Authorship Tags

It looks like a duck, quacks like a duck, but it’s not a duck. AI went from nowhere to everywhere in a hurry. No one asked if we wanted it, but someone else’s investors say we need it.

As a longtime software engineer, I see AI through the lens of automated utility. Computing for decades and decades is in service of automating away repetitive tasks. Over the last century, people have grown accustomed to having some kind of computer within reach. From the calculator in the kitchen drawer to the one on your phone, computers do the boring stuff quickly. AI is one strange calculator, though.

I’ve been writing software for fun and profit for a long time now. I have plenty of thoughts to share on AI. The most universal one I think on is the human-focused “meaning” of this tool. Should AI be allowed to write, or not? Is it art, or is it how you use it?

How do I know who made this?

“AI” circled and crossed is becoming a more common sign I see. I see it on YouTube channels, playlists, and videos. It has been appearing more in music titles. Absolutely no AI was used in the making of this.

In the office, as a software engineer, I write pages and pages of content. Documents meant for immediate consumption and some meant for long-lived systems. Some are meant for people, and more and more are becoming for AI.

How do you know if I wrote it? Does it matter?

It does matter. People are still my audience. The way I wrote code for so many years was for a human audience first and machine second. I write my documentation with the same intention.

Things get fuzzy when the machine is the audience. Part of my role is reading lots of code. I review a lot of it, too. There came a point along my career when looking at code put me at the terminal in the Nebuchadnezzar. I don’t even see the code anymore. What slows me down at the keyboard is getting first-context.

There is a non-trivial cost in time for me to become familiar with parts of an app that’s ever-changing. That cost is paid by me as anxiety and pressure when there is an emergent project. I feel like it’s my fault I can’t go fast enough to answer questions for an emergency I didn’t start. What a shame to feel shame.

AI has proven to be effective at getting me a first pass of context for very complex things. It’s not perfect. AI quality can vary and for various reasons. After first-context, I write the people side of communication. The upfront truth is I like writing, so doing this task does not bother me and is satisfying. The reason to do it is I have to read the documents that AI wrote and verify the truth.

I only trusted the machine enough to provide deterministic outcomes. I read lots of its code or text after that. Things change fast with AI in the workplace now, but for now, a large part of my role is that my name is a warranty. If I signed, it means I guarantee it. I put the same level of rigor behind reviews of my junior, midlevel, and senior peers’ code and documents.

Often, my audience does not care if the AI wrote the document or not, just that it’s useful and accurate. I still need a way to communicate, and quickly, what my involvement was on the work.

Same need for videos or music or posts or writing or whatever. What is the level of authorship in the media?

In software, this is solved with a spec. In practice, it means written rules for a convention. They’re never perfect, and the specs inherently know that and are usually updated as needed. Specs are useful guides for conventions and shorthand.

To try to better communicate how a piece was generated, I thought of the Authorship Tag. They are a text label that could be applied anywhere. In a video title, the header of an essay, the footer of a blog post, wherever an expressive work is shared.

An example would be this post. I’d label it as A0 to indicate it is fully human authored. I used standard proofreading tools like spell check, grammar, punctuation. These tools are changing too.

The current version lives at the Authorship Spec.