Edit
This proposal has not been well received. I marck the ban as the official policy.
I would allow the use of AI assistance, but I would not allow accounts based only on AI systems by users with no Earth Sciences knowledges.
I would not completely ban AI use to generate content just because GPT-3.5 fails to solve code or mathematics complex problems (it even have access to a computer to test its code).
Allowing answers generated by the AI however (AI based accounts), would be nonsensical, as a student can simply ask the AI, obtaining the same result, and we would be just copying AI output to the Stack Exchange server, wasting time, energy, and resources. The answers would also quickly become deprecated, as AI systems improve each day. Additionally, they would be link-only answers, which are discouraged here. We would also have the collateral damage of having users with no Earth Science knowledge earning reputation on the site, making it difficult for them to moderate anything related to Earth Science. So we should officially not allow AI accounts and if a user is clearly hijacking the reputation system, giving no sources to his link-only answers, and not giving signs of having Earth Sciences knowledges, moderators should ban this user.
However, asking the AI Earth Science questions and copying its answers is not the only way to use the tool. Just like a journalist, a scientist publishing content can benefit from AI in many ways.
I found six ways to use the engine as an assistant to help you generate content on ESSE:
1. Generate questions.
This is banned on Stack Overflow too.
I asked the AI to generate a question for us and it was a bit broad, so I performed my input entry, obtaining:
I find the questions generated by GPT-3.5 acceptable, but I am concerned that this could potentially be used to manipulate the reputation system by asking a large number of consecutive questions. On the other hand, ESSE does not currently have a high volume of daily questions, so if this feature is allowed, it could help our site grow. I would use it to get ideas for the title of new questions.
You can ask the AI to completely create the text of an ESSE question too:
As for literal answers, I am against allowing a user to hijack the reputation system using the AI to post literal AI-generated questions. I would allow the AI to get ideas for a question, but some own work should be shown while posting a question, it would be unfair to let a user generate dozens of question by this system.
Problem is how are we going to detect it. You can ban a user posting 20 questions in a day that are not specially elaborated, but you can have another user more cautious posting only a question a week. Moderators will have hard to detect this second user, as the current tools are not very accurated (they give false positives and false negatives).
2. Obtain partial information for your answer.
For example, to the question: "Who are winners from the global warming? Are there any?", in my answer, I could have asked the AI, "What are the global economic cost of natural disasters?" and then quoted the AI's data.
Global economic costs of natural disasters:
According to the United Nations Office for Disaster Risk Reduction (UNDRR), between 2000 and 2019, there were 7,348 natural disasters
worldwide, resulting in 1.23 million deaths and $2.97 trillion in
economic losses. This represents an increase in both the number of
natural disasters and the economic costs of these disasters compared
to previous decades.
Source: GPT-3.5
...continue with my own answer
For me there is not a big difference between quoting GPT-4 and quoting Wikipedia.
My wonder is if free GPT-3.5 would be considered a legitimate source, as it passes a history high school exam by the skin of one's teeth, but that's my problem as a publisher. But GPT4 does much more better than GPT-3.5 and its cost is 20$/month.
I asked the engine that, and apparently, it is okay to use its texts in ESSE (for ChatGPT). It says you should quote its text as "ChatGPT, an AI language model developed by OpenAI."
3. Find papers
4. Do SEDE queries
Can GPT-4 be used to do SEDE queries?
5. Detect mistakes on answers.
I asked ChatGPT if an obviously wrong answer is correct and the AI detects it is not:
I wonder here if it would be appropriate to use it to detect false information and penalize the content downvoting.
I think the site could benefict of having his own AI system incorporated to the system to detect wrong content if this technology improves its results.
6. Correct texts.
This is where the AI becomes a helpful tool for people like me who don't speak English very well. The AI can correct my texts and I have started to use it before posting. This brings me to the question of whether I should acknowledge the AI in every text I publish that has been corrected. The AI says "if you wish".
So, I think I am going to indicate in my profile that my texts are corrected by GPT-4, but I am not going to include a quote from it in the footer of my texts.
