Manifesto
Reality in the 21st Century
Sunwoo Choi
Introduction
There is a rising tide of uncertainty that has come with the movement of Large Language Models into the mainstream. Sometime around the turn of 2022, the name “ChatGPT” echoed through my 8th-grade halls for the first time, and the concept that was “AI” became something much more tangible within the layperson’s life. Now, in 2026, the imminence of AI has led to an elusive sense of doom. People like to catch glimpses of this “sense of doom” by associating specific concrete phenomena, such as the failing job market or the emergence of skynet-esque companies like Palantir, with this general fear of AI. Yet the fear of AI is not completely nor concretely defined by these sub-fears. The fear of AI is something much more general, much more abstract, much more human. The fear of AI is only the exacerbation of a fear that has been with us since the beginning of time: Human Error
But what is Human Error? Human Error is a euphemism for forgetting. A euphemism because forgetting sounds too pathetic to attribute to the incredible amount of suffering and liability brought on by the simple act of forgetting. All human errors result from forgetting. Confusion is forgetting the identity of a schema within our mind, lashing out is forgetting the long-term consequences of an action, and overconfidence is forgetting the reality of your physical being (I don’t believe in overconfidence in regards to our mental or spiritual being, because I am optimist). Forgetting is the root and the only real factor behind human error. So, in explaining the problem of Human Error, I would actually like to explain “forgetting”, and I would first like to frame “forgetting” in 2 ways- forgetting within ourselves and forgetting between individuals. These definitions of forgetting are not new. They have been true since the beginning of humanity. However, they are important to establish before I talk about how they have started to manifest in the face of recent technology.
Introspective Forgetting — Reconstruction Distortion
Research by Dr. Qi Wang of Cornell University has shown that memory is reconstructed upon recall, and this reconstruction is heavily influenced by an individual’s state at the moment of recall. Their mood, their agenda, and their culture all influence how the memory is reconstructed. For example, individuals may fail to recall certain positive aspects of a memory when they are in a negative state, or may unconsciously augment a positive aspect of a memory when recounting their memory in a positive state (augmentation itself is a “forgetting” of the extent to which a phenomenon occurred). Furthermore, individuals may forget specific aspects of a memory given their agenda. In the famous car crash example, individuals were shown a video of a car collision with no broken glass. Yet when asked the question “how much broken glass resulted from the crash”, the individuals recounted moderate to extreme amounts. Their agenda at the moment of recall was to remember glass, and they “forgot” the reality of there being no glass (again, I’m counting augmentation of memory as forgetting as well). Finally, the culture and society that surround an individual can also lead to significant faults in memory reconstruction. Thus, through many different internal processes, memory distortion occurs. The rather haphazard reconstruction of memory influences our self-identity and what Dr. Wang calls our “self-narrative”. This is not a finding specific to this paper. It’s a known phenomenon that extends to all humans, more so in some than others. In individuals with Major Depressive Disorder, rumination leads to continuously depreciating self-identities, as memories are reconstructed with increasingly negative lenses. In narcissism, the opposite can occur. In all cases, with all humans, forgetting leads to distortion which causes inaccurate views of oneself. We do not know ourselves.
External Loss — The “Bucket” Problem
Now, if memory can never be reconstructed perfectly within one’s own mind, then the threshold of the amount of truth conserved in a memory is even lower when transferring a memory to another individual. In other words, when we communicate with others, even more human error occurs. And all of this error is driven primarily by the fact that others are not ourselves.
When communicating with others, information is transferred in chunks. In emails, in meetings, in briefs. It’s all bucketfuls of information. It may contain a lot, but it cannot contain everything. Information transfer is always limited. It’s not a continuous stream of thought as it is within ourselves. I will refer to communication and information transfer using this bucket analogy from now on. When an employee is onboarded onto a team within a company, they are given a run down of what the team does. They get to see what the team has accomplished. They are given a glimpse- a bucketful- of who the team is. However, that bucket does not contain everything. It doesn’t contain all the nuances of a team, whether that be a subtle jargon that has developed amongst them or specific power dynamics within the team. This leads to onboarding members having trouble adjusting, and even becoming a liability to the team.
The “bucket” problem exists because there are physical constraints in the world. When communicating with others, information has to pass through the physical world, not just maintained as it is within our minds. This can be the aforementioned emails, meetings, briefs, conversations, even visual art and music. If individual A had the complete and full context of Individual B, then they would cease to be separate beings. Therefore, the very idea of individuals, distinct and separate, requires loss of information during communication.
The Outsourcing of Thought
So far, forgetting has been internal- faults in the preservation of reality when we reconstruct memory- and external- faults in fully communicating reality to another individual. The advent of Large Language Models has formed a 3rd state in the modern age. It is not quite an “other”, but it is also not quite “ourself”. It serves to help us think, reflect, process, but it is not internal to ourselves. Here, we see a combination of the two previous problems we have discussed.
