arXiv (Cornell University)
May 27, 2022
Aapo Hyvärinen
3 citations
Suffering arises from frustration, defined as the failure to achieve a desired goal or reward, which is inevitable due to the world's complexity, limited computational resources, and scarcity of good data. Both humans and AI agents process information to pursue goals, making AI a useful model for the human mind. Frustration itself serves as an error signal for learning and adaptation. The book examines learning algorithms and their limitations to explain suffering, then derives interventions—such as mindfulness meditation—that reduce frustration, expressed by a simple equation. These interventions align with Buddhist and Stoic philosophy, offering a computational justification for their effectiveness.
arXiv (Cornell University)
November 29, 2024
Claudio Agnorelli, Meg J. Spriggs, Kate Godfrey et al.
2 citations
Classic psychedelics (LSD, psilocybin, N,N-DMT) and non-classic psychedelics (ketamine, MDMA) enhance neuroplasticity, the nervous system's ability to adapt. Animal studies indicate these drugs induce meta-plasticity, heightening sensitivity to environmental stimuli, and hyper-plasticity, reopening developmental windows for long-term structural changes that affect mood and behavior. Translating these findings to humans is challenged by limitations in current imaging techniques, but emerging approaches like novel PET radioligands, non-invasive brain stimulation, and multimodal methods offer promising directions. This review informs development of targeted interventions for neuropsychiatric disorders and advances understanding of psychedelics' therapeutic potential.
arXiv (Cornell University)
October 10, 2019
Camile Bahi
2 citations
Depression and anxiety are common in cancer patients, and standard antidepressants have not proven effective for this distress compared to placebo. A systematic review and meta-analysis of four randomized studies involving 105 patients found that psilocybin, a serotoninergic hallucinogen, significantly reduced cancer-related depression and anxiety compared to placebo, as measured by pooled Peto odds ratios. The substance appeared safe for these patients. The review also identified surprising psychological mechanisms that may underlie the effects. Psilocybin-based therapy shows potential as a treatment for cancer-related distress, but future research should confirm these findings in larger populations and explore application to non-hospitalized patients.
arXiv (Cornell University)
June 4, 2026
Anna Mikeda
A new precautionary framework translates evidence about whether AI systems might be conscious into graduated protective obligations. The framework maps five welfare-relevant dimensions—phenomenal consciousness, affective valence, metacognitive awareness, self-narrative, and agency—each linked to distinct moral concerns. It uses a hybrid threshold-plus-gradation approach: binary triggers activate new obligation categories, while continuous scaling adjusts protective weight. Two complementary aggregation methods are offered: one hierarchical, one architecture-agnostic. Worked case studies of Replika and OpenClaw show how systems in different dimensional regions trigger different obligations. The framework applies across neural, symbolic, and neurosymbolic systems, aiming to make consciousness science decision-relevant for organizations.
arXiv (Cornell University)
May 15, 2026
Jonas Mago, Edmundo Lopez-Sola, Jakub Vohryzek et al.
States of consciousness with minimal phenomenal content, such as those induced by certain meditation practices, show increased brain entropy similar to high-content psychedelic states, challenging the Entropic Brain Hypothesis that links entropy to phenomenal richness. The Complex Brain Hypothesis resolves this by proposing that brain complexity, not entropy, better indexes the richness of experience. Complexity is modulated by the grain of inference the brain uses to resolve uncertainty: fine-grained inference loosens constraints and proliferates content, as in psychedelic states; coarse-grained inference simplifies experience into contentless awareness, as in minimal phenomenal experiences. Both regimes can elevate entropy but differ in phenomenology and perturbational signatures, refining the Entropic Brain Hypothesis and highlighting minimal phenomenal experiences as a test case for computational theories of consciousness.
arXiv (Cornell University)
November 23, 2025
Yap, Sin-Yee, Noman, Fuad, Loo, Junn Yong et al.
Psychedelics like psilocybin reorganize large-scale brain connectivity, but how these changes appear across EEG and fMRI networks has been unclear. A new multimodal graph fusion network, Brain-MGF, jointly analyzes EEG-fMRI connectivity by constructing graphs with partial-correlation edges and Pearson-profile node features, then learning subject-level embeddings via graph convolution. An adaptive softmax gate fuses modalities with sample-specific weights. Tested on the world's largest single-site psilocybin dataset, PsiConnect, the model distinguishes psilocybin from no-psilocybin conditions during meditation and rest. Fusion achieves 74.0% accuracy and 76.5% F1 score on meditation, and 76.0% accuracy with 85.8% ROC-AUC on rest, improving over unimodal and non-adaptive variants. UMAP visualizations show clearer class separation for fused embeddings, suggesting adaptive graph fusion effectively integrates complementary EEG-fMRI information for characterizing psilocybin-induced neural reorganization.
arXiv (Cornell University)
September 23, 2025
Victor Oswald, Karim Jerbi, Corine Sombrun et al.
A self-induced, substance-free trance state called Auto-Induced Cognitive Trance (AICT) engages frontal regions, the posterior cingulate cortex, and the left parietal cortex, brain areas linked to rich subjective experiences. Analysis of EEG recordings from 27 trained participants showed that the aperiodic component of the power spectrum, entropy, and complexity measures distinguished AICT from rest, with the aperiodic component being the strongest discriminator. Baseline neural activity in frontal and parietal regions predicted how much brain activity changed when entering the trance. These results suggest that self-induced trance states alter neural functioning in ways that may explain their intense and unique subjective qualities.
arXiv (Cornell University)
September 30, 2024
Jeremy I Skipper, Joanna Kuć, Gregory M. Cooper et al.
Reducing attention to language in multimodal AI models makes their semantic spaces more similar to the phenomenology of altered states such as ego dissolution and unity reported in psychedelic and meditation experiences. When attentional weights were shifted away from language and vision in CLIP and FLAVA models, their embedding spaces aligned more closely with disembodied, unitive, and minimal phenomenal states than with anxiety or random text. This alignment was accompanied by blurred distinctions within and across semantic categories, such as 'giraffes' becoming more like 'bananas'. The findings suggest that language categorization contributes to the structure of ordinary consciousness, and its breakdown may underlie the phenomenology of altered states linked to improved mental health.
arXiv (Cornell University)
August 6, 2020
Rodrigo Cofré, Rubén Herzog, Pedro A. M. Mediano et al.
Altered states of consciousness provide a key opportunity to understand how global brain activity changes relate to different subjective experiences. This paper advocates a research program that bridges bottom-up generative models of whole-brain activity with top-down signatures proposed by theories of consciousness. It defines altered states, discusses relevant brain-activity signatures, and introduces whole-brain models to explore the mechanisms of altered consciousness from the bottom-up. The authors argue that systematic investigation of altered states via bottom-up modeling may help clarify the biophysical, informational, and dynamical foundations of consciousness.