My Chat With ChatGPT About Using Ai Content
At the last PYPT the subject of AI came up a few times and generated some interesting discussions. To me, it’s an important conversation to have. Like any emerging technology there are a lot of concerns raised on how it will integrate into existing systems and what implications it has for our lives in general.
AI has the potential of finding integrationg into many areas of our lives. Like any technology, it can be used as an important tool or it can be used poorly. As our understanding develops, we need to learn how best to use it and how not to use it. I sat down today to ask ChatGPT a series of questions on the topic.
I don’t want to just verbatim drop the responses in here so I’ll largely paraphrase and summarize the responses to capture the important points.
How Does ChatGPT Learn?
ChatGPT is trained through a neural network by exposing it to a vast amount of text data. The goal of the training is to teach it to use the data to learn patterns and relationships between words and concepts. It’s trained to predict the next words in a sequence based on the context. The process is repeated millions of times with parameters adjusted over time to improve the outcome.
Where Does the Data Come From?
ChatGPT learns from a wide range of topics and concepts so it can form a broad base of knowledge. This data can come from books, articles, websites and other types of data.
Does ChatGPT Plagiarize Content it Has Learned From When Answering Questions?
No, it doesn’t repeat verbatim what it has been taught. It uses patterns and relationships it has learned to create new responses to questions. Much like you should do if you use material as a reference when researching. Material should be in your own words and style of writing.
ChatGPT draws on it’s learning and understanding from it’s training to answer questions. Some words or phrases from original text may end up being used but only for consistency within the context and meaning of the question not as copy/paste from the original.
Who owns the copyright to the responses ChatGPT delivers?
The underlying data used to train ChatGPT remains the copyright of the original creator. The specific answer generated by ChatGPT are considered to be a new creative work and the copyright may be subject to laws and regulations depending on the jurisdiction. Ultimately this may require a case-by-case analysis to determine the definitive answer as this is an emerging area.
@hivewatcher’s Post Note
After PYPT ended on Thursday a link to a post from @hivewatchers was dropped into the chat as a reference. In their post they referenced a question to ChatGPT if it was fraudulent to use AI generated text and claim to be the author of the text. I asked the question of Chat GPT and then some followup questions for clarification.
The response mentioned “unless you make it clear that the text was generated by AI and you are not the original author.” Referring back to the question above where it said the specific answer to a question generated by ChatGPT was considered to be a new creative work.
This was a bit confusing. If it was a a new creative work who is the new work’s author?
IF someone using AI generated text without indicating it is AI generated content is fraud does that mean the AI holds ownership?
It’s response was no, the AI doesn’t hold ownership, it’s merely the tool to create the work and doesn’t have legal standing. The ownership of the intellectual property rights, including copyright, would reside with the person or entity creating the original work.
I’m still not clear on who would own this new creative work. So I tried another question.
If I create the prompt to which ChatGPT responds, am I the person or organization to generate the text?
Yes, it replies. If I create the prompt I’d be considered the owner of the generated text as the prompt provided the original input or context for ChatGPT to respond to. The prompt itself would be considered a creative work I’ve authored.
Okay, that seems pretty straightforward, it’s mine. Maybe not, read on.
The ownership of the generated response is separate from the ownership of the prompt and may depend on several factors. The specific circumstances in which the response is generated would influence if I did indeed own the response.
See where this is going here? Without established law on this emerging field, this could be a minefield. But, I pressed on and tried to get a more definite answer.
What person or organization is the operator of the ChatGPT system?
ChatGPT is owned by the OpenAI research organization. It owns the intellectual property rights to the technology underlying ChatGPT.
Okay for the purposes of this interview I’ve now established what ownership OpenAI claims but I asked another question to be sure.
OpenAI owns the intellectual property to the ChatGPT system. Does it also claim ownership of the content created by the use of ChatGPT?
I’ll quote from the response here:
OpenAI does not claim ownership of the content created by the use of ChatGPT. According to OpenAI's website, "any output or other results generated by the API [including ChatGPT] are owned by you [the user]". This means that the user who generates the content using the ChatGPT system is the owner of that content.
Okay, now I seem to be back to I own the content I generate when I ask a question on ChatGPT. But, I continue to explore.
If I generate content from the ChatGPT system on OpenAI's site, am I considered to be the original author?
The answer is, if I generate the content from the ChatGPT system on OpenAI’s site I’d be considered to be the author of the content as I play a significant role in shaping the final output.
Now, that seems pretty definitive, right? Well, there was this little comment in the response which said, “depending on the specific circumstances and the applicable legal framework”. Now, what does that mean? I press on.
In the absence of a legal framework on the use of AI generated content would content I generate from ChatGPT be my original works?
