RLP 311: Giving Credit to AI Tools in Genealogy Writing

Nicole (1s):
This is Research Like a Pro episode 311 Giving Credit to AI Tools in Genealogy Writing. Welcome to Research Like a Pro a Genealogy Podcast about taking your research to the next level, hosted by Nicole Dyer and Diana Elder accredited genealogy professional. Diana and Nicole are the mother-daughter team at FamilyLocket.com and the authors of Research Like a Pro A Genealogist Guide. With Robin Wirthlin they also co-authored the companion volume, Research Like a Pro with DNA. Join Diana and Nicole as they discuss how to stay organized, make progress in their research and solve difficult cases.

Nicole (40s):
Let’s go. Today’s episode is sponsored by Newspapers.com. Hi everyone. Welcome to Research Like a Pro. Hi Mom.

Diana (50s):
Hi Nicole. How are you doing?

Nicole (52s):
I’m excellent. I’m just really enjoying our discussion on AI tools. This has been so fun.

Diana (59s):
Oh, I agree. I’m loving it too. So fun. I’m excited to continue. But first, let me just tell you, I’ve been reading the latest National Genealogical Society quarterly. Ooh. This one is volume one 12, number one from March, 2024. So I was really excited when I saw it because it has a new DNA article titled DNA Merges Families of Steven Stillwell of Dutchess County, New York, Cornwall, Upper Canada, and I don’t know how to say this, Coshocton County Ohio. Wow. Have never come across that one before.

Nicole (1m 33s):
Yeah, that’s a lot of families all over.

Diana (1m 36s):
Yeah. So it’s really fun. I’m not all the way through it, but basically there is a man who left children with five different women and

Nicole (1m 44s):
Oh my.

Diana (1m 46s):
And it is through DNA that they’re providing the, you know the final proof in situations like that. So it’s really fun to look through it and I’m reading all the documentary part first, which that’s what the article leads with, and then it has the DNA, and so there’s a figure for each one of the different women and their descendants and the DNA test takers. So wow. You know. It’s really fun to read other people’s writing about DNA So that we can learn ourselves and just see how they put things together.

Nicole (2m 17s):
Yes. That’s how I learned how to do my DNA writing was just reading a lot of it.

Diana (2m 23s):
So this one has YDNA evidence and it has autosomal DNA evidence and it has some triangulation on chromosomes. Oh, good. So it’s really got a lot of DNA information. So this one is set up with all the documentary first and then all the DNA So I always think it’s fun to see how different projects warrant different ways of organizing DNA and documentary evidence.

Nicole (2m 46s):
Yeah. Sometimes the DNA is the first thing that starts off with DNA and that guides the documentary research and then other times it’s woven all throughout and then quite often is the method you just described where it starts with documentary and then presents a hypothesis that it tries to then confirm with DNA evidence.

Diana (3m 5s):
Right. And I don’t think I even mentioned the authors yet, so this is by Patty Hobbs, who’s a certified genealogist and Barbara Garrison who is a PhD.

Nicole (3m 15s):
Oh, neat.

Diana (3m 16s):
So yeah, it’s always so neat to read these articles and see what kind of things people are working on because it’s just amazing the different kinds of cases out there. Oh my goodness. A man with children. I mean we, we see that before, but we don’t have that in our own family history. So I haven’t really worked on something like that before

Nicole (3m 34s):
That we know of.

Diana (3m 38s):
Yeah. Actually not, not, not five different women.

Nicole (3m 43s):
There was Thomas Bradley in England who had illegitimate children with several women, including our ancestor, but never married any of them.

Diana (3m 53s):
Oh, okay.

Nicole (3m 54s):
That one was fun to discover. But the records, I mean, it was a little tricky. I can only hypothesize about some of them because there aren’t super good records of all of the children being his son or his daughter, but there were like some bastardy bonds in the parish and things like that.

Diana (4m 11s):
Well, that would be fun maybe someday to track down DNA matches through him and those different women. Interesting.

