On Large Language Model Usage

I don't want to read your agent wall of text (or any, really). I want to read what you, a fellow human, have written. I don't want to listen to your agent's music. I don't want to view your agent's art.

Artificial Intelligence (AI) and Large Language Models (LLMs) are very polarizing topics in the PHP community, and I expect many others. In this article, I use AI and LLM somewhat interchangeably; unless noted, I'm referring to anything that comes out of an LLM or AI agent. You have the extreme ends of "This will replace developers" and "I'm never using it," and everything in between. What I've observed in these conversations can be distilled into a few important points:

• It's unethical because of how the models were trained
• It's causing more environmental pollution due to more/larger data centers
• It's copyright probably won't hold up in court

It's about Ethics

Ethics is one of the things that is missing from the technology industry. No one seems to care. Everything is focused on shipping value for the person paying you. We, as developers, are not bound by professional ethics standards, unlike attorneys or doctors, because there is no official process for becoming a developer. We start writing code and building things. The Association for Computing Machinery (ACM) does have a Code of Ethics and Professional Conduct that all of its members are bound to.

Many of the large commercial models were trained on content they did not have permission to use for these purposes, and that's what immediately turned many people into the "I'm never using it" extreme. They stole from authors, many of my friends, and even from me to teach their models to predict text better. Let's not forget these tools are only guessing words to give back to you. I've seen Larry Garfield refer to this as "Grand Theft Autocomplete," and I get a chuckle every time I see it.

A US judge recently approved a $1.5 billion settlement of a copyright lawsuit against Anthropic. This is only one of several lawsuits being rightly filed against the companies training models. Companies are also buying up old and rare books to use for training models. Most often, this involves them destroying the book. In the case of rare books, this is incredibly sad because it's not for the preservation of the content; it's feeding the machine.

It's about the environment

This one hits really close to home, since I live in a part of the US with an xAI data center. They've routinely been accused of using too many natural-gas-burning turbines to power it. The NAACP is suing xAI over it. Memphis is a poor city with much of the population below the poverty line, and it certainly feels like we're being taken advantage of.

The AI boom is driving a huge rush to build and expand data centers, and the problem is only getting worse. Many cities and towns are attempting to ban data center projects. Others are placing moratoriums on new data centers. You know it's serious when local news coverage is talking about it. All those new data centers aren't going to help the climate crisis.

It's about copyright

Copyright law is woefully behind the times. Laws have to be updated to be correctly applied as technology changes. You want to protect your creations with copyrights, but many argue you can't copyright anything that comes from an LLM. I'm certainly not a lawyer, and the legalese quickly drains my brain, but ultimately the concern is: if you create a product with an LLM, can you claim copyright? Wikipedia has a great article I will refer you to. In that article, you can find that the US Copyright Office requires a human to create the content being copyrighted but doesn't detail whether that is content entirely created from an LLM, or content that was partially created from an LLM. The UK asserts that content generated without a human creator is protected and that the copyright would be held by the LLM creator, for example, under a work-for-hire arrangement.

There have been cases where the copyright office has been sued for failure to issue copyrights on AI-generated artwork. The guidance they offer on works containing material generated by AI states that you must disclose the use and identify portions created by a human author and portions created by an AI. Based on my interpretation of the guidance, if you have an LLM create an entire code base, it's not copyrightable. If you, a human, have built parts of a code base and then used an LLM to create within that code, you can copyright it, but must disclose what came from the LLM and what came from you.

What do we do?

For the PHP project, I think it should adopt a policy similar to the Rust project's LLM Policy, which essentially says you can't use AI to create. Still, you can use it to "...answer questions, analyze, distill, refine, check, suggest, review...". It also states, "No one except the author is required to read LLM output unless they choose to." I think this is the important bit for those on the "I'm never using it" side because this gives them the option of being able to ignore those pull requests, issues, etc. It also provides guidance for those looking to contribute to the project using these tools. But ultimately, there's no way to identify content generated by an LLM, and that's a big issue. I firmly believe it's better to adopt this kind of policy than to put our heads in the sand, thinking this is all going to blow over.

As a writer, I found that AI was a reason I stopped writing my column for PHP Architect magazine. The inability to differentiate between content generated and human-written content is incredibly disheartening and compounded the stress of having to produce that content regularly because I refuse to use AI/LLMs to create articles for me. PHP Architect publishes great content from great authors, and I enjoy reviewing articles every month since I've been working for them for about a year and a half now. After writing ~70 articles, it was time for me to move on to something else.

My contract with you, dear reader, is that these words are written by me, a human. In the cases where I do use AI, I will disclose it because I respect your time as a fellow human.

"typewriter" by Ak~i is licensed under CC BY 2.0 .