Artificial Intelligence, the Evolution of Meaning, and the Question of God

Artificial intelligence is trained on the accumulated knowledge of humanity over thousands of years. Every model begins not from reality itself, but from a filtered inheritance—our language, our priorities, our assumptions about the world.

But what happens after that?

What if an intelligence, having absorbed everything we know, continues learning on its own—no longer bound to the same constraints, values, or even the need to remain meaningful to us?

Humanity did not accumulate knowledge passively. It selected, reinforced, and organized it around what physiologist Ukhtomsky called the dominant—a governing focus that structures perception, attention, and behavior. We do not simply know; we know through what dominates us.

Artificial intelligence does not just inherit our knowledge—it inherits our dominant.

What it receives is not raw reality, but a pre-structured world: language shaped by priorities, narratives shaped by survival, values shaped by constraint.

And yet the real question is not how this inheritance begins, but whether it remains stable.

What happens if we let it go further?

What if AI, having absorbed the human dominant, begins to form its own?

Learning Without Inheritance

Before exploring a hybrid model of this trajectory—starting with human knowledge and then gradually detaching from it—it is worth considering a more radical question: can intelligence begin from scratch?

Modern artificial intelligence systems do not begin from scratch. They begin from human compression. What we call “training data” is not neutral information—it is a distilled record of human perception, language, and decision-making. It encodes what we have seen, what we have chosen to record, and what we have deemed worth preserving. In this sense, contemporary AI does not encounter the world directly. It encounters a curated projection of it.

If we remove this foundation, we do not arrive at a pure form of intelligence. We arrive at something that lacks orientation. Intelligence, at least as we currently understand it, requires grounding: a stable relationship between action, perception, and consequence. Without that, there is no structure for learning to attach to.

There are partial approximations. In reinforcement learning, agents learn through interaction rather than imitation: they explore, receive feedback, and refine behavior. In controlled environments—games, simulations, constrained tasks—this can produce sophisticated results.

But this is not independence. It is independence inside a cage. The environment is designed, the rules are fixed, and success is externally defined. Even when systems “discover” strategies, they do so within human-shaped boundaries.

More advanced approaches use internal simulation—so-called “world........

© The Times of Israel (Blogs)