It’s response was lengthy but boiled down to the determination of ownership of AI generated content may be subject to various laws, interpretations and challenges. Are you getting the picture here yet?
One of the comments made in the answer caught my attention: “ the resulting output may contain elements that were not specifically authored by the user.” — Wait a minute, I thought I dealt with that up above. Let’s try again.
You earlier told me that ChatGPT's output didn't plagiarize the creators of the text used in training ChatGPT. If that is the case, what resulting output from prompts generated by me could contain elements authored by others?
I was correct that ChatGPT earlier had said its output doesn’t directly plagiarize from the texts it was trained on. Then it repeated the comment about the resulting output, leaving me more confused. I tried a bit different approach.
Would the resulting outputs and elements not authored by me be sufficient for plagiarism to have occurred? Or would the result be more like taking notes and writing on a topic?
ChatGPT agreed that in a sense the generated output would be similar to taking notes on a topic and then using the notes to write an essay or article. The output may contain some repeated elements for consistency but they would not be a direct copy of unattributed quotes and would be an original creation.
Are You Getting the Picture Here?
There is a multitude of reasons to not attempt to use ChatGPT or any AI tool to generate the body of your posts. The biggest one is the potentially legal minefield of who actually owns the work.
Everyone of the questions I asked ChatGPT included a clearly worded suggestion that you should always be transparent about using content generated by ChatGPT. It doesn’t present itself as a tool to replace your own ability to write content.
So What Use is ChatGPT Then?
I asked it for recommendations for a person creating content with its help. It came back with some interesting ideas:
- Brainstorming — have it generate ideas and suggest angles to explore the ideas
- Research — asking qeustions on a topic or asking for a summary of concepts to make your own notes from
- Outlining — Ask it to generate an outline on a topic or question or writing prompt.
- Editing — after you’ve written a draft, ask ChatGPT to suggest improvements of sentences or paragraphs.
- Titles and Subtitles — craft a prompt describing your post and ask ChatGPT to suggest some titles or subtitles to use.
Note, these ideas use ChatGPT as a TOOL, not as a replacement for you doing your own work. You also have to be aware that like any other tool it has drawbacks. It makes mistakes, it can incorporate biases and discriminations from it’s learning material and could generate material irrelevant to the purpose of your post.
Do your own work and where you have drawn on AI for your results, be transparent.
Shadowspub writes on a variety of subjects as she pursues her passion for learning. She also writes on other platforms and enjoys creating books you use like journals, notebooks, coloring books etc.
NOTE: unless otherwise stated, all images are the author’s.
Some of the image work may have been done in Midjourney for which I hold a licence to use the imges commercially.
How to Connect With ShadowsPub:
Twitter: @shadowspub
Instagram: shadowspublishing
Medium: @shadowspublishing
Publishing Website: Shadows Publishing
Nicheless Website: Nicheless & Loving It
(Podcast & subscriptions for: Prompt A Day, PYPT Reminder & Newsletter)
Pimp Your Post Thursday (PYPT):
join us on the DreemPort Discord12pm EST Thursdays
Get eyes on your content and meet new friends. Join DreemPort.
https://twitter.com/4815956673/status/1632506702447648770
The rewards earned on this comment will go directly to the people( @shadowspub ) sharing the post on Twitter as long as they are registered with @poshtoken. Sign up at https://hiveposh.com.
Congratulations @shadowspub! You have completed the following achievement on the Hive blockchain And have been rewarded with New badge(s)
Your next payout target is 21000 HP.
The unit is Hive Power equivalent because post and comment rewards can be split into HP and HBD
You can view your badges on your board and compare yourself to others in the Ranking
If you no longer want to receive notifications, reply to this comment with the word
STOPCheck out our last posts:
i had several interesting conversations in commments today regarding AI art... and how people really need to understand exactly what happens when people create that art... and how long it takes and how hard it is to learn to get it to actually DO what you want - and how it actually COSTS money to own the images and so on heehehe
and it was really nice to see that there were a lot of people who were more open to listening and discussing and not just lumping all "AI" into the same boat as plagiarism or lacking creativity...etc!
I know how you feel about it - because we share the same views !!! LOL so just - FYI - nice to see that there are still logical and open-minded conversations going into all of this - and not just a big AI dump!
as far as ChatGPT - I can't agree with you more (and most of the community here!) that come on - it just doesnt' even take much consideration there to see that someone asking AI to write a post for them and then just submitting it - is just downright wrong. lol
but - will be good to see how everythign settles down and people start to understand more of what's happening, what they're using it for, how a lot of actual WORK is involved with Ai Art, and how it does not replace the human element of art - at all. but can actually be a tool that many artists use too!! hmmm... well - we shall see! hehehe
also - your last tag - you have a spelling error there hehehe
yes we definitely agree with each other on the work that goes into the art. I've been weeks exploring an learning how to construct prompts to get the results that match the vision. There is also skill in crafting prompts for ChatGPT, that is why where is a new profession emerging called Prompt Engineer.
