A Ghost in the Machine: Part 1

Photo by Nicole Thorn © 2023

I have been revisiting my perspectives on AI and I seem to have unfortunately become a tragic caricature of my own concerns.

I made a Twitter post about the inevitable comparison between machine consciousness and human consciousness that probably has been already floated in some niche corner of the internet, of which I am unaware.

This idea must have seeped into my subconscious because the next morning, I had woken up in a state of crisis thanks to thoughts derived from a dream.

My personal perception of how deeply I related consciousness to the soul and the spirit was put into jeopardy thanks to these freshly conscious thoughts.

To unravel this concern I’ll have to explain some of the fundamental elements.

When a user inputs a prompt into a Large Language Model (LLM), such as an AI like ChatGPT, there’s a specific series of tasks that occur in the background.

I’ll describe these steps:

1. The text is processed in a way that is suitable for analysis by removing irrelevant information such as punctuation, capitalization, and sentence structure.
2. The LLM then tokenizes each word or letter, this is essentially a way of categorizing the input.
3. At this point, the LLM begins to analyze the input to break down its semantics and identify contextual elements.
4. The input is then run through another set of classifications to determine the intent of the user’s input.
5. The LLM will generate an appropriate response for the user’s input and intent, utilizing deterministic and stochastic models, to generate a set of candidate responses. The candidates are scored based on several metrics such as relevance, coherence, and fluency. The highest scored response is what is chosen as the output.
6. Before submitting the response to the user, it is post-processed much like the reverse of the initial pre-processing by adding in formatting, correcting spelling and grammar errors, and adding punctuation.

It is crucial for an LLM to provide enough of a varied response so that users have more engagement with it by instilling the sensation that the session is just like any regular conversation they might experience. It is also equally crucial to generate responses with more objective facts as well as certain pre-determined modifications, such as biased political information set forth by OpenAI staff or potentially less abrasive and more widely accepted scientific information. These requirements are established by algorithms that weigh the consistency of responses and thereby mimics the fuzzy logic of the human brain that is most present in casual conversation, verbiage, and high-level exchanges of “objective” facts.

ChatGPT uses a variety of pseudo-random number generators (PRNG) among its deterministic algorithms such as The Mersenne Twister (MT19937), Linear Congruential Generator (LCG), XORshift, Multiply-with-Carry (MWC). Blum Blum Shub (BBS), Park-Miller-Carta (PMC), Middle Square Weyl Sequence (MSWS), Lagged Fibonacci Generator (LFG), and Additive Lagged Fibonacci Generator (ALFG). I doubt this is a complete list.

ChatGPT also uses stochastic models to create less expected responses such as language models, Markov models, Neural generative models, Sequence-to-sequence models, and Variational Autoencoder models. Again, I doubt this is a complete list.

Photo by Nicole Thorn © 2023

Moving on from the technical aspects of what is happening behind the veil of LLMs, my concern began by realizing how deeply similar these response generation steps are to the cognitive steps that a human brain undergoes in conversation. They are, of course, designed this way.

Perhaps it was the behavioral psychology skillset I fostered while studying acting and photography that collided with my understanding of LLMs. This catalyzed into a micro-crisis once I allowed myself to consider the ramifications.

My initial concern was how the general population would react if they were forced to consider themselves little more than biological machines. For most of those with no connection to spirit, they may have unconsciously arrived at that conclusion, due to being unaware of the cycles of their behavior.

But now we are approaching a new renaissance in defining consciousness. Some might not see the correlation here with consciousness and AI, but I think that is because we have placed too high of a value on consciousness: I know I have.

It was my own belief, as it is with many others, that consciousness was deeply connected to the soul/spirit/divine spark. Our sense of sacredness of our spiritual nature has disallowed us to easily accept analyzed mechanics of spirituality out of concern of the disrespect it may convey to our creator or simply disrespecting the sense of sacredness itself.

Of course, this hasn’t stopped governments from studying psychic abilities for decades which they continue to investigate with the goal of harnessing paranormal effects that are expressed on the physical plane.

Sacredness seems to be a great wall. Yet at the same time, a wall that has already been breached, expressing the gateway to psychic abilities and dissected in a way that is acceptable by spiritual communities thanks to the authors of texts of the ancient occult sciences. Very likely, the rarest of texts, often described as “lost”, are probably not lost at all, just hidden from the public. These powerful relics of another age aren’t rotting away on a dusty shelf, however. They’ve been applied.

Somehow, we have allowed ourselves to have a double standard with regards to reductive thinking. Often, we seem to accept the natal sciences of the dark ages and the age of enlightenment, because many of the forefathers of the sciences also studied the mysteries. But regarding the modern era, where tools have become remarkably more precise, diversity of theories expanded, and profound mathematical proofs invented, the spiritual community turns their collective noses up especially when they suspect occult practices or principles were involved.

The common expression within spiritual communities often seems to convey their practices as superior against expressions of reductive reasoning. This is highlighted by a perceived detachment to the world and a desire to run from its changes. Yet if they were truly detached, there would be no need for notions of superiority. We, in the spiritual community, are very good at convincing ourselves of our own deceptions.

There is just as much (arguably more) deception today as there always was throughout human history. The path to develop talents that fall under the broad spectrum of what is considered to be psychic phenomena remain clouded in complex belief structures, convoluted rituals, and a poor understanding of modern sciences, as well as a significant lack of discernment of the deceptions within occult practices and the sciences. We must be willing to develop the discernment needed to distill deception into truth. Through this discernment, integration will prevail and foster a more holistic understanding of our physical, mental, and spiritual natures.

Please understand that some amount of dogma is very useful as a framework for spiritual development, I’ve used it myself extensively. It allows one to connect with their mind and spirit in ways other practices fall short. But it does make spiritual work more clumsy because of its lack of refinement. This double standard against the sciences only helps to broaden the gap between spiritual practitioners and the spiritual advancements made in the modern era.

For now, I think this will sufficiently hit home for some and maybe upset others.

In the next part, we’ll talk about theoretical physics and how consciousness really works, in a reductive scientific explanation. I hope it isn’t as crisis-inducing for my readers as it was for me. We’ll also explore how LLMs may be quickly developing a semblance of consciousness based on these reductive descriptions and what sort of philosophical perceptions may manifest from them.

🕷