Cutting-edge methods from machine learning could help scientists better understand the visual experiences induced by psychedelic drugs such as dimethyltryptamine (DMT), according to a new article published in the scientific journal Neuroscience of Consciousness.
Researchers have demonstrated that “classic” psychedelic drugs such as DMT, LSD, and psilocybin selectively change the function of serotonin receptors in the nervous system. But there is still much to learn about how those changes generate the altered states of consciousness associated with the psychedelic experience.
Michael Schartner, a member of the International Brain Laboratory at Champalimaud Centre for the Unknown in Lisbon, and his colleague Christopher Timmermann believe that artificial intelligence could provide some clues about that process.
“For me, the most interesting property of brains is that they bring about experiences. Brains contain an internal model of the world which is constantly updated via sensory information, and some parts of this model are consciously perceived, i.e. experienced,” Schartner explained.
“If this process of model-updating is perturbed — e.g. via psychedelics — the internal model can go off the rails and may have very little to do with the actual world. Such a perturbation is thus an important case in the study of how the internal model is updated, as it can be directly experienced by the perturbed brain – and verbally reported.”
“The process of generating natural images with deep neural networks can be perturbed in visually similar ways and may offer mechanistic insights into its biological counterpart — in addition to offering a tool to illustrate verbal reports of psychedelic experiences,” Schartner said.
A deep neural network is what artificial intelligence researchers call an artificial neural network with multiple interconnected layers of computation. Such networks can be used to generate highly realistic images of human faces — including so-called “deep fake” images — and are also being used in facial recognition technology.
In a study published in Nature Communications, researchers found a striking similarity between how the human brain and deep neural networks recognize faces.
“Deep neural networks — the work horse of many impressive engineering feats of machine learning — are the state-of-the-art model for parts of the visual system in humans,” Schartner told PsyPost. “They can help illustrate how psychedelics perturb perception and can be used to guide hypotheses on how sensory information is prevented from updating the brain’s model of the world.”
Schartner was previously involved in research that found psychedelic drugs produced a sustained increase in neural signal diversity. His colleague Timmermann has authored research indicating that LSD decreases the neural response to unexpected stimuli while increasing it for familiar stimuli.
Both findings provided some insights into the brain dynamics that underlie specific aspects of conscious experience.
But the neural correlates of consciousness are still “far from clear,” Schartner said. “The ventral visual stream in human brains seems key for visual experiences but is certainly not sufficient. Also, the exact role of serotonin in the gating of sensory information is still to be explained. Another big open question is how exactly the feedback and feed-forward flows of neural activity need to be arranged to bring about any experience.”
He added: “Psychedelics are not only an important tool for fundamental research about the mind-body problem but they also showed promising results in the treatment of depression and anxiety.”
The study, “Neural network models for DMT-induced visual hallucinations“, was published December 12, 2020.
