Recently I attended a GSQ seminar presented by Andrew Redfern on the use of Artificial Intelligence (AI) in genealogy. I thought that I would try out ChatGPT and chose to ask it about the parents of my 3X great grandfather William Thorburn. Many Ancestry trees, including my own until recently, show William’s parents as James Thorburn and Janet Hogg. But the William belonging to this family was born in 1799, which did not fit with the verified information I had about my William who was supposedly 49 when he died in 1838.
William Thorburn married Marion Lee in 1819 in Edinburgh at which time his occupation was recorded as upholsterer. William and Marion had four sons – George Lee, born 1820; John Farquharson (my direct ancestor), 1822; William, 1824; and Robert Hogg, 1827. The second name Hogg lends support to James Thorburn and Janet Hogg being William’s father. But the dates do not fit.
I gave ChatGPT the basics of William’s life – his date of marriage, his occupation, and his date of burial and asked it to find William’s parents. Although not able to find them, ChatGPT’s answer was couched in language that appeals to an experienced genealogist. It did not find the names of his parents but did suggest credible ways to find out that information, including occupational records, and pointing me to a resource I had not previously found – The Dictionary of Edinburgh Wrights and Furniture Makers. When I accessed this resource I found reference to William as a wright, upholsterer, and undertaker in 1825-1827. There was also another upholsterer with the same surname, Walter Thorburn who worked in Edinburgh from 1784 until 1830.
I do like ChatGPT as it told me ‘Genealogy Tree Sources are Unreliable’ if they ‘do not cite original archival documents’ and need ‘further verification’. It searched baptism/birth records in available digitised registers (and advised me to search Scotland’s People). It found no record of William in court or will databases. It advised further searching of occupational records. It also advised checking William’s burial record and using DNA matches to establish links with other Thorburn families.
It then asked if I wanted to do a ‘deep dive’ into available online digitised records which I agreed to. Then it listed where it had searched, after writing ‘Below is what I did, what I found, and a concrete, actionable plan to try to get a firm answer’. This deep dive resulted in definitive searches for me to do in Scotland’s People using date and place filters, and variations of the surname Thorburn and it offered to do the searches for me if I provided a means to pay for the necessary credits (which, of course, I ignored).
I then added the details from William’s marriage record and ChatGPT got very excited (in a calm AI way by writing suggestions in bold typeface and prefaced with a star ⭐) as William, at the time of his marriage, was from Lady Yester’s parish in Edinburgh. It suggested a new search for his baptism in that parish recommending the use of several surname variants – Thurburn, Thorborn, Thornburn, Thurborn. ChatGPT gave me the ‘most likely scenario’ which was that William was the son of an Edinburgh tradesman; that he served an apprenticeship probably starting around 1802-1804, and that his parents may be found in parishes near Lady Yester’s.
ChatGPT suggested that I check Edinburgh directories for William, and asked for the exact date of his marriage, burial location, and names of his other children so it could more precisely search for William’s parents. When I provided a transcript of his burial information which included his residential address in 1838, ChatGPT gave me information about the area which it said was heavily populated by various craftsmen including upholsterers. Another star ⭐ and ChatGPT provided two possible families for William.
I was impressed that ChatGPT appeared to use genealogical reasoning as it made suggestions for locating William’s parents. It asked for the names of William’s other children, mentioning the Scottish Naming Pattern. I had found only four sons born to William Thorburn and Marion Lee, between 1820 and 1827. ChatGPT made suggestions for new research angles. For example, it suggested that I look for the burials of female children of William and Marion, who may not have been baptised.
ChatGPT seemed to be quite familiar with location specific (in this case, Scottish) resources. But like all genealogical research, information provided by ChatGPT needs to be verified. Whilst it took ChatGPT only a minute or so to present me with many possible avenues of research, it took me a lot longer to locate and evaluate each of its suggestions.
ChatGPT did repeat itself, even within the same project (which keeps your conversations about one topic together) and kept asking me to check baptismal records, giving me precise surname variations and parishes to check.
And it does make mistakes! When I pasted in a screen shot of William’s presence in the ‘Roll of Edinburgh burgesses and guild brethren’, ChatGPT transcribed the list (very helpful), but misread the dates of his induction – putting the date of the person above in the list against each person’s name. It told me the William in the list was not mine, but a William who married his wife (after the Thorburn surname, other Thorburns in the list were indicated by ditto marks), who became a burgess at a different date. It thanked me for the corrections, then adjusted its thinking. The other benefit of these “conversations” with ChatGPT is that it made me re-evaluate the information I already had about my William Thorburn, looking at it more closely.
ChatGPT only uses freely available information – it cannot search behind a paywall (e.g. the Ancestry, Findmypast, or Scotland’s People websites). My next step is to personalise ChatGPT to suit my needs – you can give it clear instructions about what you do and do not want it to do. For example, you can tell it not to make up answers and to identify its own hypotheses as such. My daughter is helping me with this (ironically, she asked ChatGPT for tips on how to personalise its output). And I felt I should name it – perhaps it can tell me how it prefers to be addressed?
I would recommend the use of ChatGPT as an assistant with your genealogical research. The beginning of this century has seen big changes to how we do our family history research, for example the availability of online databases and advances in DNA matching. ChatGPT, and other whole language AI models, are the latest frontier. What a change from working with fiche and film!!
Leave a Reply