Angus Leung

Freud’s iceberg metaphor has long been used to describe (un)consciousness: a huge unconsciousness supports a tiny tip of consciousness. Inspired by this, research has assumed that unconscious discriminative processing persists into consciousness, employing contrastive analysis to isolate the neural correlates of consciousness (NCC). For example, in electroencephalography (EEG) studies, features of time series are related to unconscious sensory processes if they discriminate presence of invisible visual stimuli. According to the consistency assumption, they should also discriminate these stimuli when they become visible and task relevant. Despite its prevalence, the assumption has never been directly questioned, and its general validity for EEG studies, regardless of the exact analysis method employed, is unclear. This registered report aims to explicitly test this assumption by utilising a toolbox providing over 7000 time-series features from various analysis methods from a wide range of research fields. We apply these features to an inattentional blindness paradigm and identify features which discriminate presence of a visual stimulus which is initially invisible, and then later visible and task relevant. We track the performance of each feature in discriminating the presence of the stimulus as it varies in visibility, and use a Bayesian framework to test whether the consistency assumption is likely to be valid. Pilot results suggest it is likely invalid across the range of time-series features.

Despite the widespread application of general anesthesia since the 19th century, a fundamental understanding of its operating principle and a reliable metric of consciousness remains elusive. We here analyze Loss of Responsiveness (LoR) induced by propofol anesthesia in humans and macaque monkeys. Through application of “highly comparable time series analysis” (hctsa) over 7000 features, we aim to identify LoR metrics common to both humans and macaque monkeys. In Stage 1 analysis, we trained a nearest median classifier using 200 ms data epochs using one macaque individual, and evaluated its classification accuracy on human intracranial data, recorded from one participant. This analysis suggested the existence of univariate features capable of discriminating between awake and propofol-induced unresponsive states in both species. Stage 2 extended the analysis to additional macaque and human recordings to evaluate cross-session and cross-species generalization. We identified many hctsa features that commonly distinguished awake from unresponsive states across sessions and species, in the frontal, but not in parietal and temporal, recordings. In exploratory analyses, we aimed to distinguish loss of responsiveness from consciousness, by additionally analyzing human data under sedated yet conscious state in a within-subject manner. This analysis revealed that frontal, temporal and parietal recordings contain similarly sensitive features and channels as signatures of consciousness. These results support the existence of ECoG time-series features associated with propofol-induced loss of responsiveness across humans and macaques, while also highlighting the challenge to determine whether such features specifically track conscious experience rather than responsiveness.

Dream experiences during sleep provide an excellent opportunity to investigate the neural basis of consciousness. Previous studies have examined the neural correlates of dreaming using electroencephalography (EEG), focusing on spectral power in specific frequency bands. However, findings have been inconsistent, particularly regarding the spatiotemporal characteristics of these correlates, raising questions about whether spectral power in specific brain regions and frequency bands reliably reflects dreaming. In this study, we investigate whether neural correlates reported in a highly influential EEG study generalize to consciousness during evening naps, measured using magnetoencephalography (MEG), which offers more accurate source reconstruction than EEG. In Stage 1, the data acquisition team completed data collection, obtaining MEG data and subjective reports using a targeted serial-awakening paradigm. We test hypotheses regarding the relationship between low- and high-frequency neural power in frontal and posterior regions and conscious experience. We defined data inclusion and exclusion criteria and established pre-specified analytical procedures based on data from two participants. The pilot results from one participant were consistent with the hypotheses concerning low-frequency power across regions, but not with those related to high-frequency power. The other participant provided insufficient evidence for all hypotheses. In Stage 2, an independent analysis team will conduct group-level hypothesis testing according to the pre-specified procedure, using data from the remaining unanalyzed participants (N = 10) who meet the inclusion criteria.