Nicole (4m 19s):
Yeah, at the time I looked and there weren’t too many matches for that line, and part of it is that those people stayed in England and didn’t have very many descendants or haven’t tested. Right. Well, for today’s announcements, if you are working in Airtable and struggling to feel like you’ve mastered it, I encourage you to check out our two PDF download Airtable Guides that we have for sale on our website. They’re $10, and the titles are Airtable Research Logs for Genealogy Quick Reference, and that’s in its second edition. So we updated it just this year. The other guide is specifically focused on tracking DNA matches with Airtable, and that’s also a quick reference. So they’re short, they’re concise, they’re just only four pages So that you don’t get bogged down, but they provide a whole bunch of helpful tips.

Nicole (5m 4s):
Also, our upcoming Research Like a Pro webinar for July is Who is Grace Brown’s Mother, presented by Mark Thompson, who’s a graduate of Research Like a Pro with DNA, and also has been a peer group leader in our recent DNA study group. And he has been speaking about artificial intelligence. So he’s wonderful and has a lot of great things to say on his blog making FamilyHistory.com. So check it out. Grace Brown’s mother is a lecture about confirming Anne Hayes as the biological mother of Grace Brown. It’s a three generation study tracing a matrilineal line back to the great-great-grandmother, and Grace was born in England and then died in Massachusetts So.

Nicole (5m 46s):
that helps you kind of see what it will be about. And if you haven’t registered for this year’s webinar series, we encourage you to do so and you’ll have access to watch all the recordings for each month, even if you sign up after it’s already started. The next Research Like a Pro Study Group is starting in August of this year. Registration is already open and there may still be additional spots if you’re interested. So be sure to check our website to see if you can sign up. And we have coming up at the end of July, our artificial intelligence workshop, Research Like a Pro with AI to learn how to incorporate AI into all the steps of Research Like a Pro to help you become more efficient and accomplish more with your limited time without losing your personal style and voice and expertise and without introducing inaccuracies into your work.

Nicole (6m 36s):
So if you’re interested in AI, we hope that you’ll join us. And as always, if you would like to get our weekly newsletter, it comes out every Monday, you can join our newsletter email list, and often we include coupon codes for some of our different courses and publications. Well, we hope to see you at the Virtual Professional Management Conference by the Association of Professional Genealogists. We’re going to be attending and we’ll be virtually there. So we hope to see you there September 19th through the 21st. Well, today we’re picking up with our second episode about disclosing and citing AI tools in our genealogy writing.

Nicole (7m 16s):
That’ll be fun to talk about again today.

Diana (7m 18s):
Right. I’m loving learning about AI and it’s so fun to kind of be at the forefront of this. It’s just starting out and we’re all learning as we go,

Nicole (7m 27s):
Right? That’s what it feels like. It feels like I’m learning as I go, and I’ve already made a bunch of mistakes with using AI for different writing projects that I wouldn’t do anymore. But hey, we just have to recognize that and try to get better. Yep. Well, let’s talk about giving credit and citing AI tools and this idea of You know. How do we give it credit? It’s not a human, but it did help write some of our words. So experts generally agree that listing an AI chat bot as an author or contributor is not appropriate since an AI chat bot isn’t human and therefore cannot take responsibility for the writing. So that kind of comes into play with probably other fields where there could be legal implications for the writing or any other kind of implication where someone has to take responsibility for the writing that can also be applied here.

Nicole (8m 20s):
I think it’s good to think about the fact that anything we write, we are taking responsibility for that. And an AI chatbot cannot take moral or legal responsibility for things that it produces. So we are the human in the equation. And that’s a quote from Steve Little in the NGS course. He used to say that we’re the human in the equation. We’re the human using the AI tool, so we have to take responsibility for it, and we have to check it, and we have to make sure that the material it produces isn’t copyrighted and that kind of thing. Well, another consideration is that some of that training data that the generative text models are using came from others’ work that was published online. And often the model gives us ideas or text, but we need to cite the underlying works used as the model’s training data.

Nicole (9m 6s):
Well, how do we even know what it’s using? One way that I kind of choose to use the models is by providing my own facts and information and simply having it put it into words instead of generating new ideas and generating new text, and I don’t know where it came from. Some models can provide sources for where their information came from, and we can also mitigate this issue of copyright by providing the information and citations ourselves, like I said, and just asking the model to assist with putting it into coherent paragraphs. For example, Google Gemini, after it generates several paragraphs, you can click the check with Google button and then it can search for any sites, anything in a Google search that has that information, and you can then cite it, you can check to see if it’s copied at word for word or whatever.

