaging with my commentary and questions, Michael! I should take another look at Chapter 4. Looking forward to reading the other commentaries and your responses!