Проблемы с выбором ROI при экспериментальном сравнении предсказаний теорий сознания

Как уже сообщалось, первые результаты крупномасштабного проекта по экспериментальной проверке предсказаний наиболее влиятельных нейротеорий сознания оказались не очень впечатляющими (см. препринт и рассказ Ивана Иванчея о презентации этих результатов на конференции Association for the Scientific Studies of Consciousness). Однако обнаружились и более крупные проблемы. По большей части они связаны с выбором ROI (regions of interest) -- тех областей мозга, активность которых анализировалась. Наверное, не нужно объяснять, почему выбор ROI абсолютно критичен для результатов анализа. Однако в этом исследовании, как оказывается, он был весьма слабо обоснован, а местами еще и проводился после начала или даже завершения сбора данных, что уже совсем странно, так как исследование подается как пререгистрированное.

<...> Although the study was overall pre-registered, these critical and surprising ROI definitions were only uploaded to OSF in Dec 2022- by which point data collection was probably already completed or near completion- and made public as recently as late June 2023.

<...> for IIT, the selection of the ROIs were neither based on the mathematical formalisms of the theory (calculation of ‘phi’) nor previous empirical studies that helped to give rise of the notion of posterior ‘hot zone’ in the first instance, e.g. Siclari et al 2017 (5). Instead, curiously they look like a set of regions which anyone would have chosen in order to maximize decoding accuracy for classifying visual stimuli. <....> Notably, these ROIs also exclude the precuneus and posterior anterior cingulate which were supposed to be part of the posterior hot zone, based on previous publications from some of the authors themselves (5).

<...> For GNWT, not including parietal areas also went against how the theory has been characterized in print. I understand that the intention was to focus on the prefrontal cortex so as to allow the two theories to make different predictions. But logically this seems backwards: if the two theories are indeed empirically indistinguishable, perhaps we should just accept that fact. We can’t instead contrive to modify a theory so substantially just for this ad hoc purpose, and still convincingly call it GNWT. Relatedly, these important ROI definitions also seem to be pre-registered relatively late in the project (Dec 2022, version 4 of their OSF upload), and made public even later (late June 2023).

И т.д., и т.п. ...

Hakwan Lau. Where is the ‘posterior hot zone’? Open Review of Ferrante et al (2023): “An Adversarial Collaboration to Critically Evaluate Theories of Consciousness” (by the Cogitate Consortium). PsyArXiv, August 18, 2023 https://doi.org/10.31234/osf.io/93ufe

(по наводке Мих. Лебедева)