Nicole (9m 53s):
And so that can assist us with finding the underlying work that was used for training data. So it’s just a matter of understanding that the large language models or generative text models were trained on others’ work published online.

Diana (10m 9s):
So important to know where this information is coming from. Well, many editors of scholarly journals have come together and created a statement on the responsible use of generative AI technologies in scholarly journal publishing. And they suggest that authors who use generative AI to summarize literature, form ideas, make outlines, write drafts, and revise texts, should disclose this to their editors, reviewers, and readers. Yet, because the technology is so new, they have decided not to recommend rules for how to disclose it so they know there’s a need to to disclose it. But what are the rules? We all want rules, right?

Nicole (10m 50s):
Yeah. It would be so great to just say, okay, you just list it in the introduction paragraph, or you have to list it as a contributor. But they’ve all decided that they don’t want AI chatbots to be listed as an author or contributor. So there has to be a a different way now. So in that statement, it’s called the editor’s statement on the responsible use of generative AI technologies and scholarly journal publishing. So in that, they said, since generative AI is constantly changing and the scholarly community is only beginning to experiment with it, it is not prudent at this time to promulgate hard and fast rules for how generative AI should be disclosed.

Nicole (11m 31s):
We recommend, however, that disclosure should describe how the AI was used and should identify AI generated content. Authors should err on the side of too much transparency rather than too little. When in doubt disclose some ways of disclosing the use of generative AI could include describing the use in a paper’s introduction, methods section, appendix or supplemental material, or citing the generative AI tool in the notes or references. So I thought that was such a good idea that although we’re not sure the best way to do this yet because it is so new, here are some things we do agree upon.

Nicole (12m 13s):
And these editors are saying that they think we should definitely err on the side of too much transparency rather than too little. And there’s a lot of different places we can put it. The methods, the introduction, the notes as Genealogists. We don’t usually use methods sections, but we do have limitations sections in our research reports and in the notes and references. A really good place to disclose that.

Diana (12m 38s):
I like all those ideas. Well, there’s some additional ideas for disclosing use of AI tools and published papers that comes from the Computational Linguistics Conference. In 2023, the Association of Computation Linguistics Conference. The program chairs created a Policy on AI, Writing Assistance, and they defined six different ways AI can be used in writing assistance and proposed methods for citing the AI if necessary.

Nicole (13m 7s):
Yeah, let’s go through each of these different ones and talk about them. So the first one is language assistance. And this is one where most people don’t think that AI tools need to be disclosed. It’s like Grammarly or spellcheckers dictionaries, AI tools that are just doing simple grammar checking that doesn’t need to be disclosed. So, I thought that was helpful.

Diana (13m 30s):
Yes. And these are tools we’ve been using for years, right? I mean, I remember how excited we all were when Spellcheckers came out and the online dictionaries. I mean, there’s just so many tools we’ve been using that we probably haven’t really even thought of them as ai. I know I love my Grammarly helping me catch all those little mistakes that I make in writing,

Nicole (13m 52s):
Right? And so let’s say we write something and we upload it to Claude and we say, list any grammar problems that need to be fixed, then it would give us a list of things we might wanna fix for our grammar or spelling, and then we could fix that and that wouldn’t need to be disclosed. Okay? The next one is short form input assistance. So what this really means is predictive text tools. So if you’ve ever been writing an email in Gmail or using the Smart compose in Google Docs where you’re writing something, and then it can predict what you’re gonna say next because so many people write that thing next, or just because of the way that large language models work, they work by predicting the next token or the next string of text.

Nicole (14m 38s):
And so it’s really good at that type of thing. Using this kind of short form input assistance does not need to be disclosed according to this Association of Computational Linguistics conference.

Diana (14m 51s):
I think think that’s another one that we use all the time, even in our texting. It’s just amazing how it just helps us write our emails and text so quickly.

