When it comes to permissible uses of AI in professional writing, one philosopher argues that “anything that is acceptable to ask a human to do is, other things equal, acceptable for an LLM to do. (Equivalently, anything it would be wrong to ask a human to do is wrong to ask an LLM to do.)”
This “simple principle” was put forward by Kyle van Oosterum (Southampton) at his newsletter, Unchambered, several months ago.
The principle rules out, van Oosterum says, having an LLM write an article for you and you failing to acknowledge that. Writers are generally not morally permitted to “take credit for someone else’s work,” and that goes not just when the someone else is a someone but also when it is a something . But the principle doesn’t prohibit, he says, using an AI to “transcribe interviews, generate pushback on my… thesis, suggest trims when I’m over my word count, sharpen podcast interview questions, and perform a final fact check on columns and editorials.”
van Oosterum says:
The Simple Principle would, therefore, permit [a writer] to use an LLM to complete the very same tasks which are ‘ancillary’ to the main job. This point is crucial. The principle is compatible with the obviously-true idea that an author retains ultimate responsibility; just as one cannot blame a human assistant for a published error, the author remains the agent responsible for the final output.
Is van Oosterum’s principle a good one? The idea is to piggyback on the existing norms we have regarding authorship and assistance.
Keep in mind it was offered in something like blog post. Presumably, he would sharpen it up some were it to appear in a less casual venue. We can do a little of that work. Here is the original version again:
1. Anything that is acceptable to ask a human to do is, other things equal, acceptable for an LLM to do. (Equivalently, anything it would be wrong to ask a human to do is wrong to ask an LLM to do.)
To start, the language of “asking” is a distraction; the principle is supposed to be about the ethics of having AIs and other people complete tasks for you, not the ethics of asking them to do so. (Which is not to say that there might not be interesting ethical questions about asking.)
Now let’s turn to the ceteris paribus clause. I’m not quite sure what “other things equal” is supposed to refer to in the context of this principle. It can’t mean “provided no other possibly morally relevant issues besides the bare assigning of the task to a human or an LLM arise,” for what we’re interested in, when assessing the principle, is whether there are other morally relevant issues that tend to arise with this choice that it doesn’t adequately address. So I’m tempted to leave it out, and simply say that whatever results from our further refinement may be a defeasible principle.
Next, note that the parenthetical van Oosterum added to the principle (Equivalently…”) implies he meant it as a biconditional. Put in a form that makes that clear, and taking into account the other revisions, we get the following principle:
2. For professional writing, it is morally permissible to have an AI/LLM perform a writing-related task if and only if it is permissible to have a human perform that task.
What do you think?
A few preliminaries:
- The principle is being offered up for our consideration. That it may conflict with the extant policies of journals and other publications is beside the point.
- Humans and AIs have different capacities; each may be able to do things (or do things in certain ways, or for certain durations) that the other cannot. Does that matter for this principle? No. The principle is about permitted acts, not required ones. While ‘ought’ implies ‘can,’ ‘may’ does not.
- The principle is presumably intended for AIs/LLMs roughly as they are now. If their capacities significantly change, so too may the principle(s) that apply to it.
- That an act is morally permissible is not necessarily the only thing morally relevant thing to say about it.
Is this a good way to approach the ethics of authorship in the era of AI? Discussion welcome.
d-ai-ly nous