Belief loads of assumptions impact brain networks underlying logical reasoning: A machine learning approach

Prior knowledge and beliefs influence our reasoning in daily life and may lead us to draw unwarranted conclusions with undesirable consequences. The underlying neural correlates of the interaction between belief and logic, prior to making logical decisions, are largely unknown. In this study, we aimed to identify brain regions important in distinguishing belief load of assumptions in logical decision making. Thirty-one healthy volunteers (18-29 years old) participated in an fMRI study and were asked to respond to a series of syllogistic arguments in which assumptions were either congruent (believable) or incongruent (unbelievable) with common knowledge. An interpretable machine learning algorithm, an L1 regularized Support Vector Machine, was used to explain the discriminatory pattern of conditions given the brain activation patterns. Behavioral results confirmed that believable premises were incorrectly endorsed more than unbelievable ones. Imaging results revealed that several connectivity patterns anchored around the insula, amygdala, and IFG were deemed to be important in distinguishing believable from unbelievable assumptions at different time points preceding logical decision. Our behavioral and imaging results convergingly suggest the importance of assumptions’ belief loads in the logical decision. Our results provide new insights into the interaction between belief and logic systems and have important practical implications for social, complex decisions.