In our Research Like a Pro with AI workshop last week, I taught a lesson about using AI to help with writing. Part of that lecture focused on writing proof arguments, and the example I used was a DNA proof argument. I’m excited to share a discovery I made about using Claude.ai to generate descendancy diagrams of DNA matches! These diagrams are an important part of any proof argument incorporating DNA evidence.
In the process of writing this post, I also discovered that Lucidchart has integrated generative AI from OpenAI into their diagramming tool. See the end for a quick tutorial on how it works.
Sally Keaton Example
For an example in my lecture, I used a research report about Sally (Keaton) Reeves. Sally was the sister of my ancestor, Lucindrilla (Keaton) Welch. Sally was married to William Reeves, according to Sally’s father’s estate file in Anderson County, South Carolina, in 1830.
One of the documents states that Sally and William Reeves resided outside of the state of South Carolina during the estate division. One additional clue to their residence was found toward the estate file – a list of heirs living outside the vicinity [of South Carolina]. Next to the Reeves family’s name, it states “Tennessee W.V.” I used that as a reason to search for the Reeves family in Tennessee; but didn’t know what W.V. meant (possibly Western Virginia).
Family trees online didn’t state any children for Sally Keaton and William Reeves, nor their residence after 1830. Very little was known at the start of the project, aside from what I had in the transcription of 56 images from William Keaton’s estate file. My research objective was to identify the children of Sally and William Reeves.
Documentary Evidence
In my report, after wading through over 10 men named William Reeves in 1830 in Tennessee, I finally found couple pieces of indirect evidence that allowed me to piece together the migration route and some of the children of Sally.
First, I identified a Reeves family living near William Keaton in Anderson County, South Carolina. Then, I found that this man, Burgess Reeves, had married a woman named Francis Mauldin. Later, I found a family tree with a man named John Mauldin Reeves listed as the son of a William and Sarah Reeves. I believed this was Sarah “Sally” (Keaton) Reeves. They moved to Gibson County, Tennessee, and then to Desoto County, Mississippi.
I uploaded a PDF of my research report to Claude.ai by Anthropic and prompted,
You are an expert genealogist. I’d like to take the findings of my research report and combine them with DNA evidence to write a DNA proof argument that Sally Reeves and her descendants discovered in the report are biological descendants of William Keaton. List the documentary evidence already found in the report.
The response was:
Certainly! I’d be happy to help you create a DNA proof argument combining the documentary evidence from your research report with DNA evidence to demonstrate that Sally Reeves and her descendants are biological descendants of William Keaton. Let’s start by listing the key documentary evidence found in the report:
Key documentary evidence from the report:
-
William Keaton’s estate packet (1830) lists Sally Reeves as a daughter and heir, married to William Reeves.
-
The estate packet indicates that Sally and William Reeves were living outside of South Carolina in 1830.
-
Burgess Reeves’ will (1811) in Pendleton District, South Carolina, names William Reeves as a son.
-
Burgess Reeves was a neighbor to William Keaton in the 1790 and 1800 censuses of Pendleton District, South Carolina.
-
The 1830 census of Gibson County, Tennessee, shows a William Reeves household with ages matching the expected ages for William and Sally.
-
John Mauldin Reeves, believed to be a son of William and Sally, married Charlotte Ballance in Gibson County, Tennessee, in 1841.
-
The 1850 census of Desoto County, Mississippi, shows Sarah Reeves (age 53, born in South Carolina) living next to John M. Reeves (age 31, born in Tennessee).
-
Naming patterns in John Mauldin Reeves’ family (children named William and Sarah) suggest a connection to William Reeves and Sally Keaton.
-
Sarah Reeves (born 1796-1798 in South Carolina) is found in the 1860 and 1870 censuses in Arkansas, consistent with Sally Keaton’s estimated birth year and birthplace.
-
Gravestones in the Reeves Cemetery in Izard County, Arkansas, for Sarah Reeves (1796-after 1880) and John M. Reeves (1819-before 1900) provide further evidence of the family’s migration path.
This documentary evidence provides a strong foundation for the relationship between Sally Keaton and her father William Keaton, as well as her marriage to William Reeves and their descendants. To create a DNA proof argument, you would need to combine this evidence with DNA test results from descendants of Sally Reeves and other known children of William Keaton.
