At the conclusion of their thought-provoking paper, Johnson et al. provide some vivid imagery in an effort to contextualise their contribution. They contrast a view of Conviction Narrative Theory (CNT) as comprising “too much theoretical meat” (sect. 10.1, para. 2) with a view that the theory is “skeletal,” containing too little substance. This tension between the meat and the bones, or the details versus the framework, runs throughout the target article and is never fully resolved – and so exactly what CNT offers to scholars of decision-making remains unclear.
At the heart of this tension lies an unresolved contradiction regarding the scope of the theory. In some contexts, Johnson et al. designate CNT as merely a theory of decision-making under radical uncertainty and not a theory of the many crucial aspects of cognition – explanation, analogy, causation, emotion – that are often abstracted away in classical approaches to judgement and decision-making. In other contexts, narratives are being offered as the all-encompassing “currency of thought.” Determining how broadly CNT can contribute to our understanding necessitates consideration of how a decision unfolds.
The first task for any decision maker, as Johnson et al. note, is to understand their current situation. How people achieve this understanding is an issue that appears in various guises across cognitive science. The “frame problem” in artificial intelligence research, the “correspondence problem” in judgement and decision-making and the “problem of induction” in philosophy all ask how agents are able to identify and act on relevant aspects of a situation. To adapt one of Johnson et al.'s examples: How does the widower determine the provenance of the noise that awoke him and decide what to do?
Johnson et al.'s claim appears to be that the first thing the widower does is to realise that he is under conditions of radical uncertainty in which probabilities are unknowable. How this initial step is achieved is not specified by the theory but it seems crucial if the purview is decision-making under radical uncertainty and not all decision-making, or indeed deliberative behaviour in general.
If narratives are only invoked when probabilities do not apply then the theory needs to explain how people know they are in such a situation. If the claim is broader and that narratives, or perhaps the internal monologues that accompany them, are used as a general guide to thinking and deciding – the “currency of thought” – then there is a deep concern that the theory loses coherence and becomes a set of generic statements about various tenuously related cognitive and emotional processes. There is clearly value in pointing out the relevance of other research areas to the endeavour of understanding behaviour that is often viewed through the narrow lens of expected utility theory (EUT), but to claim that this broadening of perspective constitutes a novel theory seems like overreach.
Returning to our widower, the idea that he explicitly runs a probability calculus over the possible outcomes of different actions is, naturally, absurd. He has most likely got to this stage in his life without ever hearing of EUT or its implications – just like the vast majority of people. As such, asking whether or how he knows that he is under conditions of radical uncertainty is also nonsensical. He has heard a sound and has to make a decision. Invoking EUT to explain how he does this might provide a way to think about and analyse the potential consequences of different actions but it is not a psychological explanation of his behaviour (and was never intended to be, e.g., by von Neumann & Morgernstern, Reference von Neumann and Morgernstern1947).
We agree with Johnson et al. that narratives provide another way to think about this process, but just as probability-based theories can be critiqued for not explaining “where the numbers come from,” narrative approaches, like CNT, suffer from under-specification of how narratives are constructed (e.g., Klayman, Reference Klayman2001; Newell, Lagnado, & Shanks, Reference Newell, Lagnado and Shanks2022). By focusing only on a narrow set of all possible representations (complex narratives), CNT is unable to account for when and why people might use simpler strategies to deal with problems. Johnson et al. claim that narratives are higher-order representations that flexibly include lower-order representations, but the threshold for lower versus higher-order representations remains murky, as does the mechanism for flexible integration. This is problematic because as long as we do not understand how the relevant narratives are themselves constructed (or indeed any representation, simple to complex), we also cannot know when and why people would rely on them to make choices.
To answer the when and why questions, we would typically turn to diagnostic empirical evidence but here again the picture is rather unclear. Johnson et al. have a somewhat conflicting view stating both that lab-experimentation is “ill-suited” for testing the prevalence of narrative thinking, and that “finer-grained experimental evidence” is required to draw more definitive conclusions than those derived on the basis of qualitative interviews.
One potential solution to these problems is to start more modestly and focus on how people build up simple representations. Not all situations necessitate complex, emotionally charged simulations and evaluations, but they always necessitate the development of some kind of representation. Considering people to be “intuitive scientists” puts the focus on the process of how representations are generated and not on a particular kind of complex representation (Szollosi & Newell, Reference Szollosi and Newell2020; Szollosi, Donkin, & Newell, Reference Szollosi, Donkin and Newell2022). From this perspective, understanding how people develop relatively simple representations – such as of integer numbers (Carey & Barner, Reference Carey and Barner2019) or of the frequency of various events (Mason, Szollosi, & Newell, Reference Mason, Szollosi and Newell2022) – can provide a useful starting point by enabling clearer ways to measure or manipulate the range of factors that Johnson et al. (rightly) claim play into any decision.
Explaining how such representations emerge does not require probabilities, or utilities, or the kinds of calculations that CNT wants to eschew, and we expect that focussing at this simpler level could eventually provide a window into how more complex representations (narratives) are built from primitive information. Without thinking carefully about these initial steps of flexible representation generation and selection there is a risk that the theoretical “tastiness” of CNT will be lost in a soup of conceptual complexity.
