Patxi Elosegi

Absolutely incredible work led by @patxelos.bsky.social and @dsotob.bsky.social - such a compelling case of unconscious perception! Privileged to have been a little part of this 🧠

Our #sEEG study is now published in Nature Communications: rdcu.be/eIkoG! 🧠
Key finding: We discovered neural evidence accumulation for visual perception that's independent of report preparation—recorded from >3000 channels across 3 experiments!
#Neuroscience #Consciousness #OpenAccess

Preprint: osf.io/preprints/ps... @matthiasmichel.bsky.social @meganakpeters.bsky.social @smfleming.bsky.social @axc.bsky.social @liadmudrik.bsky.social

Finally, huge thanks to @pietroamerio.bsky.social for outstanding modeling work, and to the incomparable @dsotob.bsky.social for his mentorship throughout this project and my PhD.

Methods note: we used Bayesian optimization to titrate exposure and eccentricity to maximize the chance to dissociate perception from awareness. We also analyze spatial anisotropies and provide extensive convergent tests of the effect.

12/ The ensemble blindsight effect reported here shows that other more ecologically valid perceptual processes may be more suitable candidates to isolate behavioral traces of unconscious perception and thereby characterize the functional scope of unconscious processing.

11/ We've been stubbornly trying to find evidence of unconscious perception using single, highly degraded objects. When response criterion is controlled, results are consistently null. This doesn’t refute unconscious perception; it just shows these paradigms rarely reveal it.

10/ In contrast, the brief, masked signal from a single small object (e.g., tilted Gabor patch) could be diluted or overwritten during the feedforward pass by the more stable spatiotemporal statistics of the surrounding display, limiting its availability to unconscious processing

9/Why were ensembles successful in achieving a dissociation? We argue that the spatial redundancy of ensembles lets the system pool many noisy measurements, cancel uncorrelated noise, and yield a stronger unconscious signal.

8/ To formalize the effect, we build Bayesian ideal-observer models to test whether the detection shortfall could be mere decision-making inefficiency. We find awareness is systematically more suboptimal than ensmeble discrimination beyond expected by task-difficulty differences

7/However, 2-IFC task introduces an assymetry in the amount of the information that the perceptual and awareness test have to consider. In Exp.2. we use a single interval paradigm where both tasks are matched in the amount of information and we replicated all findings from Exp.1.

6/ In Exp.1. we find a clear dissociation between ensemble percetion and awareness. We demonstrate that this effect was not influenced by task-order, target interval order or an unstable detection criterion. We provide convergent analyses proving the robustness of the effect.

5/We tackle both issues with ensemble perception in a 2-IFC paradigm. Ensemble summaries test unconscious perception in richer, natural displays and boost unconscious sensitivity via the brain’s statistical pooling without using masking other signal degradation techniques.

4/ Criterion-content fallacy (Michel, 2023): if the awareness test relies on a different feature than the perceptual task, unconscious effects may be underestimated. This hampered prior work; even with closer matching, clear evidence has been scarce (e.g., Amerio et al., 2024).

3/ Subjective criterion problem: people may report “no experience” if sensory evidence falls below a subjective threshold. A remedy is bias-free designs like two-interval forced choice. Yet prior 2-IFC studies (e.g., Peters & Lau, 2015) found no unconscious perception.

2/ Interest in unconscious perception surged through the 2023, then cooled. One reason: despite many attempts, robust evidence remains elusive once controlling for response-criterion issues—both the subjective criterion and the criterion-content fallacy.

1/Preprint Alert🔔: Across two experiments plus a computational model, we show the visual system compresses complex scenes into summary statistics that can guide behavior without conscious access to the task-defining features. We term this the Ensemble Blindsight effect.

Large language models surpass human experts in predicting... Scientific discoveries often hinge on synthesizing decades of research, a task that potentially outstrips human information processing capacities. Large language models (LLMs) offer a solution....

"Large language models surpass human experts in predicting neuroscience results" w @ken-lxl.bsky.social and braingpt.org. LLMs integrate a noisy yet interrelated scientific literature to forecast outcomes. arxiv.org/abs/2403.03230 1/6