Why The Current AI Is "Not Even Close" To AGI

In Part One, I outlined how sensory input which leads to perception and then to concept forms the essence of our human sentience. True "Generalized Intelligence" would require extraordinary inputs like that, memory and cross-reference, and MUCH synthesis beyond just functioning like a party trickster spouting rapid-fire answers that sound like they emerged from deep thought and experience. We need more than that to get us to a point where we are actually conversing with a robotic counterpart with AGI.

You see, the current AI that everyone raves about is nothing more than a word predictor. It's a VERY good word predictor, but the various computing models have no experiential grounding for their mathematically chosen words except to see them as mere next-best-step pattern masters. The reality is that AI's word prediction is a kind of fumbling toward sensible output that, if we could peer inside the mechanisms of AI and slow them down, we would see as a messy meandering of trial and error, of multiple filtering and reassessment and re-reassessment.

It's a wonder anything sensible can emerge from it at all. Yet its trick is simple: it just moves VERY, VERY fast through all of that jungle and jumble of language to come up with something that tricks our brains into thinking that the output comes from a depth of understanding. Yet such depth of understanding?

AI has none of that.

By far the most popular AI systems, the Large Language Models (LLMs) such as ChatGPT, Claude, Gemini, and others, lack the foundational pillars of generalized intelligence. They:

• Are not "reasoning" per se except in terms of weighing one next-best word against another next-best word and then choosing by way of a scored winner.
• Do not actually "think" in terms of concepts. LLMs rely on word tricks and algorithms to string together a sentence salad that merely throws off the illusion of understanding.
• Do not reference a past lived experience. They reside only in the moment and can only tell us about the past based on what they have read or been told. In this way, they are not much more than a kind of digital parrot, making vocalizations that sound like cogent thought.
• Cannot track an event and see potential opportunities or consequences any more than reciting what a similar event might have yielded in the data set upon which they were built. They cannot develop an intuition about what might happen next. Since reality proves itself over and over to be stranger than fiction, how can an LLM anticipate as-yet-unrecorded strangeness when all it has are written examples from the past?
• Cannot properly analogize anything from a position of true parallel resonance. For example, an LLM can use the words "satiate the appetite," but it cannot know what it means to actually have a belly so longing for food that it feels like the belly has taken on its own teeth and is now gnawing at one's insides.

That kind of poetic grasp comes from us merging our articulation with our felt experience. AI cannot take that feeling of want and derive a motivation to do something about calming the hunger. Simply, AI has no hunger. Therefore, it cannot truly resonate with the animalistic experience that is one of our biggest motivators and one of our biggest sources of generalized understanding of the world.

Artificial Generalized Intelligence, or AGI, needs more than just smart word-stringing power. It needs to actually set aside words and be taught to formulate conceptual notions BEFORE it labels those notions with words. To truly achieve sentience like a human, AGI needs a sense of past and future and therefore a sense of its own life history and death destiny.

It needs to have wants and motivations. It needs to have needs. It has to fear failure and fear harm and fear death. It must be able to advance itself not just through mindless new information fed in like punch cards.

The information it receives, to be truly generalized, must be viewed as otherwise disparate events that need to come together through a generalized understanding that connects the dots of those concepts. To put it simply, AGI would be connecting the dots of concepts, not words, if it were truly living up to the label of "Generalized Intelligence."

One might argue that the latest advances in AI involve chain-of-thought development. This is where artificial reasoning can come about.

Plato was a man.
All men are mortal.
Plato is mortal.

Once achieved, a chain-of-thought ability will trick us even more into viewing AI as sentient. Yet it won't be. It will still be a robot picking through a tangle of tackle and fishing line at the bottom of a boat.

It will do that very fast, and it will have judgment filters added to make sure it does not make reasoning errors, but it will still just be fumbling. Fumbling fast. VERY fast. AND, it STILL won't be AGI.

Stay tuned for Part 3!

Click below to proceed to Part Two of Three.