Nicole (15m 1s):
Right. Well, the next one is a literature search, and I think this is more common in other research fields, but we still do it in genealogy where we’ll look up and see what articles and what abstracts are out there on our topic. But it says that the use of generative text models to help identify articles and literature to review is like using a search engine. If we’re using those models to help us do that, we should definitely read the suggested literature and still cite them appropriately. And we need to disclose that we have used the model to help us with that. It’s kind of like it’s doing research for you. So for their specific conference, they suggest that you need to disclose that you’ve used it.

Diana (15m 43s):
Hmm, that’s really interesting.

Nicole (15m 45s):
And so these are just interesting ideas that we might consider incorporating into genealogy. Well, let’s do our next one. These categories are starting at the ones that aren’t generating a lot of text for us, but are kind of just more assisting us and moving up higher into more uses of AI to generate text. So this next category is low novelty text use of generative text models to produce descriptions of widely known concepts. Well, if we use it for that, then we should consider checking the work of the model. And if the model copies verbatim from an existing article, we need to acknowledge that with this citation and use of the model for this should be disclosed.

Nicole (16m 26s):
So we’re now You know the first two grammar and language assistance and short form assistance didn’t need to be disclosed, but starting with the literature search and the low novelty text, now we’re into the level of we should always disclose it for this.

Diana (16m 38s):
So if I were to ask it to write me a definition of autosomal DNA and it would go out and get, you know an expert’s writing perhaps on a blog post and then rewrite that for me, then I would want to disclose both the original author and the AI that rewrote it for me.

Nicole (17m 0s):
Right.

Diana (17m 1s):
Interesting.

Nicole (17m 1s):
Yeah, and that kind of puts the onus on us to find out where did this come from? And it usually, in my experience, it’s taking multiple articles and definitions and kind of putting them all together. So one thing you can do is you can cite the prompt that you gave it. And then I really like the idea of using Google Gemini to search and see if there’s any of that exact wording in the response it gave you in any articles that are online and you can kind of see and compare.

Diana (17m 31s):
Yeah.

Nicole (17m 31s):
Okay. Well, the next category with even more AI involvement is novel ideas. If the generative text model produces text that gives the author new ideas that should be acknowledged, authors may also want to check for publications that contain those new ideas in order to cite them. Use of the model should be disclosed. So I thought this was interesting because in other research fields, they’re often coming up with like new ideas and new ways of doing things, which I don’t know that that really applies in genealogy because we’re just looking for historical facts and trying to write about them. So it’s not as much as like, oh, she stole my idea as much as like wanting to cite others’ words that we’ve used, which goes back to the low novelty text category.

Diana (18m 21s):
Right, right. Yeah. I don’t think we’re making up things about our ancestors or records or

Nicole (18m 29s):
I got a new idea here.

Diana (18m 32s):
That’s funny.

Nicole (18m 33s):
Well, the next category is new ideas plus new text. So this is the use of generative text models to contribute ideas and write about them. And this requires fact checking and finding and adding citations. And the Association of Computational Linguistics discourages this use for their conference papers, and they suggest that authors who do use models for this purpose should disclose its use and let the peer reviewers know that they have done their due diligence to cite any other works that the model may have been trained on. So yeah, that’s an interesting one. And then their last one in the 2024 Call for Papers, the ACL, which stands for Association of Computational Linguistics, they added a seventh usage that needs to be disclosed, and that’s assistance with code writing, which you may have heard that

Diana (19m 28s):
ChatGPT can help you write Python scripts and things like that. So if you’re doing that, you should disclose that as well. Well thank you for taking us through all of those different ideas. Certainly gives us some food for thought, doesn’t it? Well, let’s have a word from our sponsor, Newspapers.com. This episode is brought to you by Newspapers.com. Discover your mom in the paper when you search Newspapers.com. Find the stories that made her who she is today by searching the largest online newspaper. Archive Newspapers.com makes it easy to find your mom, grandma, and others in the papers. Search more than 945 million pages from major news titles to small town papers, uncover birth and marriage announcements, obituaries photos, and much, much more.

Diana (20m 12s):
Explore papers dating back to 1690 or as recent as last month. Newspapers.com is used by millions of people every month for genealogy, historical research and more. Find something you like with Newspapers.com. It’s says, snap to save and share articles with family and friends, or attach them directly to your ancestry tree. Use promo code FamilyLocket for a 20% discount on your subscription. That’s code FamilyLocket, and celebrate the stories of a lifetime with Newspapers.com.

