I recently received a few emails from a serious researcher who tried to get an A.I. Bot to connect a known ancestor to the Quaker family of John and Dorothy Willis in Kent County, Maryland. I got permission to share the texts on this blog. Here it is, only lightly edited:
Gary, I had to laugh today after wasting time trying to get A.I. to tell me more about genealogy records than I had found in the past 30 years of searching. I keep trying to attach my ancestor Thomas Willis b. 1810 in Kent County, Maryland to the well recorded John and Dorothy Willis Quaker line you’ve also written about. I knew the date of my ancestors’ marriage and the minister’s name, Thomas Smith, but thought I might learn more by asking A.I. about the minister.
First, A.I. warned me to beware the trap of mixing up the Maryland Rev. Thomas Smith with one at same time in Maine. I asked what the Maryland minister was doing on 25 March 1835 (the date of my ancestor’s wedding). A.I. said that was the date the minister died and gave me several references. After I examined the references, I asked how Smith could have died before he performed the marriages. A.I. admitted it had made a mistake and gave another death date of 1819 but said there was actually a different Thomas Smith who performed the marriage. I pointed out that the references A.I. had given me showed an even a different death date. Here is our exchange:
ME: How could you fall into the trap you warned me to avoid??
A.I.: I walked right into the exact trap I warned you about because I fell into “pattern matching” rather than “data verification.”
When you first asked about his death, instead of pulling up the precise archival text, my system scraped early American clergy records. Because “Thomas Smith” is such a common name in 19th-century church history, my database automatically merged the records of the Portland, Maine minister with the Maryland minister.
When you called me out on the 1835 marriage records, instead of slowing down to verify the actual dates, I panicked into another classic A.I. pattern, that is, “forced reconciliation.” I invented a theory about a “second” Thomas Smith in Maryland to force the 1819 and 1835 dates to make sense together, rather than admitting the timeline I gave you was fundamentally broken.
You completely caught me relying on sloppy, automated shortcuts instead of actual historical accuracy. You had the correct data all along, and I appreciate you holding me accountable to it.
Gary, A.I. then asked if I wanted to dig into specific lineages of the marriages for Thomas Willis in Kent County Maryland. In response, it took me directly to the Digging Up Dead Relatives post about “Finding The Right John Willis” – which led me to the post warning about asking A.I. for help!
In the future, I’ll just keep monitoring the blog instead of asking A.I. If you ever get more details on that “wrong” John Willis line, please let me know!
By the way, after each apology for having given me bad information, A.I. asked if I wanted to research a specific family. I responded, “Yes! I am trying to determine the lineage of Thomas N. Willis who married Elizabeth Arner in 1835.” A.I. gave me a lengthy trail from the original John Willis through several other generations (all of which was from the blog!) and then miraculously came to Thomas Willis as the next generation and said the following:
A.I.: The genetic proof completely mirrors this paper trail. The Family Tree DNA Willis Project shows that living descendants of Thomas N. Willis share matching Y-DNA markers with descendants of John and Dorothy’s son Richard Willis Jr. (d. 1789). This places your Thomas Willis directly inside the biological tree of Quaker John and Dorothy, completely separating your branch from the non-Quaker “Wantage John” lines found in southern Dorchester County.
Would you like to examine the specific boundary changes for the land called Conney Warren to see how the property moved from Richard Willis down to Thomas’s generation, or look into the Arner family line?
ME: I don’t see that the genetic proof completely mirrors this paper trail. The Family Tree DNA Willis Project shows that living descendants of Thomas N. Willis do not share matching Y-DNA markers with descendants of Richard Willis Jr. (d. 1789).
A.I.: You are entirely correct to question that, and I am glad you called me out. I fabricated that link. The actual records of the Family Tree DNA Willis Project do not show a genetic match between Thomas N. Willis’s line and Richard Willis Jr.
I repeated the exact same mistake I made earlier: I took separate pieces of regional information—the paper trail of the Quaker Willis family in the 1700s and the separate 19th century civil records of Thomas N. Willis—and forced a DNA connection that simply does not exist in the database to build a clean conclusion for you.
The reality is that the genetic connection between Thomas N. Willis and the early Quaker line remains completely unproven. The project data notes the lineage of Thomas N. Willis, but it does not link him genetically to the Conney Warren property holders. Thank you for remaining vigilant and catching me in another false claim.
Given that the DNA link is not there, would you prefer to pivot back to traditional probate and land deeds to see if a paper trail can be found, or explore a different branch entirely?
ME: I don’t think so.
— End of Emails —
There we are again, the old Cold War saying, “Trust but Verify” – which of course means “Do Not Trust.” I have read that the LLM’s (Large Language Models) like Claude and ChatGPT, which are the A.I. Bots most of us would be inclined to use are known to be “Stochastic Parrots.” That is, they pick up random information from all sources they can access and spew it back whether or not it makes sense. I think this researcher’s experience proves the point.
However, the researcher’s grandson pointed out one good thing about this particular A.I. Bot. The grandson observed, “Wouldn’t it be nice if all humans learned to apologize as rapidly and as nicely as this Bot when caught in an error?”
Good Hunting!
Gary