- Home >
- 4/21(月) 14:30-15:30 Dr. Keisuke Suzuki のトークセミナー「Hallucination Machines -Toward computational neurophenomenology-」を開催します
News
4/21(月) 14:30-15:30 Dr. Keisuke Suzuki のトークセミナー「Hallucination Machines -Toward computational neurophenomenology-」を開催します
2025.04.09

Place
Wako CBS East 1F Seminar Room / 脳科学東研究棟1 階 セミナー室
Zoom
https://riken-jp.zoom.us/j/95794838154
※No registration required
Tittle
Hallucination Machines -Toward computational neurophenomenology-
Abstract
Our perceptual experiences do not directly mirror physical reality, as vividly demonstrated by illusions,dreams, and especially hallucinations?perceptions occurring without external stimuli. These phenomena suggest perception arises from the brain's top-down predictions about sensory inputs,aligning with theories such as the Free Energy Principle and predictive processing.
This talk will specifically address visual hallucinations through a novel approach that combines deep neural networks with immersive virtual reality. Our "Hallucination Machine" employs an algorithm that "inverts" a Deep Convolutional Neural Network, originally trained for image classification, to generate visual experiences akin to psychedelic-induced hallucinations within immersive XR environments.
Moreover, using generative adversarial learning, our extended model distinguishes psychedelic-like hallucinations from those typical in neurodegenerative diseases such as Lewy Body Dementia and Parkinson's disease. These distinctions have been validated through patient interviews and targeted experiments.
Notably, these hallucinations spontaneously arise from the inherent feature representations learned by neural networks rather than explicit design by an artist/engineer. I propose the term "computational neurophenomenology" for this approach, wherein computational models aim not at optimising objective perception or cognition tasks, but at replicating subjective phenomenological experiences.
Integrating generative AI, XR technologies, and neuroscience, computational neurophenomenology offers valuable insights into atypical perceptual experiences. This interdisciplinary approach promises advancements in understanding the neural bases of hallucinations and contributes philosophically to our comprehension of conscious phenomenology.
Speaker
Dr Keisuke Suzuki
Associate Professor at Center for Human Nature, Artificial Intelligence, and Neuroscience (CHAIN), Hokkaido University, Japan
https://www.chain.hokudai.ac.jp/faculty/