Nicole (20m 40s):
Well, I just have to say how much I do love Newspapers.com and I’ve been using it so much lately. The feature to create a clipping is so wonderful. I love that you can save those and you can give them a title. And then I use the online repository assistant tool to automatically generate a citation in my preferred format using the template I set up. So that every time I have a clipping and give it a the article, like a title, ’cause the article titles aren’t always indexed, then my citation can be automatically made. It’s so great.

Diana (21m 14s):
That is so great. Well, speaking of citations, what about this idea of creating source citations for AI chatbots? How do we cite them? The Chicago Manual of Style recommends citing content generated by AI by including the model name, the prompter topic, the company, and date. Others recommend adding the name of the person using the model, since the tool can be customized to the user’s preferences as well as which version was used since ChatGPT could be either 3.5 or four, depending on whether or not the user is paying. Some style guides suggest different styles for reference note citations, depending on whether or not the prompt was mentioned in the text.

Diana (21m 56s):
And Elizabeth Shown Mills suggests adding another layer to our citations describing any source we provided to the language model using the citation format for that type of source.

Nicole (22m 7s):
Right. Wow. So there’s a lot of different elements that we can include here, and I think it is key to use the model number or the version like if you’re using ChatGPT 3.5 or four, and then your name, you’re the user and it should be in the citation that you’re the one prompting the AI tool to create that text.

Diana (22m 29s):
Hmm. That’s something I really hadn’t thought about, but I’m thinking about the five questions that we always talk about with citations. you know, who created it? Would the who be you as the person? You’re the one giving it the prompt?

Nicole (22m 41s):
Yeah, I think both. It’s you prompting the tool. That’s why I think both need to be included.

Diana (22m 48s):
Right. That makes sense.

Nicole (22m 50s):
Alright, well we’ve talked about lots of different categories and uses of AI and disclosing them now, now that we’ve decided, okay, we’re gonna go ahead and disclose that we’ve used ai, how do we do it? Well, that was a great description of how to make source citations. How can we then include that citation? So let’s think about a situation where we’re drafting some text and maybe we’re using the generative text model to help help us get started and pre-write, maybe do some drafting. And the initial paragraph that’s generated by the chat bot is usually not exactly what we want, so we’ll probably refine it quite a bit. And so the process of using it could look like this.

Nicole (23m 32s):
You provide a prompt with your own ideas of what the paragraph should say. The AI tool drafts the paragraph, then you refine the paragraph through additional prompts, and then it gives you the final paragraph. You copy and paste that into Word, and then you edit the paragraph for final use in a written product by removing any words you don’t like changing a couple things and act. So in this example, the generative text model was a tool that you used to help you write low novelty text. So, that was category number four that we talked about and the model was not coming up with new ideas, it was simply taking your idea and putting it into words.

Nicole (24m 13s):
And this is similar to the ability of like a family tree software program that is capable of generating narrative reports from the facts we provide. But it does like a little bit better than those templates. The AI generative text model is a tool to help us take our data, our structured data from like a spreadsheet or just data we’ve copied and pasted and create a readable narrative output. So unlike family tree programs where the output usually looks and sounds like a report generated from a family tree program, the generated text by large language models can sound very human-like, and often can avoid detection. So because of that, we need to disclose that we’ve used that tool.

Nicole (24m 59s):
And I kind of feel like it’s pretty important to do this because I don’t really want to live in a world where everything is written by an AI chatbot, and I don’t even know if what I’m reading is written by that or by a human. I would like to know, you know what I mean?

Diana (25m 13s):
I absolutely agree. And especially as the chatbots get better and better or people get better with their prompts, You know if it sounds like a human, I really would like it to know that it’s a human.

Nicole (25m 26s):
Yeah. So to follow the suggestions from ACL, this usage of generating low novelty text should be disclosed. And to follow the recommendations from this statement on responsible use of generative AI technologies in scholarly journal publishing, we can go ahead and disclose this use in either the introduction of the work, the appendix, the supplemental material, or the reference notes. And those are all locations that we would use as Genealogists. Usually our works will have introductions, often they have appendices and that kind of thing.

