Why Make an AI Policy

It's impossible to avoid the question of generative AI while blogging about creativity and technology. I mean, it's impossible to avoid the question of generative AI while blogging about anything.

People have very little trust in the authenticity of what they see and read on the internet. As a content creator (much as I may hate that term, that is what I am), the trust you have in me is dependent on my honesty and transparency about if and how I use generative AI tools. To this end, Damola Morenikeji encouraged bloggers to create an AI manifesto on their website, so readers would know exactly when they encountered any AI-generated material.

My Stance

I am not all or nothing when it comes to AI. The truth is, I'm not concerned about what people do in their own homes on their own computer. But if you put AI-generated content out into the world (or use AI in any way to create content to put out into the world), it carries more power, and danger. Bad actors leverage its power for their own agendas; other individuals, out of ignorance or malice, recklessly bring in their own AI contributions without considering the costs. On the other hand, I believe there is a way to adopt AI responsibly and ethically, both individually and as a society[1].

The following are some of my biggest concerns about generative AI. These issues inform a lot of my own practices, but this is no way an all-encompassing list of everything I find worrisome.

Misinformation

As social media has become flooded with AI-generated content[2], it has become a scourge of misinformation[3].

And it's not just about generative AI, either. The corporate web is transactional – the product is our data. Our posts, our likes, our follows, our browser history, our searches, etc. Companies who want to sell us something use our personalized data for targeted advertising and personalized marketing. Other entities use this information to try to rewrite the "truth" using computational propaganda ininternet manipulation schemes.

Sloppification

AI-generated content is generally of low-to-mediocre quality. And while humans are plenty capable of producing terrible content, the ceiling for human-produced work is far above what AI is capable of.

What AI offers instead is quantity. If you can churn out a lot of mediocre content really fast, you can stay at the top of the algorithm and you can fill social feeds.

Crowded-Out Creators

LLMs trained on creative works then generate outputs that mimic the style of those same works, and this devalues the work of creators. Why pay for something if you can prompt a machine to do a bastardized version for free?

This "crowding out" doesn't just affect artists who sell their work. Independent content creators suffer as well. Today, when you search something in Google, its embedded AI will greet you with an answer before it returns a list of search results. Websites that would have once pulled your attention no longer do[4].

Hallucination and Limitations

LLMs are known to produce hallucinations – to present false information as fact. All too often, people do not verify the information the AI provides. That doesn't just apply to the average person on the internet, but highly educated people performing critical functions[5].

It is important to recognize that LLMs rely on humans, and the quality and accuracy of their outputs depends on its training data. Generative AI is a recycling machine; it cannot be used to create or learn anything new. As humans we must do that ourselves.

How I Use Generative AI

Now, to be clear: all content on this website was written by a human (namely, me, unless otherwise stated).

That said, I do occasionally use LLMs (Large Language Models) in the process of maintaining this blog and creating new posts. I want to be clear on exactly how I do and don't use these technologies.

Key
This is something I do, or that I don't restrict myself from doing
This is something I don't do, and that I will never do on this site

Research/Planning

Yes/No What
Using an LLM, like Copilot, to look for sources
Using answers from an LLM as a source itself

Writing/Editing

Yes/No What
Using generative AI to draft posts or content at any stage in the writing process
Using AI tools for small edits/proofreading, such as punctuation, typos, and spelling and grammar errors
Using AI tools to analyze my writing for editing purposes
Using AI tools for rewrites or revisions, even small-scale revisions such as sentence restructuring

Coding

Yes/No What
Using a chatbot to help troubleshoot coding issues in addition to seeking answers from other sources
Asking an LLM for code examples, which may involve copying small blocks of code (<= 5 lines)
Copying large blocks of code verbatim from an LLM

Images/Media

Yes/No What
Generating images, videos, audio, or any other type of flat media file using generative AI
Using AI to "enhance" low-quality media (called "upscaling")
Using AI to isolate an element in an image in order to perform routine edits to a specific area (i.e., background removal)
Using generative AI to stylistically add flourishes or stylistic edits to media

Footnotes

  1. UNESCO adopted the Recommendation on the Ethics of Artificial Intelligence in November 2021. ↩︎

  2. AI Content Is Everywhere on Social Media, Especially LinkedIn, Pangram ↩︎

  3. The Misinformation Accelerator, Science ↩︎

  4. Nilay Patel from The Verge and the Decoder podcast has a name for the day when Google stops sending traffic to third-party websites altogether: Google Zero ↩︎

  5. Damien Charlotin maintains a database of legal cases where it was found that generative AI produced hallucinated content. To date there are more than 2000 cases. ↩︎