Lots of people are asking questions about the hottest new ethics issue on the block since The Work of Art in the Age of Mechanical Reproduction and that issue is AI authorship of science papers.
We got turned on to this question via a recent post from Professor Lance Fortnow, who might be a reader of our blog (we are responsible ethicists around here and cannot reject the possibility without evidence!).
Lance wrote about OpenAI's recent math paper where they got ChatGPT to solve a math problem. OpenAI is, of course, a very fraught company from an ethics standpoint given how they are both a nonprofit and not-a-nonprofit and also because it's unclear how much their stuff is open, artificial, or intelligent. But the special thing about OpenAI's math paper is that they didn't name anybody as author, not even ChatGPT, they just said "by OpenAI" and left that for the rest of us to figure out. Which is exactly what we are going to do today as we delve into some of the "challenging authorship issues" Lance raises surrounding the grand entrance of AI into the academia hall of fame.
Now, here at the Hades Ethics Consultancy we're big believers in liberty, so we've taken the liberty of organizing the authorship issues into two distinct questions:
- Should OpenAI or ChatGPT get authorship on science papers?
- What is "authorship"?
We're going to start with #2—applying a style of reasoning known among ethics professionals as "deductive reasoning" which basically means thinking backwards.
What is "authorship"?
Here at our Ethics Consultancy we've sure worked with a lot of academics since our founding in 2023. And we love scientists and all their quirky ethics queries! But one tendency we've noticed with academic types is that they tend to use words with meanings that are actually kind of different from the meanings normal people use with them.
Ethicists call such terms "overloaded words."
One such overloaded word is "authorship" and one example of it being loaded quite overly appears in the Authorship and AI tools report produced by the Committee on Publication Ethics and quoted by Lance. (FYI, we would be honored to be invited to any other ethics or ethics-related committees going forwards!) Here's the quote:
AI tools cannot meet the requirements for authorship as they cannot take responsibility for the submitted work. As non-legal entities, they cannot assert the presence or absence of conflicts of interest nor manage copyright and license agreements.
If we are honest: That's a whole bunch of legalese! This COPE committee says it's about ethics but, when you read the find print, what they are actually worried about is "legal entities." And the same's what you uncover when you delve into the academic debate around "authorship," too: a room full of professors with their lawyer hats on.
You see, in academia, authorship doesn't mean the normal thing of "what person or entity authored it?" Our research shows instead a broad range of applications of the "authorship" designation, ranging from humans intimately entwined with the creative and scientific process, to people with little connection to the work getting "grandfathered in" for political reasons.
What this shows is that paper authorship is a legal, not an academic concept. Specifically, it's like a homework deadline or your salary: something academics negotiate and transact in order to achieve a mutually benefitial outcome. Specifically, it's a contract.
A contract is an agreement that specifies certain legally enforceable rights and obligations pertaining to two or more parties. (Wikipedia)
Since "authorship" is a contractual agreement, it means different things in different contexts, times and places. It depends what you put in the contract!
By contrast, ethics isn't about times and places. Ethics is the study of eternal ethical truths that don't change, at least not as easy as contracts do! So:
Should OpenAI or ChatGPT get authorship on science papers?
"Get authorship" is another one of those quirky phrases that our academic clients sure love to use. In the ethics community, we think of authorship as something one does rather than receives.
That's why we're calling on the Committee on Publication Ethics to rename themselves the Committee on Publication Law, COPL, because that's what they're really concerned with, protecting the contractual interest of academics in the current environment instead of elucidating the underlying ethical issues, of which by the way there is no shortage!
If we can take one more liberty, we'd like to try flipping this question on its head: Why shouldn't an AI maker like OpenAI or an AI like ChatGPT be eligible for authorship? From a strictly ethics perspective, it doesn't harm anyone to add an entry to the authorship line—as long as the addition is honest. And ChatGPT did author the thing, didn't it?
Unfortunately, we live in a world of lawyers, not ethicists. And the lawyers want to overload "authorship" to give the term a specific legal function that has to do with suing and not-suing instead of using it as a factual attestion of how a paper was generated.
Conclusion
Now let's do the ethical thing and anticipate a counterargument: Maybe there's another reason, besides the legal mumbo-jumbo, why ChatGPT can't count as the author of a science paper. For example, maybe there is a special meaning of the word "writing" that sampling from an LLM doesn't count under.
Or maybe, there are deep ethical problems with using AI tools like ChatGPT in academic writing that stand independently of whether or not ChatGPT is listed as an author. Hint: You bet there are.
We sure love a good argument and those are both best-in-class. But neither of these arguments has bearing on the question at hand, which is about the authorship line on a science paper and the ethics of listing an AI there.
Our conclusion? It's only ethical if it's unethical, for example, if an AI is given credit to make the paper look stylish when in fact the authors wrote it all themselves. Or if authors omit the AI author as part of a strategy of covering up their reliance on LLM tools!
Crowd: We're proud of this first ever Hades Ethics Consultancy Thought Experiments and grateful that you've joined us to experience this new post format and hopefully recurring series. We welcome comments and disagreements. Always. You could say we believe in the Socratic Method; it's our commitment to you, the reader.
Right here: hades@acaciavalleyhoa.org