Diana (25m 60s):
I like all those ideas, but I’m thinking about when I’m writing a research report for myself, I really like the idea of putting that in the limitations paragraph and disclosing that I use an AI tool to draft texts. So this would help me remember that I use that for one thing while I’m reading it, and then it would clue other people in as well that there could be some errors due to some of the text being written by a machine and not by a human. And as the human using the AI tools, it really is my responsibility to fact check the information. So when I first started, I thought it was really fun to have the chatbot generate writing, and sometimes it sounded better than my own.

Diana (26m 41s):
But lately I’ve just realized I prefer to write myself just like this report I’ve been writing. I just like putting it into my own words. I like having it be the way I think, and I like to do my own writing with my research reports. So even though the chat bot can have flowery language and have all these fancy ways of saying things, I think that with genealogy writing the more simple and clear the better. And from my prompts, I’ve read some writing that was just hard to even understand what it’s trying to say. It’s just using a lot of big words. So it’s, you know, I obviously need to practice more, but with my genealogy writing, I have decided to pretty much for most of my narrative, use my own unique style.

Nicole (27m 29s):
Right.

Diana (27m 30s):
But I do like the idea of putting the disclosure in if I am going to use it for some portions of the report.

Nicole (27m 35s):
Right. And the limitations section, such a great place for that with the research report because we already have a limitation section where we’re talking about I was limited to 20 hours or this timeframe from February to May, or I wasn’t able to visit this onsite repository that had relevant records and machine helped write some of the information

Diana (27m 58s):
Right, right. That’s perfect.

Nicole (28m 1s):
Well, I would only feel comfortable using AI tools for client reports at this point for simple uses like summaries of long transcriptions that I made, or that I’ve worked on with Transkribus, and then generating a results summary after I’ve already written the report myself. So I would cite my use of the tool for this in a reference note citation. You know you have maybe a introduction of a probate file and then a paragraph that summarizes it, and then a footnote that says that this was a summary written by Claude version Opus at Anthropic and the transcription was created by Nicole Dyer from this probate file.

Nicole (28m 44s):
So you have like three layers of the summary, and then the person who made the transcription and then the original source of the probate. So you could do it all in one reference note. And this is just a great way to incorporate some time saving ways of using AI into our client reports and then appropriately disclose and cite it. And I think the client would really appreciate that because instead of having to read, you know a four page transcription, which they still will have, you know, we still give that to them, but it will help the research report flow and then that can be included in an appendix or an attachment.

Diana (29m 25s):
Well, I am thinking about some AI that I used in my research report, and I had found a deed on FamilySearch’s new full text search. And FamilySearch is using AI to basically transcribe that document for you. So I had taken that transcription, that FamilySearch gave me, put it into a chat bot. I used Chat GPT and asked it to summarize it, and it did a really nice job of summarizing that for me. So now I can go and make a citation that I use that to create a summary.

Nicole (29m 58s):
I think that’s perfect. Yeah. So going back to client reports, we would want to put in the Limitations for that that AI was used and just spell out exactly what it was used for. You know it was used for helping summarize this transcription and for creating the results summary or the executive summary or whatever the case may be.

Diana (30m 17s):
It can save a lot of time. And if you are kind of gun shy of trying to create your own abstracts or your own summaries of something like a deed, which has a lot of boilerplate language, that is such a good way to use AI because it can take that out and just leave the real nuts and bolts of the actual document for you. But then you would of course always want to fact check it and make sure that you haven’t lost any meaning in that summary.

Nicole (30m 47s):
Yeah. Or you have to watch out for inferences, and I’ve been playing with this quite a bit, and I use it a lot in my own research and summarizing transcription for my research log. I like to have a little idea of what was in that source in the research log, but I don’t always want my full transcription in the log. Sometimes it will add in a little sneaky thing here and there, so you have to walk out for those inferences or those additions. So fact checking it is really important.

Diana (31m 15s):
And this is another example where if you’re just a newbie genealogist, maybe you need to do the transcription and summary yourself because you are learning as you do that. Some of these tools I think can really speed it up if you are more advanced and you have done a hundred deeds. You know what a deed is. But when I think back to when I just started working with deeds, I really needed to do the work myself to learn better. So just something to think about.

