As I defined in my earlier submit, I lately tasked AI with evaluating two transcripts of the 1807 treason trial of Aaron Burr. My final query is, what do I do with the doc that resulted? And that breaks down into two sub-parts. First, do I publish this, both simply on-line informally or with some sort of journal? And second, how do I describe what my relationship is to it? Am I a co-author? The writer? Only a prompter?
Let me begin by explaining how the memo was created, after which flip to the questions I’ve.
I. How the AI Memo Was Created
I would like to begin with what I did to assist create the memo, as that could be related to my questions. The transcripts that wanted to be in contrast had been .pdfs of two-volume books from 1807 and 1808 that go for a whole lot of pages, though the one components I cared about had been the components on the privilege towards self-incrimination. I used Claude (Opus 4.6 prolonged), and I tasked it with evaluating the discussions of the authorized arguments concerning the privilege towards self-incriminagtion to get a greater sense of whether or not my 2021 article on these arguments based mostly on the Robertson transcript was correct in gentle of the Carpenter transcript.
To say that “I tasked Claude” covers up lots of element, although. I went by means of round 30 rounds of prompting with Claude, over the course of some hours. As I went alongside, I discovered about what Claude may and could not do and pushed it to do a greater job when it was immune to do extra. For instance, once I first requested Claude to check the paperwork, it declined, saying it was simply too large a activity to take the 2 lengthy pdfs, to make them readable, after which to check them. So I began with a neater activity: Take my 2021 article, learn it, and perceive what it claims concerning the Burr trial, after which learn the Carpenter transcript and write an article presenting a comparability. The primary draft reply was a begin, and made me assume that the enterprise could be in the end helpful. But it surely left so much to be desired.
Over time, I got here to comprehend that there was an artwork to getting Claude to make the comparisons I wanted. Finally it agreed to do a direct comparability of the 2 transcripts based mostly on the claims I had made in my 2021 article. And I shortly realized that the comparability actually wanted a number of direct quotes and web page references for every little thing, in order that took lots of additional time: Amongst different issues, Claude had stunning problem with web page numbers, partly as a result of one of many two .pdfs had two volumes again to again and Claude couldn’t work out the pagination. I needed to spot examine, and I stored discovering errors, which Claude needed to hold correcting.
A giant breakthrough got here once I realized by means of trial and error that Claude may do comparability screenshots. That’s, as an alternative of simply telling me what the 2 transcripts stated, Claude may take screenshots of the related discussions and present them side-by-side. That approach I may very well be extra assured that I used to be getting an actual comparability. Even then, the screenshots wanted lots of correction: Claude began off giving me solely about 1/3 of the comparisons appropriately, and I needed to hold telling it to return and ensure it was screenshotting the precisely equal sections.
As I neared the tip, I additionally added extra duties. For instance, I requested Claude to learn the Carpenter transcript and inform me if my 2021 article had precisely summarized the arguments, in addition to whether or not there have been any components of the argument from the Robertson transcript I had missed. I additionally requested Claude to say if there was any authorized supply Robertson had reported than Carpenter hadn’t and vice versa, and any authorized argument that one had reported that the opposite habn’t.
After about 30 rounds of prompting, I ended up with a 22-page comparability memo. A typical web page of the memo seems to be like this:
The excellent news, at the very least accordiing to Claude, is that all the substantive factors matched actually properly. The variations between the Carpenter transcript are apparently very minor, the sorts of stuff you may anticipate with two unbiased human beings attempting to jot down down hours of courtroom proceedings and listening to some small issues in a different way. The components I cared about had been a match.
II. Ought to I Publish This, At Least in Some Means?
This brings me lastly to the large queston I’ve, what do I do with this 22-page AI-generated memo? On one hand, should you’re excited by my 2021 article, I believe the AI comparability is of scholarly curiosity. The comparability is fairly noteworthy, at the very least for the small variety of nerds who care concerning the substantive matter. However, the AI-generated memo would not slot into any conventional understandings I’ve of both scholarship or non-scholarship. So I do not know what to do subsequent.
