Using AI: The Assur, the Muttar, and the Safek
I believe it is time to have a more nuanced conversation about the role of AI in our work as writers and producers of ideas in general, and as writers and producers of Torah in particular.
___
We can probably all agree on the extremes.
On the one hand, having AI generate an entire Dvar Torah on the parsha or a complete teshuvah on a particular topic, and then presenting the result as our own, is definitely wrong. It is unethical and immoral.
On the other hand, if we write a paper, Dvar Torah, or teshuvah and ask AI to review the syntax, check the grammar, or correct mistakes, that is entirely legitimate. AI is doing basic editing.
At these two extremes, it seems fairly obvious what is acceptable and what is not.
The question we need to grapple with is what happens in the broad gray area between these two poles. That is where things become complicated. Within that space, where exactly is the line separating legitimate from illegitimate uses of AI?
My personal experience as a writer complicates this question for me significantly.
I am an ESL writer. English is my second language, and it therefore took me many years of hard work to become what I want to believe is a pretty decent English writer. The same goes for Hebrew. Satmar, of course, did not train me to write in Modern Hebrew. Learning how to write well in Hebrew was therefore a long and arduous process.
Whatever ability I have developed as a writer, however, owes a great deal to the extraordinarily helpful editors with whom I spent many long hours over the years. Sometimes they merely corrected a typo or minor mistake. At other times, they did much more, making suggestions about syntax, formulation, sentence structure, or even the whole architecture of an essay.
I had many editors over the years, but the two most helpful were my wife, who is a genius editor, and a personal editor who worked with me for many years and is also a very close friend. They were able to be so helpful because they both knew me well, not only through the sum total of my writing, but as a person. They knew my ideas, beliefs, and sensitivities, as well as the way I think and the manner in which I like to express myself. They could therefore point out things that might not have been obvious to someone who had read only the particular piece in front of them.
They would sit with me and think through a topic and say, “Maybe this is not how you want to say it. Perhaps you should express it differently. This comes across as too strong, too weak, too blunt, or too hesitant,” and so on. Their suggestions emerged from a broader knowledge of who I am and what I was trying to say.
Good editing, then, entails more than knowing how to edit a piece of writing. It also requires knowing the person who wrote it well. The result was a significantly better piece, still written by me but immensely improved by extensive editorial corrections and changes.
This, in turn, raises the question: if this level of editing is legitimate when done by a human being, does it become illegitimate when done by a machine?
Perhaps not, since for those of us who use AI in many different parts of our lives, it can gradually acquire something of the broader knowledge that made my wife and editor so helpful: knowledge of the writer beyond the individual piece. It can bring to the editorial process not only its technical prowess, but also something of the whole gestalt, the larger image of the writer and the writer’s work. That can allow it to offer advice that is highly personalized, useful, and productive. So why not have it do the kind of editing my wife or editor would otherwise provide? I would still be the author and the arbiter of the suggested corrections, deciding which edits to incorporate and which to reject, with AI playing a robust editorial role.
__
For me, this is not an abstract question. Utilizing AI in that role would be a tremendous boon. It could save a family member’s time or the cost of hiring a professional editor. After close to forty years of teaching and writing, I am now preparing the Torah I have produced over those years for publication.
Yet even after writing for so long, I still prefer to have my work edited extensively because it makes the final product so much better. That need for editing arises for me in several concrete ways.
Beyond preparing older material for publication, that need also arises in my ongoing writing of Divrei Torah. Sometimes I write a new Dvar Torah, but at other times I repurpose one that I wrote many years ago. In those instances, I need to turn to an editor. The core text is good and worth sharing, but it still needs serious reworking, either because the content no longer works as well or because the audience is different. The original audience may have been less knowledgeable, while the new audience is more knowledgeable. Or perhaps the original audience was more aligned with my philosophy and therefore more readily accepted the thesis I was trying to convey, whereas the new audience requires more persuasion. Either way, the piece needs serious editorial work.
Suppose, then, that I sit with AI and use it to help me undertake that serious editorial work. The original Dvar Torah is mine, and I remain personally engaged in reworking it, paragraph by paragraph and sentence by sentence, considering each suggestion and deciding what to accept, reject, or revise. Would the final product not still be considered my work, with AI serving as an editor?
The same editorial question arises in relation to my Facebook posts.
As some of you may know, over the years I have written a great many Facebook posts. I am now thinking about turning some of them into a more formal body of work. The challenge is to take posts written at different times and in different contexts and organize them as parts of a larger whole. I could use AI’s help to identify recurring themes and think through the connections between the pieces, exactly the kind of work an editor would help me do.
So, would these be legitimate uses of AI?
Personally, I think the answer is affirmative: these are still acceptable uses of AI. The editorial process is exactly what it always used to be–but for the fact that it is performed by a machine.
But at the same time, these uses require extreme caution and a high degree of honesty and integrity.
The reason is that such robust editing can bring one very close to the moral precipice: the point at which editing turns into rewriting. Instead of merely suggesting a better way for the original to be written, the machine can begin to rewrite the piece, whether a sentence, a paragraph, or the entire work. The line separating the former from the latter can be very thin and difficult to see, making it all too easy and tempting to cross.
__
This is my considered view (I have thought about this long and hard), but I realize that others might disagree, perhaps vehemently.
Either way, not having an opinion at all is not an option, especially for those of us in the business of producing content, who are dedicated to doing the best work possible while adhering to the highest moral and ethical standards.
__
One Further Question
All writers and editors have to struggle with these questions, and we who write Divrei Torah are part of that larger conversation. But I also think there is a question specific to Divrei Torah. Are the rules for writing Divrei Torah different? If so, in which direction? Is there more leniency or more room for flexibility when writing Torah? Or is there less?
__________________________________________
(Full disclosure: I composed this entire piece myself and used AI in two ways. First, I dictated the initial draft, and AI transcribed it, saving me considerable time; dictating a piece takes much less time than writing it. Second, I used it to perform soft edits.)