Reading the article from the perspective of a non-mathematician the line that rang true for me was "I realized they derived joy, satisfaction, and meaning from the long journey toward understanding." My eldest child loves math and her whole life has been chaos except when she is solving math problems. I see that human element in science and I know deep down we need that.

I've been in the software industry for 30 years and I understand that sentiment perfectly on a personal level. However, also having used ChatGPT for the first time this year to help solve a technical problem, I was surprised to learn that feeling didn't go away. In fact, it was because I leaned on my experience and technical understanding to get to the point where I realized I needed help that I decided to try a new tool. I didn't feel any shame or disappointment in myself, rather I felt excited to learn something new that came out of the solution. It sent me on a new path of learning.

Now, that was research and not implementation. While plenty of code options were presented by ChatGPT, I analyzed the solution and educated myself from it. My final fix looked very different from the proposal because I do things a certain way, based on experience and learning from my many mistakes. In this scenario I was the secondary verification. My peers the tertiary. AI made the proposal, humans did the verification and could not have done so without cumulative knowledge and experience.

I have to assume that all fields utilizing AI will remain as they have always been, human education and experience will come first no matter what the tools available, because we are the ones impacted by the data produced by AI. As many math-oriented commenters here have already noted, human verification is a necessity and to do that, you must understand the discipline within which the data is being produced.

Personally, the idea of reaching solutions in math and computing (for example) exponentially beyond human capacity is exciting; I want certain answers before I die! But it still must be human-verified and the solutions should be for humans, not for machines, and not for "time to market". Repositories full of unvetted AI-generated code is bad enough, but once you start engineering structures, spacecraft and medicine strictly with AI, well...