You have
Access
At the conclusion of their thought-provoking paper, Johnson et al. provide some vivid imagery in an effort to contextualise their contribution. They contrast a view of Conviction Narrative Theory (CNT) as comprising “too much theoretical meat” (sect. 10.1, para. 2) with a view that the theory is “skeletal,” containing too little substance. This tension between the meat and the bones, or the details versus the framework, runs throughout the target article and is never fully resolved – and so exactly what CNT offers to scholars of decision-making remains unclear.
At the heart of this tension lies an unresolved contradiction regarding the scope of the theory. In some contexts, Johnson et al. designate CNT as merely a theory of decision-making under radical uncertainty and not a theory of the many crucial aspects of cognition – explanation, analogy, causation, emotion – that are often abstracted away in classical approaches to judgement and decision-making. In other contexts, narratives are being offered as the all-encompassing “currency of thought.” Determining how broadly CNT can contribute to our understanding necessitates consideration of how a decision unfolds.
The first task for any decision maker, as Johnson et al. note, is to understand their current situation. How people achieve this understanding is an issue that appears in various guises across cognitive science. The “frame problem” in artificial intelligence research, the “correspondence problem” in judgement and decision-making and the “problem of induction” in philosophy all ask how agents are able to identify and act on relevant aspects of a situation. To adapt one of Johnson et al.'s examples: How does the widower determine the provenance of the noise that awoke him and decide what to do?
Johnson et al.'s claim appears to be that the first thing the widower does is to realise that he is under conditions of radical uncertainty in which probabilities are unknowable. How this initial step is achieved is not specified by the theory but it seems crucial if the purview is decision-making under radical uncertainty and not all decision-making, or indeed deliberative behaviour in general.
If narratives are only invoked when probabilities do not apply then the theory needs to explain how people know they are in such a situation. If the claim is broader and that narratives, or perhaps the internal monologues that accompany them, are used as a general guide to thinking and deciding – the “currency of thought” – then there is a deep concern that the theory loses coherence and becomes a set of generic statements about various tenuously related cognitive and emotional processes. There is clearly value in pointing out the relevance of other research areas to the endeavour of understanding behaviour that is often viewed through the narrow lens of expected utility theory (EUT), but to claim that this broadening of perspective constitutes a novel theory seems like overreach.
Returning to our widower, the idea that he explicitly runs a probability calculus over the possible outcomes of different actions is, naturally, absurd. He has most likely got to this stage in his life without ever hearing of EUT or its implications – just like the vast majority of people. As such, asking whether or how he knows that he is under conditions of radical uncertainty is also nonsensical. He has heard a sound and has to make a decision. Invoking EUT to explain how he does this might provide a way to think about and analyse the potential consequences of different actions but it is not a psychological explanation of his behaviour (and was never intended to be, e.g., by von Neumann & Morgernstern, Reference von Neumann and Morgernstern1947).
We agree with Johnson et al. that narratives provide another way to think about this process, but just as probability-based theories can be critiqued for not explaining “where the numbers come from,” narrative approaches, like CNT, suffer from under-specification of how narratives are constructed (e.g., Klayman, Reference Klayman2001; Newell, Lagnado, & Shanks, Reference Newell, Lagnado and Shanks2022). By focusing only on a narrow set of all possible representations (complex narratives), CNT is unable to account for when and why people might use simpler strategies to deal with problems. Johnson et al. claim that narratives are higher-order representations that flexibly include lower-order representations, but the threshold for lower versus higher-order representations remains murky, as does the mechanism for flexible integration. This is problematic because as long as we do not understand how the relevant narratives are themselves constructed (or indeed any representation, simple to complex), we also cannot know when and why people would rely on them to make choices.
To answer the when and why questions, we would typically turn to diagnostic empirical evidence but here again the picture is rather unclear. Johnson et al. have a somewhat conflicting view stating both that lab-experimentation is “ill-suited” for testing the prevalence of narrative thinking, and that “finer-grained experimental evidence” is required to draw more definitive conclusions than those derived on the basis of qualitative interviews.
One potential solution to these problems is to start more modestly and focus on how people build up simple representations. Not all situations necessitate complex, emotionally charged simulations and evaluations, but they always necessitate the development of some kind of representation. Considering people to be “intuitive scientists” puts the focus on the process of how representations are generated and not on a particular kind of complex representation (Szollosi & Newell, Reference Szollosi and Newell2020; Szollosi, Donkin, & Newell, Reference Szollosi, Donkin and Newell2022). From this perspective, understanding how people develop relatively simple representations – such as of integer numbers (Carey & Barner, Reference Carey and Barner2019) or of the frequency of various events (Mason, Szollosi, & Newell, Reference Mason, Szollosi and Newell2022) – can provide a useful starting point by enabling clearer ways to measure or manipulate the range of factors that Johnson et al. (rightly) claim play into any decision.
Explaining how such representations emerge does not require probabilities, or utilities, or the kinds of calculations that CNT wants to eschew, and we expect that focussing at this simpler level could eventually provide a window into how more complex representations (narratives) are built from primitive information. Without thinking carefully about these initial steps of flexible representation generation and selection there is a risk that the theoretical “tastiness” of CNT will be lost in a soup of conceptual complexity.
Financial support
Funding from the Australian Research Council (DP1901675) and from the EPSRC (EP/T033967/1) is gratefully acknowledged.
Competing interest
None.