The neural mechanisms of consciousness remain elusive. Previous studies on both human and non-human animals, through manipulation of level of conscious arousal, have reported that specific time-series features correlate with level of consciousness, such as spectral power in certain frequency bands. However, such features often lack principled, theoretical justifications as to why they should be related with level of consciousness. This raises two significant issues: firstly, many other types of times-series features which could also reflect conscious level have been ignored due to researcher biases toward specific analyses; and secondly, it is unclear how to interpret identified features to understand the neural activity underlying consciousness, especially when they are identified from recordings which summate activity across large areas such as electroencephalographic recordings. To address the first concern, here we propose a new approach: in the absence of any theoretical priors, we should be maximally agnostic and treat as many known features as feasible as equally promising candidates. To apply this approach, we use highly comparative time-series analysis (hctsa), a toolbox which provides over 7,700 different univariate time-series features originating from different research fields. To address the second issue, we employ hctsa to high-quality neural recordings from a relatively simple brain, the fly brain (Drosophila melanogaster), extracting features from local field potentials during wakefulness, general anesthesia, and sleep. At Stage 1 of this registered report, we constructed a classifier for each feature, for discriminating wakefulness and anesthesia in a discovery group of flies (N = 13). At Stage 2, we assessed their performances on four independent groups of evaluation flies, from which recordings were made during anesthesia and sleep, and which were originally blinded to the data analysis team (N = 49). We found only 47 time-series features, applied to recordings obtained from the center of the fly brain, to also significantly classify wake from anesthesia or sleep in all 4 of these evaluation datasets. Most of these were related to autocorrelation, and they indicated that signals during wakefulness remained correlated to their past for a longer timescale than during anesthesia and sleep. Meanwhile, time-series features related to well-known potential markers of consciousness, such as those related to complexity or spectral power, failed to generalize across all the flies. However, many of these complexity and spectral features have a consistent direction of effect due to anesthesia or sleep across flies, suggesting that even slight variations in experiment setup can reduce generalizability of classifiers. These results caution the current state of frequent discoveries of new potential consciousness markers, which may not generalize across datasets, and point to autocorrelation as a class of dynamical properties which does.

The neural mechanisms of consciousness remain elusive. Previous studies on both human andnon-human animals, through manipulation of level of conscious arousal, have reported thatspecific time-series features correlate with level of consciousness, such as spectral power incertain frequency bands. However, such features often lack principled, theoreticaljustifications as to why they should be related with level of consciousness. This raises twosignificant issues: firstly, many other types of times-series features which could also reflectconscious level have been ignored due to researcher biases towards specific analyses; andsecondly, it is unclear how to interpret identified features to understand the neural activityunderlying consciousness, especially when they are identified from recordings whichsummate activity across large areas such as electroencephalographic recordings. To addressthe first concern, here we propose a new approach: in the absence of any theoretical priors,we should be maximally agnostic and treat as many known features as feasible as equallypromising candidates. To apply this approach we use highly comparative time-series analysis(hctsa), a toolbox which provides over 7,700 different univariate time-series featuresoriginating from different research fields. To address the second issue, we employ hctsa tohigh-quality neural recordings from a relatively simple brain, the fly brain (Drosophilamelanogaster), extracting features from local field potentials during wakefulness, generalanesthesia and sleep. At Stage 1 of this registered report, we constructed a classifier for eachfeature, for discriminating wakefulness and anesthesia in a discovery group of flies (N = 13).At Stage 2, we assessed their performances on four independent groups of evaluation flies,from which recordings were made during anesthesia and sleep, and which were originallyblinded to the data analysis team (N = 49). We found only 47 time-series features, applied torecordings obtained from the center of the fly brain, to also significantly classify wake fromanesthesia or sleep in all 4 of these evaluation datasets. Most of these were related toautocorrelation, and they indicated that signals during wakefulness remained correlated totheir past for a longer timescale than during anesthesia and sleep. Meanwhile, time-seriesfeatures related to well-known potential markers of consciousness, such as those related tocomplexity or spectral power, failed to generalize across all the flies. However, many of thesecomplexity and spectral features have a consistent direction of effect due to anesthesia orsleep across flies, suggesting that even slight variations in experiment setup can reducegeneralizability of classifiers. These results caution the current state of frequent discoveriesof new potential consciousness markers, which may not generalize across datasets, and pointto autocorrelation as a class of dynamical properties which does.

Mediano et al. (1. Mediano 2022 TICS) separate out strong and weak flavours of the integrated information theory (IIT) of consciousness. They describe “strong IIT” as attempting to derive a universal formula for consciousness, and “weak IIT” as searching for empirically measurable correlates of aspects of consciousness. We put forward that their overall notion of “weak IIT” may be too weak. Rather, it should be separated out to distinguish “aspirational IIT”, which aims to empirically test IIT by making tradeoffs to its proposed measures, and “IIT inspired” approaches, which adopt high level ideas of IIT while dropping the mathematical framework it reaches through its introspective, first-principles approach to consciousness.

