Why I’m opposed to AI writing but not AI coding
Someone asked me recently why I’m opposed to AI writing in blogs and newsletters but I’m fine with it for generating code. This is my attempt to answer that question. I’m fine with AI coding because no one cares what the code actually looks like except for other software engineers. The customers don’t care. The managers don’t care. They want the software to work. They want it to look good and be fast and reliable. They don’t care how the sausage is made.
Code written by an AI can meet the criteria above as long as a real human engineer wrote the specs and enforces the tests. Of course if you let the AI run wild you can easily end up with AI slop that has only the veneer of quality software, but that’s a discussion for another time. AI, thoughtfully applied, can make software better, or at least faster to develop.
AI writing on the other hand is bad. I hate it. I hate it with the burning passion of a million flaming GPUs.
Writing is Thinking
Writing is thinking. If you didn’t write the words yourself then you didn’t think about it. When I write it is a process. I have to put down the words and edit them to really think something through. Even if I did a bunch of research first, handing that research to an AI tool wouldn’t result in the same ideas as when I write it myself. I have to work through the idea. The writing process sharpens the ideas and smoothes the prose. Without the writing process what you end up with isn’t writing. It’s just cheaply outsourced ghost writing. Slop. And this leads to my second point.
If an author didn’t take the time to really write something, to actually think about what they are trying to say, then why should I bother to read it? I take pride in my writing. My blog voice is different than my corporate or academic voice, but they are always my voice. I’m putting my name on this writing. If I didn’t actually write it then I didn’t actually think it. If I didn’t write it then the output doesn’t actually contain my thoughts, so why should anyone else bother to read it.
AI as Assistant, Not Author
I can see a few uses for AI when engaging in the writing process, just not the writing part. Collecting research, for example. When I recently started working on my career portfolio, I used an LLM to help me collect all of my work. I have gigabytes of documents scattered across my hard drive from 30 years of programming. Having the LLM help me organize and pull out examples was very useful. It did produce prose as part of the research, but I immediately deleted and rewrote all of the prose because that’s the thinking part. Even with data collected I still have to think though my projects, decide what is important and what should be left out. AI let me focus on the really important parts, the curation and reflection. In other words, the thinking part. No one, and no thing, can do that for my career other than me.
So back to software. Generating the code was never really the thinking part. The process of working through some code issues and the rapid iteration certainly is an aide to thinking, but so is the writing of unit tests and creating requirements documents and architecture diagrams. I have found that what I miss from the writing of code I get back through the writing of documents and the rapid iteration of testing. I’ve been able to write React based analytics tools in a tenth the time by not writing the code, but this only works because I am specifying the user interface. I am making requirements for storage and latency. That’s where the thinking part is, not in the writing of the code itself.
Human Ownership
I understand it’s easy to get lazy when using AI to generate code. It’s happened to me too. What keeps me honest is that I put my name on the code. It doesn’t matter if the bot generated it or my fingers. I own it. I’m responsible for it. If it breaks, I’m on the hook. If someone asks how it works then it’s up to me to explain.
This has caused me to think again about the OpenAI attacks on Hugging Face. If a human had done this it would be criminal, so why weren’t charges filed? While it wasn’t malicious, someone at OpenAI allowed this to happen. They didn’t expect the AI to do what it did, but they put it in an environment where it could escape. They can’t blame the AI anymore then you can blame a car for being defective. Perhaps it should be negligence instead of malfeasance, but a human somewhere must be held responsible when AI does bad things.
What do you think? Should AI be allowed to be an autonomous corporation? We must always keep humans in the loop, and legally responsible.