Conference Agenda

The visual awareness negativity (VAN) has been proposed as a neuronal correlate of consciousness (NCC). It is typically obtained by comparing trials in which a stimulus was perceived with trials in which it was not. However, if the VAN represents a true NCC, it should be detectable in conditions where participants consciously perceive stimuli despite their physical absence, such as during visual illusions. Our study (N = 60) aimed to test this hypothesis by employing the sound-induced flash illusion (SIFI). In SIFI, a single briefly flashed white disk is paired with two brief sounds, generating the illusion of a second flash. This paradigm allows for investigating event-related potentials (ERPs) associated with visual illusions, as they coincide with a precisely timed auditory stimulus, enabling the inference of the illusory percept's onset. We employed a modified perceptual awareness scale (PAS) to obtain trial-by-trial reports of subjective perception. ERPs were analyzed by comparing trials with and without illusory percepts under the same physical conditions. We observed an enhanced negativity over occipital-parietal electrodes in the typical VAN window (250 to 300 ms). Our findings suggest that not only veridical perceptions enhance negative potentials over posterior areas in this time window but also illusory percepts, fostering the interpretation of the VAN as an NCC. We suggest that to be considered a neuronal correlate of consciousness, a potential candidate must be detectable not only through the traditional hit-versus-miss comparison but also within illusion paradigms such as the one utilized in this study.

We can rapidly learn repeating patterns in our environment. These learned patterns are often used to form expectations about future sensory events. Several influential predictive coding models posit that stimulus-evoked neural responses in the visual system are reduced when an expected stimulus appears (expectation suppression). However, there is currently scant electrophysiological evidence for genuine expectation suppression in the visual system when relevant confounds are taken into account. This empirical evidence is critical for constraining and developing predictive processing models of conscious experience. To test for expectation suppression we performed three predictive cueing experiments (n=48, n=48, n=60) while recording electroencephalography. Participants learned cue-stimulus associations during a training session and were then exposed to the same cue-stimulus pairs in a subsequent experiment. Experiment 1 presented faces, whereas experiments 2 and 3 presented oriented gratings. Participants viewed a cue followed by an expected or unexpected S1 stimulus. They were then required to report whether that stimulus matched a subsequently presented S2 stimulus. In experiment 3, noise was added to the S1 grating stimuli so that participants heavily relied on the predictive cues while performing the task. Across the three experiments we did not find evidence that expectations influenced event-related potentials in the first 300ms after S1 stimulus onset (i.e., during afferent visual responses). Bayes factors generally favoured the null hypothesis during the peristimulus time window. Our findings do not support predictive coding-based accounts that specify reduced prediction error signalling when perceptual expectations are fulfilled.

Recent experiments demonstrated that instrumental conditioning and responding to previously learned stimuli cannot take place in the absence of stimulus awareness (Skora et al., 2023, 2024). These findings prompt a re-evaluation of older empirical work, which reported motivational effects in response to monetary incentives, even for unreportable stimuli (Pessiglione et al., 2007). These effects may have resulted from masking techniques allowing for residual stimulus awareness, potentially distorting the study’s findings.

We revisited the original paradigm with more rigorous methods aimed at controlling for total absence of stimulus awareness, combining minimal exposure durations with subjective trial-by-trial awareness ratings: employing very brief presentation times in the µs-range renders a stimulus invisible to the participant without having to rely on potentially confounding masking techniques (Lanfranco et al., 2024). Meanwhile, awareness ratings allow to strictly divide trials into below- and above detection threshold.

Across presentation times and awareness ratings, we compared behavioural (physical force exerted on a hand-dynamometer) and electrophysiological measures (EEG components previously implicated in reward processing) between high- and low-reward trials using Bayesian analysis methods. We found no evidence of motivational processing in trials where stimulus awareness was absent.

Our findings suggest that monetary rewards can influence behaviour only if the subject is at least partially aware of them, challenging previous claims about unconscious motivation. Our results underscore the critical role of conscious perception in reward processing, highlighting how crucial it is to rigorously ensure the absence of stimulus awareness in experimental paradigms investigating potentially unconscious processes.

Imagine deciding if a mosquito is present before going to sleep—are you sure that nothing is there? In many contexts, we need to be confident in our inferences about the presence or absence of multisensory stimuli. Here we sought to characterize how humans form such confidence judgments. Participants performed a detection task of audiovisual, visual, and auditory stimuli, at unimodal detection thresholds intensities. They indicated whether a stimulus was present or absent (irrespective of modality) before reporting their amodal confidence on a continuous scale. Finally, they provided modality-specific detection judgments and confidence on a two-dimensional, audiovisual, scale. In two pre-registered experiments (N = 47 and N = 52), participants detected audiovisual stimuli better than unimodal ones, and provided confidence ratings that better tracked objective accuracy following audiovisual than unimodal stimuli. Surprisingly, confidence judgments were both higher and more aligned with objective accuracy following absence than presence judgments. To fit these data, we extended a recent ideal-observer model of visual detection to multimodal detection. The model assumes that only evidence for presence is accumulated, while absence is inferred from counterfactual detectability (i.e., “I would have perceived it if it was present”). Our model successfully reproduced amodal and modality-specific detection decisions, and, despite being fitted to decision and decision time data alone, captured population variability in modality-specific confidence ratings. The model however failed to account for amodal confidence. This suggests that sensory evidence from different modalities may be integrated differently when making detection judgments and when monitoring the accuracy of such judgments.

Human decision-making often occurs in noisy environments, yet individuals can still gain metacognitive insight into their performance even when outcomes are difficult to predict. In this study, we investigated computational and physiological indices of learning that contribute to metacognition. Specifically, we tracked pupil responses to stimuli, choices, and outcomes when participants were unaware of the environmental regularities. Additionally, we collected confidence reports.

In Experiment 1, participants (N=39) learned to predict probabilistic outcomes from combinations of four cues. Experiment 2 (N=20) used the same task with confidence reports. Experiment 3 employed a standard reinforcement learning task, where participants (N=53) chose between two stimuli with different reward probabilities and reported their confidence. Across all experiments, computational models quantified prediction errors and stimulus values.

Experiment 1 showed that pupil responses were sensitive to implicit surprise, but this sensitivity diminished once participants became aware of the environmental regularities. Experiments 2 and 3 demonstrated that surprise also influenced confidence reports—specifically, confidence levels were higher following positive prediction errors on the preceding trial.

These findings contribute to ongoing discussions on reinforcement learning and human cognition (Chew et al., 2021). Our data also support the dissociation between metacognition and performance (Shekhar & Rahnev, 2021). It also demonstrates separate mechanisms of conscious and unconscious learning. Taken together, these results provide new insights into the relationship between performance, metacognition, and physiological arousal markers. A critical next step is to examine how these indices relate to the subjective representation of learning variables (e.g., computational phenomenology; Palminteri & Cecchi, 2023).