My uncertainty breaks down into two questions. First, do I publish this? And second, if I publish it in some form of approach, what ought to I state as my relationship to it? Let me clarify my considering as to each questions.
First, on the query of whether or not to publish this, I’m conflicted.
A part of me thinks this memo must be just for my inner use. In some unspecified time in the future, I ought to simply do the work of studying the Robertson transcript once more and studying the Carpenter transcript for the primary time, and I ought to write up a comparability myself. Maybe the memo ought to simply be a useful information as to what I’d anticipate finding, a form of map that factors the way in which and makes the duty a bit faster. That is akin to how I have a tendency to make use of analysis assistants: Write me a background memo for me, which then offers me an useful thought of what I ought to search for once I do the work myself.
Alternatively, I may submit the AI-generated memo on SSRN.com, the place lots of draft articles are posted, however not attempt to publish it with any precise journal. As soon as it’s up at SSRN.com, anybody can discover it with a Google search. From a scholarly perspective, that’s the essential half; that approach, individuals within the matter can see the comparability. However there isn’t any want to really publish the essay in a journal. It isn’t even my work product, however Claude’s work product, so there’s not a lot level in getting it formally revealed.
Or perhaps I may (at the very least attempt to) publish it? It appears to me it is an academically fascinating doc. Perhaps there is a journal that will be excited by publishing it, and if that’s the case, why not? That will give the memo a extra everlasting house, and it’d draw extra consideration to my 2021 article and the problems I used to be attempting to deal with.
III. What’s My Relationship to the Memo? Am I The Creator? An Creator? Neither?
Lastly, there’s the set of questions on authorship. If I simply hold this as an inner memo, granted, I haven’t got to fret about that. But when I submit it on SSRN, or (actually) if I attempt to publish it, I do. Am I an writer? A co-author? A prompter? What am I?
One factor that appears clear to me is that I mustn’t publish it the memo as an article single-authored by me. Maybe I’ve an excessively romantic notion of authorship, however I really feel like authorship implies the second of sitting in entrance of a clean web page and placing my phrases on it. There must be an authenticity behind that, and prompting Claude to jot down one thing (even many occasions) would not really feel prefer it makes me the writer. Even when I checked it, I did not write it.
One other chance is that perhaps I’m a co-author. Perhaps my course of the mission, and my repeated prompting, made me a co-author together with Mr. Claude Opus, the precise author. That appears higher than saying I’m the writer, as at the very least I’m attempting to disclose how the memo got here to be. Though a co-authorship strategy is just a little bizarre: It isn’t like Claude and I are two students who labored on the article collectively. I do not even know if SSRN would enable me to state “Claude Opus” as a co-author. So I am undecided that matches.
A 3rd chance is that roles like mine are one thing new, and we have to give you a brand new vocabulary for it. Perhaps I did not writer the article, however quite I’m the prompter of the article. Perhaps I did not write the article, however quite directed it. Maybe, in my function as prompter/director, I shoould write an introduction that explains my targets and the way the AI-generated memo got here to be. Mainly, I ought to summarize what I’ve written in these weblog posts up to now. After which I connect the AI-generated memo, for which I take no authorship credit score. That approach, the reader is aware of who did what and the place the memo got here from, in addition to its limits. There is not a job of prompter-director now, however perhaps there must be?
Proper now, at the very least, my intuition is that I first have to assess how a lot time it could take to do that myself. If it will not take an excessive amount of time, and if I’ve the time, I ought to simply use the AI-generated memo for my very own inner use as a information for once I do the mission the old school approach. What sees the sunshine of day will probably be my very own human-reasoned and human-written article as an alternative. Alternatively, if I believe the time dedication is an excessive amount of given different obligations, I believe I am going to attempt to take the prompter-director function: I’ll write the intro and fix the memo, posting them collectively on SSRN, with the entrance web page saying “introduction and prompting by” me however the article clearly labeled as written by AI.
These are my instincts, at the very least. However I do not know. What are your ideas?