Thanks for the catch on the tag... solverbloggers.. maybe we need that community for solutions LOL
ROFLLLLLLLLLLLLLLLLL YES!!!!!!!!! a new help community hhahahahahahahahaah
and yes - i tell you new opportunities open up everywhere!!! we CAN make this work for us if people are just willing to keep open minds and work diligently and be fair and equitable.
maybe some of us will just keep trying to lead the way! hehehe
I'm starting to repeat myself but this is something overlooked by seemingly everyone in this Hive discussion. You can "collaborate" with ChatGPT by using something most people call "Data priming". In other words, you do the research yourself and compile notes about it, list facts, input data, and give all of that to the AI and then tell it to construct an article or essay or a video script or whatever you need. This way you avoid plagiarizing other people's content 100% as the AI is using your data and your words.
I've not actually come across that being done but it's certainly possible with the right prompt engineered to handle it. That's something a lot have missed is that with enough skill creating a prompt, AI could be being used and would not be detectable.
One of the huge downsides to trying to create policies against an emerging technology is that the skilled users get away with going around them while the casual user gets caught up in it.
There will always be those who use emerging tech to abuse other systems. The challenge is to deal with them without hurting others.
It's definitely fun to play with. I don't ever foresee myself publishing any words not written with my own fingers though. Well, maybe a STT on occasion for a message, but that's rare as it is. I do have fun with the graphics they generate though.
I agree. I've use it to collect data, ask questions about details, create outlines, titles etc. In other words as a time saving tool. Any 'content' I've seen it output for the most part I'd not want to publish under my name without an extensive rewrite which then makes it mine.
The other issue is that there are certain whales that feel compelled to get rid of it completely off of the blockchain, and the other one is another issue altogether, but with the same idea, where I totally agree with your view, that it is nothing more than a tool to use to aid in better posts.
Another one of my concerns is that the tools for detecting the AI usage are not all that accurate at this time, and there will no doubt be massive false positives that will occur.
Also, the people that are not English first use them to help with their posts as well, and I am not sure what the full trigger is to set off a detection tool's results.
I love your list of legitimate uses for it, from making suggestions and outlines, that is an excellent purpose. I'm with Witty, I would never use this to make any of my text for me, my writing comes from my own hand with no other input.
Beautiful that you took this on and brought it out into the open more, it is needed.
Thank you Shadows.
You know, might be just me but, is there not a huge irony that some whales want to ban an emerging technology from use on a platform that is built on emerging and sometimes controversial technology?
A complete ban is not only pretty much impossible but will hinder continued growth of the platform. I know some of the whales consider content creation a nuisance but really, it's what has brought most of the current users in and continues to be the draw.
Taking the approach of transparency when it's used and then let people choose to upvote or not is a more sane approach. Medium's policy on it is failure to disclose the use of AI could get an account sanctioned.
As for the AI detection tools, their accuracy is spotty. And really, some smart cookie could use an accurate tool and have the AI rewrite until the tool doesn't detect itself. Sounds crazy but not impossible. The AI continues to learn as its used.
I think we're better in the long run to encourage appropriate use and then come down on those who are found to abuse it.
I consulted with an English professor here at the University a few days ago, as to how students should cite ChatGPT.
Her suggestion about using ChatGPT was to cite it as personal communication with ChatGPT, all the while realizing ChatGPT is a known liar (who spouts lies and inaccuracies with an incorrigible air of confidence).
I’d say that sums it up pretty well.
Thank you for this clear post @shadowspub I am very interested and excited to use AI as a pair programmer, and an always available Jeeves to my bumbling Bertie Wooster. But the entire discussion around plagiarism had been getting me very worried. Personally I find that using it creatively is an interesting challenge, and trying to cut off any entire section of people from using it felt like not allowing people to use the first cars or the first printing press. However in my thoughts my worry was never about crediting as whenever I work with AI I clearly inform that it is the case. Just like you quoted the ChatGPT output I thought it was natural to say "Hey I got this information from my man Jeeves here". Having read this post which clarifies so many of my questions, and hivewatchers post about AI I am now relieved that the main thrust of plagiarism was about uncredited work. I find that alright for now, though it will be interesting in the future if we will continue to put the same amount of importance to non machined original work. Probably it will be a small scale industry like the many old trades are currently. Cheerio!