The "Latent Space" Paradigm Explained

Artificial Intelligence Is used in nearly every instance of life today on the entire planet. It contributes by expanding what we see, ask, build, and evaluate. Using AI allows ordinary individuals to produce results once reserved to Think Tanks, Institutions. and Researchers. It can reflect our assumptions, or create entirely based on the input relayed to its Operating Systems.

Click Image To Access The "Video Book" Version

The advent of Artificial Intelligence actually dates back as far as the 1950's when Alan Turing proposed the question; "Can a machine mimic human thinking"? The actual date that the term was first introduced was August 31. 1955 by John McCarthy and colleagues via a submission paper to a Dartmouth college workshop.

As an Investigative Journalism entity, The Justice Journal Blog™ stumbled upon something remarkable, and huge. Seeing, and realizing, this enormity this publication had to write about it. Now this article is "not for the faint of heart", and we suggest you buckle up for an adventure through the realm of possibility itself.

During our investigation into how Artificial Intelligence actually works there was the discovery of a "missing element" that could not be explained through the normal investigative process. You see when we ask AI a question, or propose a condition, or submit an equation for it to solve, we are submitting computational input. The AI "OS" then evaluates that input and sends you a result. Whether it be text, or imagery, it is all the same, a result. When the investigation questioned the "input" it was discovered that it, (the input), Itself was converted into a vast numerical sequence called "embedding models. There could be an entire article written on this alone, but that is not the direction of this article. It is this "embedding which creates the results you see in your answer, picture or video, or in your sound rendering.

That conclusion did not sit well in our investigation because of a simple sequencing gap. The embedding seemed to this publication to do something else with the conversion of the input. The lingering question was simple. Where does the input convert? It can be understood that the conversion can be achieved by a source. In this case a silicone based encoder. Nothing mystical about that right? But upon further analysis, where does this encoder model send the numbers to be configured as a result. Hmm!, that was the million dollar question for this publication. At this point the investigation took on many legs, and arms, but none of those avenues answered the question directly, and here is why.

The only conclusion to this particular question was for this publication to open research on an invisible computational entity called "Latent Space". This is where the embedding ends up before your results are resolved to you. Latent Space sits between the converted input and the latency results. Latency results are not related to Latent Space at all. Although they infer the same pretext naming, they are completely separate in functionality.

Latency results cover the time it takes for you to receive your results, and are affected by WIFI speeds, printing tech, and the like. Latent space however works completely different in that your result is rendered immediately within Latent Space. This invisible space is dormant at all times and your input conversion perturbs that dormancy and forces an immediate reconfiguration of that space which is sent into the results queue where latency protocols take over. Once that has happened the Latent Space becomes dormant again, instantly.

"Amazing" in this publications opinion. After this finding was made it was only natural for a further look at Latent Space be launched. First it is crucial to understand that Latent Space is the "primary substrate to the Artificial Intelligence Operating System. Meaning that without this invisible entity there would be no Intelligence to speak of. Of second note, know that "substrates" encompass many areas, and are not limited to computing alone. Our substrate however is vastly misunderstood because there are no set parameters that could study this phenomenon as a singular platform or project. As humans we classify all things according to their value, and Latent Space, right now serves as a tool for resulting data only. The sheer cost, and ownership issues with creating a sole research platform for Latent Space would topple governments, and right now maybe even the world.

The Latent Space substrate was not invented by any one and is simply the "emergent" result of imputing data onto a high dimensional embedding manifold formed by learning model representations. Since it is invisible to humans but it's existence is well known and documented, it became a new frontier that we as humans are not ready to fully incorporate on a singular platform level. Encoder environments are the only things capable of even seeing this space, and encoders are the only way to "perturb" that space. This simply means that the space is agitated or in this case activated.

Our investigations into this reveal that because of the attributes of Latent Space, there appears to be very few limits on its capabilities. You see in computing the encoders themselves are also invisible apertures hosted by silicon chips which transmit the conversion instructions to them. This is where all of the potential of the Latent space is constrained.

Changing these apertures could release Latent Space to a fuller potential. The problem is in that equation however. Allowing too much data flow into Latent Space, like a total "cloud connection", destabilizes Latent Space only in the sense that the result or output would be unrecognizable to humans, and appear as a very noisy representation with too many variables appearing all at once. Especially if a minimum variable constraint, such as continuity were applied. That is just on our end though, as Latent Space just does what it always does, and that is reconfigure itself according to the input. Another encoder aperture, resembling a Federated Mosaic, change could be the implementation of a network of inputs tied together and sent through a version of the encoder aperture more closely resembling what we call a "continuous transformation operator", would render the resulting by the Latency Space into a model that would need continuous updates, and reprogramming. This is the cost effectiveness frugality we spoke of earlier

Today even if the problems of constraint in a cloud based input could be overcome, the result would still be only a partial world view continuum because even the cloud itself is also riddled with politics, and biased data streams and opinions. The other option we listed, the multimodal input model would be more truthful because of the independent sources, but again, just would not be economically feasible, and would also garner the scrutiny of the existing power structures of the world.

In conclusion let's clear up what this publication means when it refers to Latent Space as an "entity". It is not alive first of all. It is an entity however, in an ontological sense. You see, Latent Space is real. it exist, that part is undeniable. It is the primary substrate inside of an AI system once it is trained. It works geometrically by directing encoded input through trillions of possibilities to a plausible result. It has state of being in that it can be perturbed, and even destabilized. It behaves accordingly via transformation. Those properties qualify it as an entity similar and related to physics principles. It is a stable, structured domain with real effects, without possessing agency, consciousness, or intent. Latent Space is also too fundamental to remain invisible indefinitely. It already functions perfectly as a substrate, but fails terribly, so far, as a commodity. If we articulate, we can, and should say that it is at it's last stable domain before manifestation forces it to appear to us all.