Nicole (31m 40s):
That’s such a good tie back to our episode before this where we talked about, you know when students are learning how to write, that’s not a good time for them to use AI tools because they have to actually do it to learn it. Same here that doing the transcriptions ourselves and the abstracting ourselves is how we learn. And then yeah, once we’ve done it a hundred times, like yeah, we can really speed up the process by having the abstract and stuff be done by the AI tool. We can fact check it and make sure it looks good and then move on.

Diana (32m 9s):
Right.

Nicole (32m 10s):
For the NGS workshop that I taught about Using Research Logs and Timelines to Help You Generate Research Reports with AI Tools, I used one of my previous research projects on Baldy Dyer. I had created a timeline for him and a research log, but then I never wrote the report. I just must have gotten busy. And it was a few years ago. And so in the meantime, I had received a record that I ordered from National Archives, a Bounty Land Warrant application. He died in the war of 1812, and so I had added that to my log. And then I decided to go ahead and try using Claude by Anthropic to help write the research report. So I went ahead and uploaded a CSV file of my starting point timeline and asked it to generate sentences and a paragraph for the beginning of the report about what was known at the beginning of that research session.

Nicole (32m 58s):
And it did a great job with that. And then I uploaded a CSV file of my research log and guided the chatbot through creating sections of my report row by row of the log with footnotes from my citations column. And I asked Claude to use my own transcriptions and comments from the log. And then Claude helped smooth out some of the text and grammar and added topic sentences and things like that. One thing I noticed is that several rows in my research log were for negative results. And so Claude didn’t really know what to do with that, and so I had to redo that section. But I guided Claude through creating a bulleted list of the negative searches, which is how I decided would be the best way to present it.

Nicole (33m 40s):
And my log actually didn’t even have complete source citations for some of these negative searches where I had just browsed the images in FamilySearch for different collections like the wills of Davidson County, Tennessee and that kind of thing. And so I just put the collection title for that microfilm and then what I noticed is that when Claude made the paragraph with the footnote citation, it created full consistent citations following the format of my other ones. It was amazing. It had taken the date where I did the search and added that in and added the word accessed and like just put in the root URL, like FamilySearch.org. And so it did a good job.

Nicole (34m 22s):
I was very impressed.

Diana (34m 23s):
Wow, that is amazing that it was smart enough to fix your citations up. Yeah.

Nicole (34m 28s):
That’s so cool. They are these, these models are really smart. It’s exciting to see what the potential is there. Well, the way that Claude and ChatGPT create footnotes is with a markdown language. And so you’ll see like the asterisk and carrots and brackets of things for different types of formatting and markdown and so I used Writage from Writage.com, and this is a plugin for pasting markdown into Microsoft Word. So I used the free trial of that, and I actually ended up purchasing it because I loved it so much. But it can copy the text and the footnotes generated by Claude and ChatGPT, and it can paste it into Word and preserve the formatting so that the footnotes were just super automatic.

Nicole (35m 14s):
I didn’t have to do anything with that. It just took the footnotes from my log and just eliminated that whole step of you having to copy and paste it into your report. And so after getting all the paragraphs generated, I did one paragraph at a time row by row from my research log. I then added my objective to the top of the report and saved it all as a PDF. Then I uploaded that PDF to Claude and said, can you help me write a conclusion by summarizing the findings section and list any progress made on the objective to trace Baldy’s children forward? And it did a great job with that. And I just kind of tweaked it a little bit.

Nicole (35m 54s):
And then usually in our client reports and things, we will add a results summary section or an executive summary at the top of the report to kind of give an idea of what was found. So I asked Claude to read the PDF and make a bulleted list with action verbs about what was found in the findings section So. that was also very quick and super helpful. And then my report was done, and I haven’t finished fact checking every sentence in the report, but it was really helpful to see how keeping a detailed research log with complete citations can then be used to generate a report with an AI tool. And in the limitation section of the report, I did mention that I used Claude to write the report So that since this report was for myself, then I can remember that I did that.

Diana (36m 44s):
That’s great. And I’m just looking at your report. So for our listeners, if you go to the blog post, there is a link there to Nicole’s report that she did with AI generative text. So my question is, how did you do your future research suggestions?

