ChatGPT and other large language models have the potential to take information and produce concise summaries with proper punctuation, grammar, sentence structure, etc. For some researchers, this could be tempting to avoid the burden of a difficult task. But, there are considerations for how ChatGPT or other large language models can be used.
- Incorrect or biased information. Authors should be careful when using presumed facts stated by ChatGPT. ChatGPT can provide information that sounds authoritative, but is ultimately incorrect or biased, so human authors need to check ChatGPT's facts.
- Transparency. If an author uses ChatGPT in the course of producing their manuscript, they should disclose how ChatGPT was used in the cover letter and submitted work.
- Authorship. Most scholarly journals do not allow ChatGPT to be an author of an article because an author must be able to be accountable for all aspects of the work. ChatGPT cannot be held accountable since it is not human.
- Confidentiality. The contents of submitted manuscripts should be kept confidential until public release. If using ChatGPT to spell check, edit, proofread, etc, authors must keep in mind that confidentiality may not be assured when using these technologies. Conversations with ChatGPT's open model are being recorded and used to refine the model's training.
- Plagiarism. ChatGPT and other language models may regurgitate information that they were trained on without properly citing it. So, if you use ChatGPT's script in your manuscript, you may accidentally plagiarize someone else's work.
- Copyright. The issue of who owns the copyright on ChatGPT's responses could be complicated. US Copyright Law is predicated on humans owning copyright, so it seems clear that a large language model cannot hold a copyright on the material it produced. So, who holds the copyright on the responses ChatGPT produces? OpenAI has stated that it assigns the account holder all "its" rights, title and interest in the output generated based on the input. So, this sounds like the user holds the copyright. However, since ChatGPT learned on information which was presumably copyrighted by many other people, OpenAI might not hold the rights on the original data that ChatGPT may be using in its responses. This area of the law will evolve, so it's important to consider whether you own the copyright on content you take from ChatGPT's responses.
