Spatiotemporal Signatures of Surprise Captured by Magnetoencephalography

Abstract

Surprise and social influence are linked through several neuropsychological mechanisms. By garnering attention, causing arousal, and motivating engagement, surprise provides a context for effective or durable social influence. Attention to a surprising event motivates the formation of an explanation or updating of models, while high arousal experiences due to surprise promote memory formation. They both encourage engagement with the surprising event through efforts aimed at understanding the situation. By affecting the behavior of the individual or a social group via setting an attractive engagement context, surprise plays an important role in shaping personal and social change. Surprise is an outcome of the brain’s function in constantly anticipating the future of sensory inputs based on past experiences. When new sensory data is different from the brain’s predictions shaped by recent trends, distinct neural signals are generated to report this surprise. As a quantitative approach to modeling the generation of brain surprise, input stimuli containing surprising elements are employed in experiments such as oddball tasks during which brain activity is recorded. Although surprise has been well characterized in many studies, an information-theoretical model to describe and predict the surprise level of an external stimulus in the recorded MEG data has not been reported to date, and setting forth such a model is the main objective of this paper. Through mining trial-by-trial MEG data in an oddball task according to theoretical definitions of surprise, the proposed surprise decoding model employs the entire epoch of the brain response to a stimulus to measure surprise and assesses which collection of temporal/spatial components in the recorded data can provide optimal power for describing the brain’s surprise. We considered three different theoretical formulations for surprise assuming the brain acts as an ideal observer that calculates transition probabilities to estimate the generative distribution of the input. We found that middle temporal components and the right and left fronto-central regions offer the strongest power for decoding surprise. Our findings provide a practical and rigorous method for measuring the brain’s surprise, which can be employed in conjunction with behavioral data to evaluate the interactive and social effects of surprising events.

A taxonomy of surprise definitions

“…Importantly, most of the previous experimental studies have focused on one measure of surprise and its role and signatures in behavioral and physiological measurements. The examples that considered more than one surprise measure [Mars et al, 2008, Ostwald et al, 2012, Kolossa et al, 2015, Gijsen et al, 2021, Mousavi et al, 2022 have mainly focused on model-selection methods to compare different models and did not look for fundamentally different predictions of these measures -see Visalli et al [2021] for an exception. Even if two surprise measures are formally distinguishable, it may be that, in a given experimental set-up, the number of samples or effect size are not big enough to extract the quantitative differences between the two.…”

“…The Bayesian network [Barber, 2012] corresponding to the most general case of our generative model in Equation 1and Equation 2. The arrows show conditional dependence, the grey nodes show the hidden variables (C 1:t+1 and Θ 1:t+1 ), the red nodes show the observations (Y 1:t+1 ), and the blue nodes show the cue variables (X 1:t+1 1A [ Adams and MacKay, 2007, Fearnhead and Liu, 2007, Nassar et al, 2010, Wilson et al, 2013, Liakoni et al, 2021, C. generative model for modeling human inference about binary sequences in experiments like the one in Figure 1B [Meyniel et al, 2016, Maheu et al, 2019, Modirshanechi et al, 2019, Mousavi et al, 2022, Gijsen et al, 2021, D. generative model corresponding to variants of bandit and volatile bandit tasks like the one in Figure 1C [ Behrens et al, 2007, Findling et al, 2021, Horvath et al, 2021, where the cue variable Xt = At is a participant's action, and E. classic Markov Decision Processes (MDPs) to model experiments like the one in Figure 1D [Sutton and Barto, 2018, Schultz et al, 1997, Gläscher et al, 2010, Daw et al, 2011, Huys et al, 2015, Lehmann et al, 2019, where the cue variable Xt = (A t−1 , Y t−1 ) consists of previous action and observation. See subsection 2.2 for details.…”

“…Shannon surprise, also known as surprisal [Barto et al, 2013], is a way to formalize this concept of surprise. It comes from the field of information theory [Shannon, 1948] and statistical physics [Tribus, 1961] and is widely used in neuroscience [Mars et al, 2008, Kopp and Lange, 2013, Kolossa et al, 2015, Konovalov and Krajbich, 2018, Meyniel et al, 2016, Modirshanechi et al, 2019, Maheu et al, 2019, Mousavi et al, 2022, Gijsen et al, 2021, Visalli et al, 2021.…”

“…The first definition is more consistent with the original definition of the Bayesian surprise [Baldi, 2002, Baldi and Itti, 2010, Schmidhuber, 2010 applied to our generative model because the belief before the observation should include the knowledge that the environment is volatile. However, the second definition looks more intuitive from the neuroscience perspective [Gijsen et al, 2021, Mousavi et al, 2022. Note that, in Equation 29 and Equation 30, the observation y t+1 does not appear explicitly on the right hand side; the observation has, however, influenced the update of the belief to its new distribution π (t+1) .…”

“…While our proposed taxonomy is solely conceptual and based on the theoretical properties of different definitions, we note that there have been a significant number of studies investigating the neural and physiological correlates of prediction [Mars et al, 2008, Kopp and Lange, 2013, Kolossa et al, 2015, Modirshanechi et al, 2019, Gijsen et al, 2021, Maheu et al, 2019, Meyniel, 2020, Mousavi et al, 2022, Konovalov and Krajbich, 2018, Loued-Khenissi and Preuschoff, 2020, Gläscher et al, 2010, change-point detection [Nassar et al, 2012, Xu et al, 2021, Liakoni et al, 2022, confidence correction [Gijsen et al, 2021], and information gain [Ostwald et al, 2012, Kolossa et al, 2015, Gijsen et al, 2021, Visalli et al, 2021, Nour et al, 2018, O'Reilly et al, 2013 surprise measures (Figure 1). We, therefore, speculate that at least one measure from each of these categories is computed in the brain but potentially through different neural pathways and to be used for different brain functions.…”