Nicole (36m 60s):
Well, in my research log, I have a column for to do next, and So I just tried to put the ideas of what to do next in the research log. So I kind of updated the research log a little bit when I was getting it ready to put into ai. And I added a couple steps in the log for each thing. And then when it generated the paragraph, so my instructions and my prompt to Claude was to start with the header, then do a block quote, and then like a summary of that, and then list fan club members, and then list ideas for future research. So each section has any of those ideas for future research. And then when I uploaded the PDF to Claude at the end, I said, make a list of future research suggestions by finding all of the future research suggestions that I put into the findings section, and then you know, make a bulleted list at the end.

Diana (37m 51s):
That’s great. I can tell you had too much fun creating this. This is awesome.

Nicole (37m 56s):
Yeah, it was pretty fun. It was so fun.

Diana (37m 60s):
Well, let’s talk a little bit about various ways to disclose the use of AI tools. And I think Nicole, you’ve decided after all your study that it’s almost always a good idea to disclose this when you’re using it to generate text for many uses. You can just do a note at the end of the article that you use ChatGPT to help draft some of the texts. But here’s some other ideas. If you are doing video and podcast descriptions, you could do an acknowledgement at the end of the summary. If you’re writing a blog post, it could put that acknowledgement in reference note citations or an acknowledgement at the end of the article. If you’re writing it to yourself a report, like a research report, you could put it in the reference notes and the limitations section.

Diana (38m 46s):
If you are using this for a client report, reference notes for summaries of long transcriptions, a reference note for generating a bulleted list of results or reference notes for explanations. So lots of different places you could put it in a client report

Nicole (39m 1s):
And the limitation section, I should update that

Diana (39m 5s):
And the limitation section. And then in a syllabus, you could put a reference note and a note in the footnote section of the first page. If you’re writing a journal article to be published, you would have those in your reference notes and a note in the footnotes section of the first page, if the journal allows AI writing assistance. And so something to think about before you start writing and using ai.

Nicole (39m 31s):
Right? Yeah. Well, hopefully this discussion gave everyone some ideas for when we need to disclose the use of AI and how to site it. And I really found those six categories of AI use and writing assistance from the Association of Computational Linguistics Conference to be especially helpful in determining when to disclose and just thinking about the different levels of usage. And sometimes we’re using it just for grammar checking, and other times it could be generating new ideas and new text. So it’s really important to consider what type of usage we’re doing and then treat it differently. So let us know how you’re using AI. We’d love to hear from you as we all navigate this world of AI tools and genealogy writing together.

Diana (40m 14s):
Absolutely. And I love how at the very end of the blog post you wrote, AI tools were not used to generate text for this blog post. Oh, that’s funny. I wonder if we’ll get to a place in the world where we actually have to state that, I wrote this myself, you know that we’re also used to using AI. Right?

Nicole (40m 33s):
I know. I saw somebody else do that, and I was like, oh, that’s interesting. Yeah. So I’m gonna do that too, because since I’ve been talking so much about using it in this article, I need to say, just so you know, I didn’t use it for this.

Diana (40m 44s):
I wrote it myself. Every word came outta my head. Well, Thank you so much for writing the article and doing the research and just giving us so much food for thought. I am excited to move forward and learn better and better how to use AI to speed up my research and my writing, but also to keep my unique voice there. You know, what’s the balance? I think that’s what we’ll all be working on in the future. So, thank you, thank you. And thanks everyone for listening. We hope you’ve enjoyed this discussion on ai and we will talk to you next time.

Nicole (41m 20s):
Alright, bye

Diana (41m 21s):
Bye-Bye

Nicole (41m 21s):
Thank you for listening. We hope that something you heard today will help you make progress in your research. If you want to learn more, purchase our books, Research Like a Pro and Research Like a Pro with DNA on Amazon.com and other booksellers. You can also register for our online courses or study groups of the same names. Learn more at FamilyLocket.com/services. To share your progress and ask questions, join our private Facebook group by sending us your book receipt or joining our courses to get updates in your email inbox each Monday, subscribe to our newsletter at FamilyLocket.com/newsletter. Please subscribe, rate and review our podcast. We read each review and are so thankful for them. We hope you’ll start now to Research Like a Pro.