Written in 2023-10-01. Just wanted to archive my old thoughts.

Today I watched a video and what struck me the most was this insight: In the end, what AI does is pattern recognition.

“If I had to explain what AI does in one sentence, I’d say: AI discovers latent patterns.
In any domain where such latent patterns exist, AI will soon surpass human ability—and eventually replace humans.”
From the video

This might sound like an oversimplification, but I believe one of the main reasons humans have progressed this far is our ability to recognize and classify patterns.

Ray Dalio refers to this as principles.
I’ve expressed it in terms of template-ization.

Of course, all living beings operate based on inherited instincts, but I think what sets humans apart is our ability to share subjectively constructed patterns through communication. (In fact, I believe the ultimate goal of all learning is the communication and sharing of knowledge.)

In other words, we interpret the complex world around us in a way that aligns with our goals (survival and reproduction), then we:
→ compress that interpretation
→ recognize patterns
→ build frameworks (or theories)
→ communicate those frameworks
→ and increase our collective knowledge and efficiency.

This is how I believe human civilization has developed.
But if AI can now learn these patterns better than us—especially in the case of unsupervised learning, where AI can discover patterns even humans can’t— then it will dominate in any domain where things can be patternized.

In the end, anything that can be patternized will eventually be replaced by AI.

This leads to a fundamental and chilling question:

Is there anything that cannot be patternized?

If the answer is “no,” then ultimately everything is replaceable by AI.
And honestly, that seems highly plausible.

The world is often described as a complex system.
It’s not governed by rigid rules like the game of Go, but it’s also not pure chaos with zero structure.
It lies somewhere in between—a blend of structure and randomness.

In other words, the world contains both regularity and probability, and from that standpoint, I’d argue that nothing is truly patternless.

That doesn’t mean AI will perfectly solve everything (after all, probability deals with trends, not certainty), but at the very least, it’s likely to far surpass human capabilities.

Even setting aside AlphaGo—where the rules are formalized—consider autonomous vehicles.
They still get into accidents, yes.
But humans get into far more.
As noted in a 2018 article, “37 out of 38 autonomous vehicle accidents were due to human error.”

Conclusion

So if AI potentially has the ability to recognize patterns (or, in my terms, build better templates) in every aspect of life, then I seriously question whether I should keep manually reverse-engineering, building, and refining those templates myself.

Wouldn’t it make far more sense to design an AI that can build those templates for me?

That’s something I’d really like to research and experiment with further.