DMT simulated with Artificial Neural Networks simulates DMT

Recently more and more researches have been going on in the field of psychiatry and psychopharmacology about various Psychedelic substances. Psychedelics are starting to be decriminalized and used for pharmaceutical purposes. Psychedelics based medicines have received ‘breakthrough status’ by the FDA for alleviating depression. Now there is even an ETF (Exchange-traded fund) on pharmaceutical companies related to Psilocybin and other Psychedelics based medicinal startups.

The Holy Medicine of the Holy and Divine Mother Ayahuasca is already well known for ‘Healing’ and therapeutic roles.

N,N-Dimethyltryptamine or DMT is the active hallucinogenic compound of the Holy Medicine. Together with the therapeutic drives and mental health effects, DMT introduces intensive and highly visual hallucinations. Mostly these hallucinations are geometric patterns and saturation of visual spaces.

How DMT causes 3D patterns

Structurally, as DMT is similar to serotonin, it leads to several relevant therapeutic benefits, especially against depression. Due to its presence in the brain, normal visual processes do not work properly. Other additional inputs from other sensory systems are interpreted as visual, which is more like a form of induced synesthesia. To be specific, those geometric patterns, which are seen or experienced are the results of normal visual processes being disturbed. This includes 3D pattern filler being over-applied, persistence of images overlaying, and difficulty in distinguishing edges, so things melt and blend together. Simultaneous movement of both eyes between two phases of fixation in the same direction creates saccade, leading to synesthesia, which causes other sensory input to be interpreted as visual.

Simulating DMT in computers

Recently Michael Schartner and Christopher Timmermann, who are well known for their researches about DMT, have created a model using two generative deep convolutional networks to reproduce the hallucinations related to DMT. This attempt is not new, and Google DeepDream had already attempted it in 2015. In this study, the authors elaborated a bit more on the effect of conscious experience and serotonin disturbance.

Why trying to recreate those disruptions with artificial intelligence?

Nowadays, artificial intelligence and machine or deep learning is rampant. Generative Adversarial Networks (GANs), a kind of artificial network that produces images, are used commonly not only to classify. Recently, it is observable that one of the growing trends of neurotech companies is the use of Psychedelic for medication use.

Scientific pieces of evidence about the benefits of Psychedelics for major depression are piling up. Figuring out what happens to the brain during those powerful treatments will profoundly assist therapists and physicians, even if the hallucinations are considered as the side effect of the treatment.

Copying ‘perception’ model from reality to regenerate with artificial intelligence

Though Psychedelics and AI both are completely separate from each other, they can be used in symbiosis to design better treatments.

According to Karl Friston, the framework of predictive coding states that the human brain generates a model of the world by constantly combining prior beliefs with sensory information. The resulting model is partially consciously perceived and subject to report. Each experience depends on a balanced weighting of prior and sensory information. Jacob and Trulson, in one of their research, showed that this balance could be disturbed by classical Psychedelics, which act primarily via the serotonergic system.

Inspired by the usage of deep convolutional neural networks to model Psychedelic hallucinations done by Mordvintsev and Suzuki and increasing evidence on the role of the serotonergic system in sensory gating information, the authors of this study suggest two recent generative deep convolutional neural network architectures to illustrate the disturbance of the balanced integration of sensory and prior information associated with visual perception.

According to the researchers, GANs have a discriminator model, which is a neural network firstly trained by exposing it to thousands of images to learn a specific pattern of feature, for example, trained to find cats in a picture. Another model has a generator that continuously produces fake images from random values until the discriminator cannot distinguish the true from the produced fake images. This second model has become very popular recently, allowing deepfake to create pictures of animals and humans that do not exist.

Specific use of GANs is known as the “style transfer”, where the content of an image is conserved while the style coming from another image is integrated. For example, Mona Lisa in the style of Van Gogh’s Starry Night. In this case, the researchers perturb an image until the discriminator is not convinced that the perturbed image belongs to a specific style. This is the essential part of this study. By doing this, the researchers are aiming to answer the question of “What happens if given a traditional picture seen by the brain perturbed it, transferring the DMT style?” In this way, the GANs models are practically generating the produced synesthesia given by the DMT disrupting the brain.

According to the researcher, the human visual stream is generally considered the key for visual experiences, and this study is promising as a computational approach to reproduce disruption in the brain with DMT.

Conclusion

The authors of this study mentioned that this study is certainly not enough for a whole DMT and other Psychedelics demonstration, as it does not fully show all effects of DMT on the serotonin receptors need to be understood. They also emphasize that science is still far from a complete understanding of consciousness or perception expansion and perceiving, but according to them, it is a step towards it.

They also expressed hope that now the times are different; with current open-mindedness, this can be the start of a more serious computational representation of DMT.

References

  1. Alan K. Davis, Sara So, Rafael Lancelotta, Joseph P. Barsuglia & Roland R. Griffiths. (2019). 5-methoxy-N,N-dimethyltryptamine (5-MeO-DMT) used in a naturalistic group setting is associated with unintended improvements in depression and anxiety. The American Journal of Drug and Alcohol Abuse, [online] Volume, 45(2), p. 161-169. Available at: https://doi.org/10.1080/00952990.2018.1545024 [Accessed 11th July 2021].
  2. Friston K. (2018). Does predictive coding have a future? National Neuroscience, [online] Volume, 21(8), p. 1019-1021. Available at: https://doi.org/10.1038/s41593-018-0200-7 [Accessed 11th July 2021].
  3. Jacobs BL, Trulson ME. (1979). Mechanisms of action of LSD. American Scientists, [online] Volume, 67(4), p. 396-404. Available at: https://pubmed.ncbi.nlm.nih.gov/533027/ [Accessed 11th July 2021].
  4. Schartner, M. M., Timmermann, C. (2020). Neural network models for DMT-induced visual hallucinations. Neuroscience of Consciousness, [online] Volume 2020(1), Available at: https://doi.org/10.1093/nc/niaa024 [Accessed 11th July 2021].
  5. Tero Karras, Samuli Laine, Timo Aila. (2019). Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Computer Vision Foundation, [online] Volume, 2019, p. 4401-4410. Available at: https://openaccess.thecvf.com/content_CVPR_2019/html/Karras_A_Style-Based_Generator_Architecture_for_Generative_Adversarial_Networks_CVPR_2019_paper.html [Accessed 11th July 2021].