AI and the end of understanding

AI and the end of understanding

Lost in the hype of what AI can do for us is what this technology might end — understanding itself.

Humans are machines that were built by forces devoid of intelligence. Our biology, as opposed to culture, is the product of a machine learning that operates without any understanding of what it is learning.

Now humans are also creating machines that are a black box to us. When a machine learning chess AI gives us a list of top moves, we don’t know how that list was reached or why those moves are preferred.

Unlike most of the history of science up until this point, no human has painstakingly mapped the underlying principles or plans to do so.

Where scientists have gone to great lengths to see and try to understand the molecular world from gluons to mRNA, machine learning could just tell us how to do things without revealing why the approach it recommends works.

Human power could increase dramatically without any expansion of human insight. The age of understanding could end.

Of course the age of a single human understanding has already ended some time ago. As fields of specialty became smaller and smaller, there are many systems that no one human fully understands. The clothes you wear are the output of agriculture, materials science, and manufacturing that require dozens if not hundreds of humans to fully understand. For each different question you have to ask a different expert and even he would consult a book or database.

However those experts exist and they had the fun and satisfaction of learning the details of their respective fields. Going forward those details maybe abstracted away.

Is there even such a thing as a chess expert anymore? As agadmator’s chess channel’s commentator says, “We know this is the correct move because our digital overlords tell us so.”

A person could still devote his life to understanding chess theory if he wants. Perhaps someone might consult this human chess theory expert eventually or read their book, but perhaps not.

AI is not the first time computers have abstracted theoretical details. Computer assisted proofs go back at least to 1976 and the four color theorem.

Mathematicians still have the final say on what is proven. A logic checker AI would usurp that role; the logic of the proof then verified by AI. That doesn’t mean an AI that can create the proof is coming any time soon, but imagine such a machine was built.

Ask a true or false mathematical question and the AI prover submits a solution to the AI logic checker and, if successful, your question is answered without you understanding why that answer is correct beyond the machine says so.

Human mathematical expertise would then be reduced to following along, much as we can now marvel when two high level chess AI play each other. Though sometimes as we watch the why of a computer move does become apparent.

To be fair our own brains are also black box machine learning devices. When chess master Hans Niemann beat world champion Magnus Carlsen he couldn’t articulate how he did it and so said, “The chess speaks for itself.”

That answer didn’t satisfy Carlsen and they ended up in court after Magnus accused Hans of cheating. Similarly in a world without human experts and understanding already very low trust levels will plummet further.

We will once again live in an age of magic where even the hope of one day understanding will be gone.