Mediano et al. (The strength of weak integrated information theory. Trends Cogn Sci 2022;26: 646–55.) separate out strong and weak flavours of the integrated information theory (IIT) of consciousness. They describe ‘strong IIT’ as attempting to derive a universal formula for consciousness and ‘weak IIT’ as searching for empirically measurable correlates of aspects of consciousness. We put forward that their overall notion of ‘weak IIT’ may be too weak. Rather, it should be separated out to distinguish ‘aspirational-IIT’, which aims to empirically test IIT by making trade-offs to its proposed measures, and ‘IIT-inspired’ approaches, which adopt high-level ideas of IIT while dropping the mathematical framework it reaches through its introspective, first-principles approach to consciousness.

How a system generates conscious experience remains an elusive question. One approach towards answering this is to consider the information available in the system from the perspective of the system itself. Integrated information theory (IIT) proposes a measure to capture this integrated information (Φ). While Φ can be computed at any spatiotemporal scale, IIT posits that it be applied at the scale at which the measure is maximised. Importantly, Φ in conscious systems should emerge to be maximal not at the smallest spatiotemporal scale, but at some macro scale where system elements or timesteps are grouped into larger elements or timesteps. Emergence in this sense has been demonstrated in simple example systems composed of logic gates, but it remains unclear whether it occurs in real neural recordings which are generally continuous and noisy. Here we first utilise a computational model to confirm that Φ becomes maximal at the temporal scales underlying its generative mechanisms. Second, we search for emergence in local field potentials from the fly brain recorded during wakefulness and anaesthesia, finding that normalised Φ (wake/anaesthesia), but not raw Φ values, peaks at 5 ms. Lastly, we extend our model to investigate why raw Φ values themselves did not peak. This work extends the application of Φ to simple artificial systems consisting of logic gates towards searching for emergence of a macro spatiotemporal scale in real neural systems.

How a system generates conscious experience remains an elusive question. One approach towards answering this is to consider the information available in the system from the perspective of the system itself. Integrated information theory (IIT) proposes a measure to capture this, integrated information (Φ). While Φ can be computed at any spatiotemporal scale, IIT posits that it be applied at the scale at which the measure is maximised. Importantly, Φ should emerge to be maximal not at the smallest spatiotemporal scale, but at some macro scale where system elements or timesteps are grouped into larger elements or timesteps. Emergence in this sense has been demonstrated in artificial binary systems, but it remains unclear whether it occurs in real neural recordings which are generally continuous and noisy. Here we first utilise a computational model to confirm that Φ becomes maximal at the temporal scales underlying its generative mechanisms. Second, we search for emergence in local field potentials from the fly brain recorded during wakefulness and anaesthesia, finding that normalised Φ (wake/anaesthesia), but not raw Φ values, peaks at 5 ms. Lastly, we extend our model to investigate why raw Φ values themselves did not peak. This work extends the application of Φ to simple artificial systems consisting of logic gates towards searching for emergence of a macro spatiotemporal scale in real neural systems.

The physical basis of consciousness remains one of the most elusive concepts in current science. One influential conjecture is that consciousness is to do with some form of causality, measurable through information. The integrated information theory of consciousness (IIT) proposes that conscious experience, filled with rich and specific content, corresponds directly to a hierarchically organised, irreducible pattern of causal interactions; i.e. an integrated informational structure among elements of a system. Here, we tested this conjecture in a simple biological system (fruit flies), estimating the information structure of the system during wakefulness and general anesthesia. Consistent with this conjecture, we found that integrated interactions among populations of neurons during wakefulness collapsed to isolated clusters of interactions during anesthesia. We used classification analysis to quantify the accuracy of discrimination between wakeful and anesthetised states, and found that informational structures inferred conscious states with greater accuracy than a scalar summary of the structure, a measure which is generally championed as the main measure of IIT. In stark contrast to a view which assumes feedforward architecture for insect brains, especially fly visual systems, we found rich information structures, which cannot arise from purely feedforward systems, occurred across the fly brain. Further, these information structures collapsed uniformly across the brain during anesthesia. Our results speak to the potential utility of the novel concept of an “informational structure” as a measure for level of consciousness, above and beyond simple scalar values.