top of page
Search
  • johnmcgaughey255

Neural interpretation and psychedelics

From the Amanita Muscaria found in damp forests in the northern hemisphere to the Ayahuasca drink, rich with dimethyltryptamine (DMT), psychedelics and psychoactives have played a critical role in our history and arguably our evolutionary psychology. Autonomously, amazingly, almost every tribal group has participated in the ingestion of psychoactive or psychedelic substances. Native Americans with the ingestion of peyote, a desert cactus which contains high levels of the psychedelic compound mescaline. Evidence dated 9,000 years ago strongly suggests psychedelic use in the Sahara desert, inferred by the murals in the area, this due to the psilocybe mairei fungus. What do we make out of this? Is the question of importance really important to ask? What do communities and tribal entities have to gain from this practice? Is it somehow mind expanding, lead us to a deeper understanding? Whatever it is, there is convincing evidence that psychedelic's role in our evolution plays a fundamental role in how we perceive this world. Psychedelic consumption was, by my hypothesis, the change in human behavior that allowed us to surpass the other primates, and step onto the path of exponential technological growth. The subconscious conceptualization of generalization, on multiple levels fluidly from the physical to the abstract, is in my opinion due to psychedelic's effect on our brain.

The article I was reading introduced three speculative claims of neural networks, some really interesting concepts. The first claim: features, which can be understood and studied are the fundamental unit of neural networks. The second claim: Circuits are the features connected by weights, and they can be understood and studies. The third claim: Universality: analogous structures of features and circuits across a wide domain of networks. The claim of feature recognition is not widely accepted by research scientists in the domain of deep learning. It is pretty commonly accepted that specific digital neurons react to edges and curves, and put together, these individual neurons are part of a larger circuit in feature detection. The controversy of these curve detectors straddles between the accepted idea of edge and boundary detection and more trivial ideas such as ear or face detection as a feature. Understanding what happens inside the black box of neural networks is important not only to understanding how the model functions and generalizes, but it is important to understand the essence of the net to improve upon it. The usefulness of viewing a generalization from the perspective of a specific neuron is to understand to what extent the neuron captures the essence of the training data. Perhaps we want to train a network to understand and generalize across Christmas trees, looking at the generalization from the perspective of a neuron can give us incredible insight to how to machine understands the essence of what a Christmas tree is. We can then, understand how we can change our training data to influence that neurons generalization and improve our model that way.

The psychedelic idea from the first paragraph may seem random and weird, but I assure you there is an interesting though process behind both of these ideas. It is almost evident to me that psychedelic let us see a more generalized version of what it is we are looking at or thinking of. Just as we can look inside of a neural network to understand the models understanding of the essence of a particular thing, I believe that taking psychedelic allows us to see our individual generalization of a phenomenon, a visual representation of the essence of an idea. In this picture below, a model was trained to recognize dogs, and in order to do that it must form an appropriate understanding of the essence of what a dog is, a working generalization. More approximately, the model wants to generalize across dogs in different poses and orientations. If we look at the feature visualization for neuron 4b:409, it is very informative about what the neuron is looking for. Perhaps now you can understand the connection between psychedelic visualizations and feature visualizations. It is something about ingesting psychoactive substances that allows us to visualize a phenomenon in a more generalized and abstract form. Perhaps this is not simply with objects, like Christmas trees, or boots, or dogs, but with our ideas. Maybe it

is with ideas, breaking out of our normal modes of thinking and introducing to us a new picture of how we should live our lives given the abstract generalization we have been shown about our lives.


3 views0 comments

Recent Posts

See All

Meta-heuristic optimization

These are all drafts, by the way. They are not meant to be perfect or to convey all the information I wish to convey flawlessly. My blogs are just a way for me to get ideas and my thoughts realized as

Dimensionality reduction and model interpretability

I would say that the main purpose of communication is to give a universal understanding of abstract ideas. An abstraction is, for my intents and purposes, a lower dimensional encoding of a higher dime

bottom of page