I noticed a key piece of evidence missing, so I said,
We need to add a piece of information about Burgess Reeves’ wife, and her surname Mauldin. This is key for connecting John Mauldin Reeves to his proposed mother, Sally (Keaton) Reeves, whose father lived near Burgess.
Claude responded:
Here’s a summary of the relevant documentary evidence, including this new information:
-
Burgess Reeves’ marriage: Burgess Reeves married Frances Mauldin in Granville County, North Carolina. This explains the origin of the “Mauldin” middle name in John Mauldin Reeves.
-
Proximity of Reeves and Keaton families: In the 1790 and 1800 censuses of Pendleton District, South Carolina, Burgess Reeves was a neighbor to William Keaton (Sally’s father). This proximity suggests a close relationship between the Reeves and Keaton families.
-
William Keaton’s estate records: Sally Reeves is listed as a daughter and heir of William Keaton in his 1830 estate packet. She is mentioned as being married to William Reeves.
-
Burgess Reeves’ will (1811): Names William Reeves as a son. This William Reeves is likely the same person married to Sally Keaton.
-
John Mauldin Reeves: Born around 1819 in Tennessee, according to census records. His middle name “Mauldin” appears to be derived from his grandmother’s maiden name (Frances Mauldin).
-
Migration patterns: The Reeves family moved from South Carolina to Tennessee before 1819 (John Mauldin Reeves’ birth). This aligns with information in William Keaton’s estate records indicating that Sally and William Reeves were living outside South Carolina by 1830.
Addition of DNA Evidence
One of the future research suggestions I had listed at the end of this report was to search for DNA evidence. While I was chatting with Claude, I decided to check for any DNA matches that were easily findable in my mom’s DNA matches. I checked Thrulines – no luck. Since no trees had Sally Keaton with children listed, there were no DNA matches descending from her. I will need to add my hypothesized son, John Mauldin Reeves, and wait for the Thrulines to generate.
I looked at my colored dot group for known Keaton descendants and then checked their shared matches. I filtered the shared matches of these matches by anyone with the surname Reeves in their tree. This brought up a few matches. I checked their trees until I found one that had John Mauldin Reeves in their tree! Then I thought I would ask Claude.ai if it could generate a descendancy diagram to show how my mom and the match descend from William Keaton through Sally and through her sister Lucindrilla. I was surprised and excited to see that Claude was able to generate code that would then generate an SVG image file with the diagram including labeled colored boxes!
The diagram was not perfect, but learning that Claude is capable of this was an exciting discovery. Here’s how I prompted Claude:
A DNA match, “Match 1,” a 3rd-great-grandchild of Sally (Keaton) Reeves, and Diana, who is a 3rd-great-granddaughter of Lucindrilla Keaton, has been found. They share 11 cM at AncestryDNA. Diana’s line goes from herself > Bobby Gene Shults > Ettie Belle Harris > Dock Harris > Malissa Welch > Lucindrilla Keaton. Match 1 is the son of “Unknown Twilley” > Elza Twilley > Martha Reeves > John Mauldin Reeves > Sally (Keaton) Reeves. Lucindrilla and Sally were likely sisters. The common ancestors are William Keaton and Katy Gresham, making them 4th cousins. Can you draw a diagram of this relationship to use in the proof argument?
The resulting diagram included the phrase “4th cousins” below the diagram, which turned out to be incorrect, because I had pictured the relationship in my mind, and tried to count the generations without a diagram, and got it wrong. In this process, I noticed that while Claude did well with drawing the diagram, it didn’t do well with understanding the relationships between people in the diagram.
I continued to find additional matches. As I asked Claude to add them to the diagram, I simply said their line back to the common ancestors.
Another descendant of Lucindrilla, Sidney’s son, also shares 11 cM with Match 1. John is the son of Sidney Francis Bone > Margaret C. Harris > Malissa Welch. Please add his branch to the diagram.
After this prompt, Claude made a couple duplicates at first, which I asked it to combine:
There are duplicates in the diagram. Merge the two boxes for Lucindrilla into one. Merge the two boxes for Malissa.
