Rational analysis has found multiple applications in generating behavioral predictions based on task structure (e.g., Oaksford & Chater Reference Oaksford and Chater1994; Sims et al. Reference Sims, Neth, Jacobs and Gray2013). Yet, the initial two uses of this method have demonstrated its potential for more. First, the rational analysis of memory (Anderson & Milson Reference Anderson and Milson1989; Anderson & Schooler Reference Anderson and Schooler1991) did not only predict behavior, but, in fact, developed a theory of that cognitive capacity. This theory has been successfully applied to many memory (e.g., Anderson et al. Reference Anderson, Bothell, Lebiere and Matessa1998; Schneider & Anderson Reference Schneider and Anderson2012) and decision-making tasks (e.g., Dimov & Link Reference Dimov and Link2017; Fechner et al. Reference Fechner, Pachur, Schooler, Mehlhorn, Battal, Volz and Borst2016). Second, the rational analysis of categorization (Anderson Reference Anderson1991) also achieved more than predict behavior: Keeping considerations of cognitive plausibility in mind, an algorithm was developed that sequentially assigns stimuli to categories. To summarize, in addition to predicting behavior, rational analysis can be used to develop theories of cognitive capacities given environmental constraints and, second, it also penetrates the algorithmic level when necessary to explain experimental data.

With their resource-rational analysis, Lieder and Griffiths extend rational analysis by including cognitive constraints into the optimization function. The viability and explanatory power of resource-rational analysis is well supported by many successful applications. However, many of these applications have one thing in common: They are relatively independent of the underlying cognitive capacities. Thus, one major constraint of resource-rational analysis – computational resources – is often avoided, in many cases reducing the approach to that of finding the optimal algorithm under task constraints – a procedure strikingly similar to rational analysis. The reason, also acknowledged by the authors themselves, is that measuring cognitive constraints has progressed slowly.

Behind this slow progress lies a fundamental difficulty in reverse-engineering the nature of the cognitive system's components. As argued by Newell (Reference Newell1990), each psychological experiment produces output that is the joint product of several cognitive processes. Consequently, data collected to advance our understanding of one cognitive process is marred with unexplained variance from several others, which renders determining the exact structure of the process under investigation problematic. To address this issue, Newell proposed iteratively developing and refining a model of the entire cognitive system – a unified theory of the mind – that provides a unified account of an ever-increasing number of psychological tasks. By jointly carving away unexplained variance from all components of the mind, such a theory would enable each subsequent experiment to ask more specific questions about the psychological process it investigates.

Newell's behest was followed by several cognitive architectures, the most developed among which is likely ACT-R (Anderson Reference Anderson2007). This architecture has incorporated the rational analysis of memory into the earlier ACT* (Anderson Reference Anderson1983), added the perceptual and motor processes, meticulously developed for the EPIC cognitive architecture (Meyer & Kieras Reference Meyer and Kieras1997a; Reference Meyer and Kieras1997b), and linked its components to regions in the brain (Anderson et al. Reference Anderson, Zhang, Borst and Walsh2016; Borst & Anderson Reference Borst and Anderson2017). Currently, it is able to account for behavior in hundreds of tasks in various fields, which include language learning and comprehension (Budiu & Anderson Reference Budiu and Anderson2004; Taatgen & Anderson Reference Taatgen and Anderson2002), decision making (Marewski & Schooler Reference Marewski and Schooler2011), driving (Salvucci & Taatgen Reference Salvucci and Taatgen2010), and many others (see http://act-r.psy.cmu.edu/publication/ for a list of publications categorized by field). This likely makes ACT-R the best source of cognitive constraints for resource-rational analysis.

Whether ACT-R is the cognitive theory of choice or not, Newell's arguments remain valid today: addressing the identifiability problem (Anderson Reference Anderson1990), the irrelevant specification problem (Newell Reference Newell1990), or the problem of amortization of theoretical constructs (Newell Reference Newell1990) is likely to be most successful with a unified theory of the mind that progressively incorporates multiple constraints from experiments, evolutionary arguments, and functional considerations. In my opinion, we should be devoting more efforts to develop such theories to accelerate our understanding of the mind as even the leader, ACT-R, despite its many successes, is still far from complete: It lacks theories of some fundamental components of the mind, such as emotions and tactile and other sensations, whereas many of the currently included components will likely be subjected to multiple refinements and extensions as this architecture is put to the test in new tasks.

Advancing a unified theory of the mind will naturally benefit approaches such as resource-relational analysis. Moreover, I believe that this approach might play a role in unveiling the structure of the mind similar to the role rational analysis played in developing a theory of memory. Specifically, if we maintain the assumption of optimality, we can ask under what cognitive and task constraints the empirically observed algorithms would be optimal, which could allow us to narrow down the plausible region in the space of possible computational resources. Such synergies between optimization approaches and cognitive architectures coupled with growing efforts in developing the latter will likely lead to considerable advancements in our understanding of human cognition and behavior.