Generated Code

Today, when I was asked about the software we’re making in the university research group, I couldn’t answer well. I wanted to build something and wanted to see it quickly, so I had an AI agent do part of the work and only looked at the artifacts it produced.

But when someone asked, “Can you explain what logic it’s running on?”, I was stumped.

soumame

The output is there, but I don’t know why it’s there. Of course I can read the code, so I’d understand if I looked… wait a sec… (checked for about 30 seconds)

Two years ago, when I did Mitou Junior, this didn’t happen. When asked, the answers were in my mental drawers, so I could answer immediately.

But what about now? In my mental drawers, I only have the definition of the artifacts produced by AI-written code, so it’s not uncommon that I don’t know why they work (whether or not I could understand them).

Even though it’s an app I’m making alone, it feels as if multiple people were making it. Because not everything is in the palm of my hand, I sometimes have others build parts for me and I don’t grasp the details.

So, maybe I should be making sure I know those things myself, but I feel like I’m increasingly neglecting that. The world is changing so fast, and we think too much about just making and releasing things, so often the attitude becomes, “as long as it runs for now, it’s fine.”

soumame

From now on, will I proudly tell outsiders that the outputs that feel like other people’s work — the ones I had the AI create — were made by me?

If this AI agent were human, I might have eaten with them, had small talk, picked up even a faint sense of how they think, and could have entrusted work to them somewhat without checking. But AI feels like someone who just does what they’re told and leaves without saying much. So you don’t know what they’ll produce, and it ends with “well, it runs.”

As I’m writing this, I’m getting confused myself. Oh well. I’m going to sleep.