When we communicate with Large Language Models, we are using that “bucket”, just like any other type of communication. In this case, that bucket is the prompting window. There are things that are included in the bucket, but also things that are left out of the bucket. Developers like to use the word “context”, but I’ll prefer to use the term I’ve been using so far- memory. Large Language Models are external to us, and therefore undergo the “bucket” effect. However, the problem is humans are starting to treat it like it is internal to ourselves. And with that, our identity automatically becomes entangled in the mix. Large Language Models over-validate our ideas. Large Language Models hallucinate. Large Language Models returned biased information based on the objective of the prompt. And all of this is internalized in our brains just as distorted memories are internalized in our brains during introspection. We start to develop inaccurate views of our own selves. It is the distortion that I talked about in point I, but aggravated to a great degree.
In simple words, the emergence of Large Language Models has brought about a quicker, faster distortion of reality within our minds because it combines the lack of context of external communication with the thought distortion of introspection. Our thoughts themselves are being outsourced, and with that more opportunities to lose memory and internalize distorted memory.
The Solution
I have already managed to accurately define the problem of our exponentially growing integration with technology. It is twofold: It involves the inaccuracy of memory reconstruction and the limited amount of information that can be communicated within the real world. Then, the solution must also be twofold:
In Relation to Reconstruction Distortion
To combat the inaccuracy of memory, the easiest, non-neurological way of doing this is having thought and experience being recorded in the moment. Memory reconstruction is where the fault of human error takes place. The closer the moment of recall is to the moment of event (the T0), the less the distortion. Everything that happens externally, in reality, can be recorded through video and audio using modern technology. This has already been an idea that has been implemented through governmental surveillance technology and recently within the consumption world. As the 21st century has progressed, consumers have also begun to record more and more of their lives, from digital cameras, to iPods, to iPhones, to now Meta Glasses. Using such technology, a repository of every external event that happens to an individual is possible.
However, what about things that happen internally? This is where technology lacks the ability to completely cut out distortion. Every thought we have is our own interpretation of reality. Without getting into the metaphysical implications of our perception, what I’m saying is this: our thoughts are impossible to un-bias, because when we verbalize or transcribe our thoughts, that itself is reconstruction. It’s something internal that is becoming external. However, we can mitigate the extent of human error to the highest degree by removing the time between thought and recall. Essentially, crystallizing one’s thoughts at the moment of occurrence can not eliminate, but minimize the “forgetting” distortion of a thought. Traditional journaling attempts to do this, however there is a threshold of inconvenience that prevents a lot of individuals from doing it moment-by-moment. Opening a journal, taking out a pen, and writing one’s thoughts down is highly cumbersome to do multiple times a day. Even with note-taking apps on devices such as a phone, there is still an element of inconvenience that arises out of opening up the device, clicking on the application, and writing down the notes. This slight barrier of inconvenience prevents any note-taking or thought transcription from being fully integrated in moment-by-moment life. However, it is true that certain devices such as iPhones and smartwatches have become omnipresent in our lives. Utilizing voice-transcription technology on these integrated devices can minimize the friction of thought recording, which can make near-T0-recording on a moment-by-moment basis possible. With such technology, an internal thought repository of an individual becomes feasible as well.
In Relation to the Bucket Problem
Now addressing the problem of the bucket, the solution is to eliminate the limit of information that can be transferred. This is philosophically impossible unless we fully integrate ourselves with a Large Language Model and become one being. However, we are able to set that limit as high as possible, especially using modern day cloud technology. A moment-by-moment transcription of our thoughts not only solves the reconstruction problem by minimizing the time between occurrence and recall, but it also solves the bucket problem by maximizing the amount of information (context) shared.
If we are able to incorporate hardware such as the aforementioned iPhones, Meta Glasses, etc… to record as much reality as possible, and voice-transcription technology on integrated devices such as smartwatches to record as many thoughts as possible, then we are essentially able to form a comprehensive repository for an individual. The internal thought repository and the external experience repository can go hand in hand to form an accurate portrayal of an individual’s life. This allows for improvements in all three entities I have previously mentioned: it improves an individual’s sense of self, it improves others’ sense of teammates and acquaintances, and it improves the Large Language Model’s understanding of the users. In terms of the last point, a fully comprehensive database of an individual can allow for the Large Language Model to become the singular most powerful tool in the history of humanity.
Future Framing
Technological advancements in neuroscience can allow for the thought repository aspect of my proposed solution to take a step forward. Neuralink has already opened doors for immediate brain-to-database communication. At that point, the bucket becomes an individual’s entire brain. Although the technology is not sufficient enough to fully transfer the entire mental processes of an individual onto a “thought repository”, and although there are significant hurdles regarding what the nature and form of the content transferred would be, this framing proposes a potential future jump that could integrate individuals with technology to perhaps a final form.
