Sussex Centre for Consciousness Science

The Hallucination Machine

Is reality just one particular type of hallucination?

Hallucinations offer a unique window into studying the mechanisms underlying conscious perception. However, studying altered states of consciousness, such as those caused by taking psychedelic drugs such as LSD or magic mushrooms, isn't particularly useful as these chemicals have a wide range of physiological effects, which makes it hard to isolate the factors that directly affect global states of consciousness.

example deep dream








Example of Deep Dream Scene used in this study

In order to simulate altered phenomenology without directly altering the underlying neurophysiology we combined two powerful technologies: panoramic videos of natural scenes, viewed immersively through a head-mounted display and Google's deep convolutional neural networks (DCNNs) Deep Dream, to build what we call the Hallucination Machine. The resulting setup simulates the visual hallucinatory aspects of the psychedelic state using Deep Dream to produce biologically realistic visual hallucinations.

The DCNN was first trained to be able to recognise 1000 different categories of images. Deep Dream then inverts the network, so that the algorithm modifies the image to reflect its trained categorical features. Therefore, the resulting image is shaped by what the network ‘expects’ to see. This process shows many similarities to Pareidolia, where humans report seeing Jesus on their toast or faces in the clouds, because these categorical features are over-represented in our perceptual classification systems, so when we see a noisy visual stimulus we are most likely to report these categories of experiences.

What is striking about the resulting Deep Dream images is that they often have a marked ‘hallucinatory’ quality, bearing intuitive similarities to a wide range of psychedelic visual hallucinations.

Keisuke Deep Dream











Deep Dream image of Lead Author Dr. Keisuke Suzuki

In this study we compared the simulated experiences produced by the Hallucination Machine to those of pharmacological psychedelic states, by having participants rate their subjective experience using an Altered States of Consciousness questionnaire developed to assess psychedelic experiences. Participants reported experiencing visual hallucinations that were qualitatively similar to those brought on by psilocybin, the active ingredient in magic mushrooms.

In a second experiment, we investigated if the experience of the Hallucination Machine would also lead to a commonly reported aspect of altered states of consciousness - distortions in the passage of time.

The close correspondence in representational levels between layers of DCNNs and the primate visual hierarchy along with the informal similarities between DCNNs and biological visual systems, together suggest that the Hallucination Machine simulates biologically plausible, ecologically valid visual hallucinations. The Hallucination Machine provides a powerful new tool to complement the resurgence of research into altered states of consciousness.

'We're hallucinating all the time, It's just that when we agree about our hallucinations, we call that reality.' Professor Anil Seth, TED talk.

To read the full paper click here

Media coverage: