Уважаемые коллеги, доброго времени суток! Представляем вам британское научное издание Behavioral and Brain Sciences. Журнал имеет третий квартиль, издаётся в Cambridge University Press, его SJR за 2022 г. равен 0,505, импакт-фактор 29,3, печатный ISSN - 1469-1825, электронный - 0140-525X, предметные области - Поведенческая неврология, Физиология, Нейропсихология и физиологическая психология. Вот так выглядит обложка:
Здесь два редактора - Пол Блум, контактные данные - paul.bloom@yale.edu
и Барбара Финли - barbara.finlay@cornell.edu.
Дополнительный публикационный контакт - bbsjournal@cambridge.org.
Это всемирно известный журнал с инновационным форматом, известным как Open Peer Commentary. Публикуются особо значимые и противоречивые работы исследователей в любой области психологии, нейробиологии, поведенческой биологии или когнитивных наук, а также 20-40 комментариев к каждой статье от специалистов в рамках этих дисциплин и за их пределами, плюс авторские ответы на них. Результатом является увлекательный и уникальный форум для общения, критики, стимулирования и, в частности, объединения исследований в области поведенческих наук и наук о мозге - от молекулярной нейробиологии до искусственного интеллекта и философии разума.
Адрес издания - https://www.cambridge.org/core/journals/behavioral-and-brain-sciences#
Пример статьи, название - Decisions under uncertainty are more messy than they seem. Заголовок (Abstract) - Conviction Narrative Theory (CNT) is conceptually so multifaceted as to make critical evaluation difficult. It also omits one course of action: Active engagement with the world. Parsing the developmental and mechanistic processes within CNT would allow for a rigorous research programme to put the account under test. I propose a unifying account based on active inference.
Behaviour under uncertainty is messier than proposed, and while laudable in its breadth, Conviction Narrative Theory (CNT) risks specificity for pluralism. Indeed, narratives themselves are presented in the target article as multifaceted, enveloping the functions of structure, simulation, affect-evaluation and communication. This fuzziness allows it to include a broad variety of cognitive functions and mechanisms, which while useful for more general descriptions, also makes its strengths and weaknesses as an explanation difficult to evaluate in detail. I propose that in addition to relying on narratives in situations of radical uncertainty we also have another option: Random exploration by acting on the world.
Consider, for example, an infant who in a situation of truly radical uncertainty (i.e., a situation with very shallow schemas, limited priors or disjointed narrative structure) acquaints with the world by putting things in their mouth, by wiggling and whacking the world. This is an example of model-free learning, where in novel situations we prod and poke to reveal the structure of the situation and possibly to alter it into a more familiar choice set. Only after we have inferred some structure to guide our predictions do we begin to form more complex model-based predictions to guide our actions (Castro-Rodrigues et al., Reference Castro-Rodrigues, Akam, Snorasson, Camachi, Paixão, Maia and Oliveira-Maia2022). Now it is not necessary that the inferred structure corresponds to any so-called ground truth, merely that it provides enough structure for action, that is, the reduction of uncertainty may be illusory. Alternatively, we may uncover this structure by asking others, of using others' representations as scaffolding for one's own.
Behind both attempts is the aim to reduce the uncertainty by acting on the world or by updating our internal model. Behind a variety of such uncertainty (or free energy) minimization accounts lies the free-energy principle (FEP, Friston, Reference Friston2010), a broad formalization which arguably also engenders decision accounts from expected utility to softmax accounts (Friston et al., Reference Friston, Schwartenbeck, FitzGerald, Moutoussis, Behrens and Dolan2014). It proposes predictions to be hierarchically structured (Friston, Reference Friston2010) with the task of the organism to reduce uncertainty thus maintaining the environmental fit of their “generative model.” At the higher hierarchical end, predictions take the form of beliefs, with narratives as socially reinforced and integrated belief landscapes. These belief landscapes compete for valuation in a multidimensional matrix (for a main valuation account, see Sharot, Rollwage, Sunstein, & Fleming, Reference Sharot, Rollwage, Sunstein and Fleming2022). A mismatch in the belief and the actual state of the world calls for one of two lines of action: Revising and updating the internal generative model; or active inference. Active inference refers to action selection as a process of imposing structure on the generative model organized from sensory data (Friston, Da Costa, Hafner, Hesp, & Parr, Reference Friston, Da Costa, Hafner, Hesp and Parr2021). From this point of view, CNT would fit the FEP fold. The utility of such an account is singularly to reduce uncertainty in a fashion that can be approximated by Bayesian methods yet account for bounded rationality. FEP accounts sidestep the critique provided by the authors against Bayesian accounts that in situations of radical uncertainty priors cannot be set. Considering CNT a higher order method for uncertainty reduction, that is, as a way to increase the model fit of the environment and the generative model, the problem is removed: From the first-person view of the generative model, we can impose a model on the situation, even if its initial first-iteration fit were poor; to launch the competition for the best fit. After all, at some point, we stop dealing with uncertainty by putting things in our mouth for inquiry and utilize other methods.
CNT is praiseworthy in its scope and explanatory breadth. However, there are several avenues to begin clarifying these functions and the mechanisms. It is noteworthy that if were to replace the term “narrative” with that of “reflective consciousness” in the definition and describe it as:
…structured, higher-order mental representations incorporating causal, temporal, analogical, and valence information about agents and events, which serve to explain data, imagine and evaluate possible futures, and motivate and support action over time (target article, Table 1 and sect. 5, para. 3)
…we would not be that far removed from various accounts of consciousness overall. For example, biological realism considers our conscious experience as a spatiotemporal virtual reality, a simulation of the world beyond our senses (Revonsuo, Reference Revonsuo2006).
With such limitations in mind, I propose CNT would gain in explanatory depth by integrating a developmentally informed account of how we move from model-free learning to narratives under uncertainty. One could make the claim that only once the social complexity increases and our mental representations acquire complexity probabilistic reasoning becomes both limited and a resource strain – and when active engagement fails – do we switch to narratives, again unified under the function of uncertainty reduction and thus formalizable (for children faring well in probabilistic reasoning see Girotto & Gonzalez, Reference Girotto and Gonzalez2008; Riggs, Reference Riggs2019).
A more thorough analysis would illuminate how the proposed mechanism functions: What is necessary and what is sufficient for CNT and its subcomponent dynamics. For example, it is suggested we simulate the structure of the mental representation of reality already before birth (Hobson, Reference Hobson2009), whereas narrative communication develops later as it requires theory-of-mind, and causal and temporal reasoning abilities (Stadler & Ward, Reference Stadler and Ward2005). Before such parsing, CNT is at risk of falling into the same mechanistic trap it sees as the demise of competing theories with the broader terms doing a lot of work.