Then I simply typed the line of descent of each new match I discovered into Claude conversation, and the code was regenerated. Once the code was generated, the image appeared. I can toggle back and forth between the code and the diagram.
The final diagram with all 5 matches is below.
The result was a good diagram that I can actually use in my report. Here the benefits I noticed in using Claude to generate this diagram:
- It decided to color code each branch, to easily distinguish descendants through independent child lines
- I didn’t have to spend time creating boxes and lining them up.
- While Claude was generating code and building the diagram, I was searching for more matches. It was like a co-worker, one of the roles Ethan Mollick mentions AI can play in his book Co-Intelligence.
- When I needed to anonymize the names of matches for this blog post, I could simply type in who should be Match 1, Match 2, and so forth, and Claude generated another version.
- Anthropic does not use conversations with Claude.ai users to train their AI model unless you use the thumbs up or thumbs down buttons; or the conversation is flagged for trust and safety review. This is a good place to work with DNA matches who are often living.
- Each time a DNA match was added, a new version of the diagram was created. Each previous version was saved in the conversation as an artifact that I can continue to access.
- When the diagram was incorrect or had text spilling out of the box, I was able to type prompts to correct the issues, like “merge the two boxes for Malissa,” and “change Sally (Keaton) Reeves to Sally Reeves.”
- The code used to draw the diagram seems like a great way to potentially change the look and feel of the diagram myself, if I can learn how it works. However, prompting the chatbot with plain English is a pretty easy alternative.
- When I want to add a new branch, I don’t have to move branches around and try to fit people in.
Drawbacks of using Claude for this diagram include:
- Writing sometimes went outside the lines of the box
- Claude consistently failed to recognize the relationships in the diagram. As you can see below the diagram, it created a list of the relationships, and they are almost all incorrect.
- At first, Claude created a duplicate of Malissa Welch and Lucindrilla Keaton.
- Every time you add a new person, the code and diagram regenerate, which takes time.
- The output is a SVG file, so you cannot edit each individual box or colors without adding a new text prompt. SVG files cannot be easily turned into diagram files (as far as I could determine in my research). Claude didn’t know how to save it as a diagram file. Importing the SVG into Lucidchart resulted in one image, not a diagram with editable shapes and text.
Generating Diagrams with AI in Lucidchart
It seems that many companies are incorporating AI into their existing tools. Canva and Airtable, two of the tools I use all the time, have generative AI tools already integrated. And look – Lucidchart does too! Lucidchart is using the Microsoft Azure OpenAI Service. Lucidchart states that their AI generated diagrams will be editable within Lucidchart. Also, our data stays private – they don’t send users data back to OpenAI to train the model. The AI capabilities are currently available to all users but may be subject to a paid subscription model in the future.
In Lucidchart, start a new diagram, then look for the AI symbol on the sidebar (two four-point stars). In the “generate diagram” box that appears, type a text prompt for how you want your diagram to appear. Here’s the prompt I gave it:
Draw a descendancy diagram showing how two DNA test takers descend from William Keaton. The first DNA test taker is Diana, son of Bobby Gene Shults > Ettie Belle Harris > Dock Harris > Malissa Welch > Lucindrilla Keaton > William Keaton. The other DNA test taker is Match 1, the child of “Unknown Twilley” > Elza Twilley > Martha Reeves > John Mauldin Reeves > Sally (Keaton) Reeves > William Keaton.
Here is the result! Interestingly, this diagram was created with the common ancestor on the left, and the descendancy lines flowing horizontally to the left, instead of the top down diagram Claude generated.
Below the text box to type your prompt is the option to choose an orientation. Switching the orientation from horizontal to vertical and then clicking “generate” resulted in the following diagram. To enhance the diagram’s readability, I adjusted the size of the font and resized the entire diagram to be less tall.
Read more about AI at Lucidchart here: https://lucid.co/blog/lately-at-lucid-spring-2024.
Conclusion
We no longer need to draw our own descendancy diagrams. Gone are the days when we have to move boxes around, try to line up the generations, or move whole branches to make space for new DNA matches! This is an exciting use of AI. Share in the comments below any discoveries you have made with drawing diagrams with AI.
Learn More
Learn more about using AI tools in our hands-on workshop, Research Like a Pro with AI.
11 Comments
Leave your reply.