Background Photo by Google DeepMind on Unsplash. Composite image developed by the author using Canva.
Manifesto Version 1, Published June 10, 2023
Introduction
I am beginning to write about my experiences with using ChatGPT and other large language models. I also will be sharing images generating using tools like Midjourney throughout this work to highlight some of the capability of these tools.
That said, before I dive into the use of the tools, I want to talk about the ethics. And, this is where I think everyone should start. Consider how you will use these tools responsibly and ethically. Draft your own ethics statement or personal manifesto. If you do, please comment or link to it or drop me a note. I’d love to read it and learn from you!
I chose to structure this in the form of my personal manifesto for ethical use of AI tools. The headings provide the summary but I also want to give you the reader as well as myself a touchstone to why I chose that specific statement or thought. This is a document I plan to consider and appropriately revise as I learn more and explore these tools and research about these tools in more depth. I was very pleasantly surprised to see all of the attention given to helping organizations like the MLA (see comments) further clarify what is possible and what is appropriately for citations. I will join these organizations in grappling with leveraging this evolving technology responsibly. When you add comments or questions, and when I conduct research and experiment on my own, I will add, edit, and delete sections of this document.
I avoided using ChatGPT or other generative tools in the writing of this piece. I conducted some Google searches to become more knowledgable about current research on these topics. Where this work is relevant to my manifesto, it is attributed. Otherwise, these thoughts reflect my personal beliefs and current understanding only.
1. Practice currently accepted citation standards, appropriate to that publication, when publishing ChatGPT and other AI output.
As I began researching my manifesto, I wanted to know what some of the preeminent academic and writing style experts say about the use and citation of AI tools.
1.A. LINK DIRECTLY TO THE OUTPUT OF AN AI TOOL WHEN POSSIBLE. ALWAYS STATE EXPLICITLY WHEN AN AI TOOL WAS USED, WHAT AI TOOL WAS USED, AND THE DATE I USED THE TOOL TO GENERATE THE INFORMATION. CONSIDER WHETHER APPROPRIATE TO SHARE THE PROMPT.
For much of my work, all of which is published in the form of blog posts and web articles, I link directly to sources rather than practicing a specific form of citation. That said, I will always be completely explicit about what text or images were generated by an AI tool, what tool was used, and the date I created the work. In many cases, I will also share the prompt I used to generate the work.
1.B. WHEN APPROPRIATE TO THE WRITING TYPE, USE THE CURRENTLY ACCEPTED CITATION STANDARDS FOR GENERATIVE AI TOOLS.
I conducted google searches to ascertain the thoughts on behalf of the American Psychological Association (APA), Modern Language Association (MLA), and the Chicago Manual of Style. These are three main bodies that I have utilized when publishing work in the past.
American Psychological Association - How to cite ChatGPT
Modern Language Association - How do I cite generative AI in MLA style?
Chicago Manual of Style - A Q&A on AI
The MLA’s was the most interesting to me simply based on the adaptations they appear to have made based on comments. I also really appreciated this line from the very loosely defined Chicago Manual Q & A page - “Many scholarly publishers are requiring its identification though also requiring human authors to take responsibility for it and will not permit the AI to have “authorship.”. That element of the human author taking responsibility for all use of the AI is one of the core reasons I decided not to publish any of my work about AI until I wrote this piece first.
2. Use the direct output of AI tools with intention and thought.
I use ChatGPT and other Large Language Models as well as image creation tools like Midjourney as thought partners and inspiration tools. I do my due diligence to take the output of the prompts I have these tools and turn those into ideas that reflect only my beliefs, thoughts, and knowledge of the source.
There are limited cases where it is appropriate to use the direct output of an AI tool.
2.A. USE THE DIRECT OUTPUT OF GENERATIVE AI TOOLS FOR PERSONAL KNOWLEDGE ENHANCEMENT.
First is when that output is solely for my personal research or knowledge. ChatGPT has largely replaced Google as a place I start when I want to research a topic. For instance, the other day, I was writing an article on debt consolidation and I wondered what the differences between a home equity loan and a home equity line of credit (HELOC) were. I’ve used these terms interchangeably but knew that wasn’t accurate. So, I asked ChatGPT that question. It wasn’t for the article - it was just a question for my own knowledge at the time since the question came to me.
2.B. REVIEW EVERY WORD, EDIT APPROPRIATE, AND ACKNOWLEDGE THAT THE GENERATIVE AI TOOL DID THE HEAVY LIFTING.
Second, when I ask ChatGPT to do some highly time consuming work for me, I will take the time to review every word of the text before publishing it with appropriate attribution (discussed more fully below).
For example, I am working on an article series and other content on job interviewing. I fed ChatGPT the following prompts in succession:
Prompt 1: Would you please provide me with a list of at least 100 typical behavioral job interview questions?
Prompt 2: Would you be able to reuse some of those questions and/or generate additional questions in groups of 10 questions each with questions specific to the following topics: Written Communication, Oral Communication, General Communication, Teamwork, Collaboration, Problem Solving, Process Improvement, Change Management, Personnel and Supervision, and Leadership? Would you be able to reuse some of those questions and/or generate additional questions in groups of 10 questions each with questions specific to the following topics: Written Communication, Oral Communication, General Communication, Teamwork, Collaboration, Problem Solving, Process Improvement, Change Management, Personnel and Supervision, and Leadership?
