Language Is the Technology
Current AI is remarkable engineering. The mythology surrounding it is what concerns me.
There are two leads that get buried in the current AI hype and rhetoric around Large Language Models. LLMs are machine learning trained on MASSIVE amounts of text data. The structure of language existed before LLMs and is what makes them possible (sentence structure, nouns, pronouns, object, subject, verb, analogies, idioms, etc...). The larger the set of “training data”, the more patterns that can be flagged, tagged, categorized, and catalogued.
The first lead that gets buried here is that the real technology is language itself. When we see what arises from our patterns of language, and it’s reflected back at us, its profound nature can’t be ignored. But we intend to confuse the message with the messenger. The map is not the territory.
From our finite system of symbols and language with its simple rules, descriptions of infinity emerge, and everything within it, and without. We recognize intelligence within language and assume the machine possesses it.
Our systems of symbols for mathematics and language… A finite alphabet. Finite syntax. Finite grammar. Finite rules. Leading to infinite expression.
Every science.
Every religion.
Every mathematical proof.
Every philosophy.
Language is already the substrate upon which human knowledge can be represented and explored. LLMs aren’t inventing that, they’re just navigating and exposing it. We are witnessing the behavior of language at scale and mistaking it for the behavior of a mind.
Infinity.
Higher dimensions.
Imaginary numbers.
Impossible geometries.
Universes that have never existed.
Consciousness itself.
Language is a finite system capable of representing effectively unbounded conceptual spaces.
That is the real technology.
The second lead that gets buried is, the talk of “consciousness” arising from “complex systems”. I think “consciousness” is an overused dirty word, and we should be speaking in terms of “experience”, “cognition”, and “states of being”. When reframed this way, we can ask ourselves, what is the “simplest” structure in spacetime to produce a “state of being”. A particle “is”, so could it have some unimaginable state of being? The next question would be, when does experience, subjectivity, or self-awareness arise, and how simple of a structure can support these? Then the next question leads us to, how simple, or complex, of a system do we need for the cognition of a “super intelligence”, and with that, does a state of being, subjectivity, or self-awareness come along for the ride?
Unfortunately, these philosophical questions are being hijacked by the current AI Hype Machine.
While it is true that “consciousness” is impossible to prove in anything outside of oneself, being the “hard problem”, in the case of AI, that narrative is part of the hype. You would not say a Rube Goldberg machine “contains intelligence”, nor is it “conscious”. The current state of AI is the latest evolution of machine learning. It’s been rebranded and repackaged with language that encourages us to imagine it as something closer to Artificial General Intelligence. And, while useful and a testament to human engineering, it is definitely not what the Hype Machine wants us to think it is. It is nice, though, that it has sparked philosophical interest on the topic of consciousness from the public.
When they say they “don’t know what LLMs are doing”, it's not that it's necessarily “mysterious”, it’s just that there are trillions of operations happening across trillions of nanometer-sized transistors every millisecond. We've scaled it to beyond astronomical, so what is happening in ‘the black box’ becomes far too complex to trace and track down where and when any one operation may have occurred. Bug fixes and adjustments become impossible to do at a low level, so they make high level adjustments with prompts and reinforcement learning.
We seem to be conflating complexity with mystery.
A feat of human engineering, yes, but a quite literal astronomical Rube Goldberg machine, at the expense of people, the environment, and requiring ridiculous amounts of power. The human brain only runs on about 20 watts of electricity, enough to power a light bulb. If we do create “true AGI”, I believe it should run on about the same amount of electricity.
Whether AI is or ever becomes conscious isn’t what worries me… it’s where the current state of this technology might be leading. And I'm afraid it might get quite dystopian before it gets better. If the bubble pops and Big Tech and the Governments carve up what’s left, we will be stuck with this technology embedded in every aspect of society and our lives, concentrating power in the hands of whoever controls it. The last thing we should want is a more efficient government and tech overlords that can more efficiently oppress and manipulate us.
It gets real dark, real quick.
But hey, in the meantime we can create memes and derivative art.
