#320 — Constructing Self and World

Episode Summary

Executive Summary: Sam Harris and Shamil Chandaria discuss Chandaria’s path from mathematical economics and AI to meditation, psychedelics, and philanthropy, centering on predictive processing and the free energy principle as a framework for understanding perception, selfhood, and consciousness. The conversation connects neuroscience, Bayesian inference, Buddhism, and AI to explain why experience is a constructed model rather than a direct readout of reality.

Main Topics: Chandaria’s multidisciplinary path (Priority: 4/5): Chandaria traces his route from mathematical economics and finance to DeepMind, philosophy, neuroscience, and philanthropy, explaining how these domains converged around AI, brain theory, and human flourishing. The free energy principle and Bayesian brain (Priority: 5/5): He presents the brain as an inference machine that predicts causes of sensory data, using approximate Bayesian inference to minimize prediction error and maintain a workable model of the world. Predictive processing and hierarchical perception (Priority: 5/5): Harris and Chandaria explore how top-down generative models and bottom-up prediction errors operate across layers of cortical hierarchy, reversing the common-sense feed-forward view of perception. Meditation, non-duality, and the self-model (Priority: 5/5): Chandaria links meditation traditions, emptiness, and non-dual practice to the idea that the brain constructs a phenomenal self-model and a phenomenal world. Psychedelics and neuroplasticity (Priority: 4/5): They discuss psychedelics as a tool for rapidly perturbing predictive priors, potentially opening windows into neuroplasticity and changing self-experience in durable ways. AI, DeepMind, and human-level inference (Priority: 3/5): Chandaria notes that machine learning architectures and large neural networks provide practical analogies for understanding brain computation and hierarchical feature extraction. Philanthropy and effective altruism (Priority: 3/5): He explains how effective altruism shaped his funding of neglected research areas such as psychedelic science and meditation neuroscience, including founding key research centers.

Key Arguments: Perception is not a direct window onto reality; it is an internally generated simulation constrained by sensory input. Bayesian inference offers the best current high-level account of how the brain infers the causes of its sensory data. The free energy principle formalizes how the brain balances prediction error against prior expectations to maintain a coherent model of the world. Hierarchical predictive processing explains why higher cortical areas send more feedback than feed-forward information. Attention can be understood as precision weighting: increasing the influence of sensory evidence relative to prior beliefs. Meditation and psychedelics can reveal the constructed nature of perception and selfhood by weakening ordinary predictive constraints. The phenomenal self is likely a model generated by the brain at higher levels of the inference hierarchy. Deep learning systems are useful analogies for brain function because they also build hierarchical representations from low-level features to abstract concepts. Psychedelic research and meditation neuroscience are important, relatively neglected areas that can benefit from targeted philanthropy. Effective altruism motivates funding where marginal impact is likely highest, especially in underfunded foundational research.

Data Points: Years meditating: 35 years - Chandaria says he began meditating 35 years ago, starting with Transcendental Meditation. Years since first Buddhist retreats: ~20 years ago - He notes his first Buddhist retreats began around two decades prior to the conversation. Years at Gaia House focus: 7–8 years ago - He describes spending substantial time at Gaia House in the UK in the recent past. DeepMind tenure: 2015–2021 - Chandaria served as a strategic advisor at DeepMind during this period. PhD institution: London School of Economics - He earned a PhD in mathematical modeling of economic systems. Master's institution: University College London - He completed a master's in philosophy there. Oxford funding year: 2018 - He helped endow the Global Priorities Institute at Oxford in 2018. Psychedelic research center founding year: 2019 - He was a founder of the Center for Psychedelic Research at Imperial College London in 2019. Conference size: ~70 people - Harris recalls the 2015 Puerto Rico conference where they likely met. Visual field detail: Tiny macula/foveal region vs. broad peripheral field - Used to illustrate that conscious vision is constructed rather than directly read out from retinal input. Deep learning depth: Close to 100 layers - Chandaria references transformer architectures underlying models like ChatGPT and Google’s Bard. Top-down connections: ~10 times more feedback than feed-forward - Harris cites this as a notable neuroanatomical fact supporting predictive processing.

Pivotal Quotes: "we are already brains in vats" — Sam Harris: Harris emphasizes the philosophical point that our access to reality is mediated by sensory signals, not direct contact with the world. "the brain is having to construct or fabricate or simulate a world, a phenomenal world, and a phenomenal self" — Shamil Chandaria: Chandaria summarizes the core predictive-processing view of perception and selfhood. "what we actually do with the sensory data is use it to calibrate our simulation model" — Shamil Chandaria: He explains the relationship between sensory input, priors, and the brain’s generative model.

Implications: The discussion suggests that perception, selfhood, and even spiritual insight can be studied as inference processes. For AI, neuroscience, and psychedelics research, predictive processing offers a unifying framework with major implications for treatment, cognition, and understanding consciousness.