ChatGPT did exactly as I asked. I could have written this list of questions, but it would have taken a significant amount of time and I may very well not have generated as good of a list. In this case, I let ChatGPT do some of the more monotonous labor of the task for me. Before I will publish that list of questions, I will read through it myself and make edits and determinations about the groupings. I know from the first glance through the list that my edits will not be significant at least related to those areas. I will also fully acknowledge that anyone could generate a list of their own and will likely share the prompt I used to generate the list as well as which tool was used to generate it.
I am also considering the financial repercussions of using text generated with and in cooperation with LLMs and other such AI tools.
3. Take care to properly identify and compensate other authors and creators for their work whenever possible.
First, I want to fully acknowledge, I have not yet figured out how to do this. This section is definitely a work in progress.
3.A. BALANCE THE USE OF GENERATIVE AI CREATED CONTENT WITH THE USE OF PURCHASED AND LICENSED FOR USE WORK.
Large language models like ChatGPT and image generation tools like the popular Midjourney are invoked using a prompt of my design or choosing. The tool then generates text or images based on that prompt. What is returned to me as the output is based on whatever was used to train that model - a bulk of which is copyrighted and proprietary work.
This is the single biggest concern I have regarding using these tools. I must always ask myself - how will I use this output in a way that honors the integrity of the creators whose work I stand on top of? For now, I will strike a personal balance between using generative AI and purchasing and licensing work directly from others. A few ways I will go about this:
If I am using generative AI to develop content, I will make effort to keep generative AI content to less than 10% of that total work.
If I am trying to create purposeful images (logos, banners, etc.) used for marketing purposes, I will self create and/or pay fair market value for creation of those.
If my goal is to create content specifically for sale, I will use no more than 5% generative AI content in that content. The only exception to this is where AI is being used as an example in and of itself. For example, in a course where I am teaching about the use of generative AI, there is no choice but to include more than 5% of the work from the generative tool. In these cases, I will identify a suitable charity benefiting creators to donate 25% or more of the proceeds (not profit) from the sale of that content.
I don’t anticipate this will cover all potential cases and so I will also consider additional cases that they arise. As these generative tools evolve, this is an area that will continue to evolve. I am intrigued at the potential for blockchain and crypto to allow creators to license work for use in generative AI engines and then to receive a portion of proceeds from the work. I don’t know if anyone is working on this but I hope someone with the technical and business acumen to consider such an idea.
3.B. ASK GENERATIVE AI ENGINES FOR SOURCE ATTRIBUTION WHERE POSSIBLE AND DETERMINE IF DIRECT COMPENSATION IS APPROPRIATE.
This idea is inspired I by the article by the American Psychological Association on the current style considerations for citation of work generated by AI tools. The author suggested that you should ask the model what sources it used in generating the content and then review those primary sources as a part of your research process. The article suggests that this is specific to validation of facts, but it also has relevance to compensation of creators. I have already been asking ChatGPT to provide links to research it finds relevant on a specific topic. I will now update all of my prompts to ask ChatGPT to provide a list of sources it used for generating its response. I have no idea how that will change my search but it is a good practice to get into now as I’m still early in my exploration and use. The author of the APA article noted that this is another case where ChatGPT occasionally makes up facts which I’ll discuss in more depth in the final point of my manifesto.
4. Validate the accuracy of output of a generative AI tool.
The importance of the fine print on OpenAI’s ChatGPT tool cannot be ignored when using ANY generative AI tool. As of the May 24 version of Open AI’s ChatGPT’s free preview version, this is what that message looks like:
The text says, “ChatGPT may produce inaccurate information about people, places, or facts.”
My goal is to do no harm with my content production. That does not mean I won’t pose controversial ideas. It does mean that I will, to the best of my ability:
Ensure accuracy of all of the content I publish.
Highlight areas where I may be glossing over or leaving out key details for the sake of length, focus, or another reason
Conduct my own research and engage with text generating AI as only a writing and thought partner.
I will be responsible for reading and updating any content that I generate if using it near verbatim. Moreover, I will do my best, as noted above, to limit verbatim content reproduction. Any time verbatim content is used it will be properly attributed using the conventions of the medium.
5. Clarify my expectations regarding the use of generative AI if I will be assessing the work of others.
I am a teacher at heart. I currently teach one university course and two courses of my own design with several more in development.
Specifically in the university course, I will abide first by the university policies on use of generative tools as well as plagiarism. Should I teach for other universities that do not have their own policies, I will clearly state my policy in the syllabus. I will always use a first suspected or even verified “offense” as a teaching opportunity rather than as a punitive opportunity. As something “new” in the world, I want to lead with grace towards myself and others as we learn together.
In the courses I design and teach which do not carry credit grades with them, I will generally teach about the use of generative AI when it makes sense to the content of the course and will introduce the topic by suggesting participants draft their own ethics statement.
Finally, in places where I am purchasing or otherwise acquiring work from someone else, I will be clear about my expectations of whether generative tools are to be used and if so, what my allowable uses are as the person contracting the work and thus responsible for its use.
6. Continue learning about generative tools and the ethical considerations of their use.
This version of my AI Ethics Manifesto is version 1. I anticipate many, many updates in the weeks, months, and years to follow. These updates will be based on my own experimentation and learning as well as research from reputable sources.
📰 Want to make sure you never miss a story from me? Subscribe to my newsletter. I send out a fresh new article and do a round up of everything I published that week. It is released for free every Monday morning.
🌐 You may find other copies or versions of this article at places where I publish my work. You can find me at Medium, on the Hive Blockchain, on my personal website, on my business website, on LinkedIn, and on Twitter.