Saturday, January 25, 2025

Ask AI: "Provide a Summary and a Source Citation For An 1851 English Census Record For John Richman Household"

The Artificial Intelligence tools (like ChatGPT, Claude, CoPilot, Gemini, Grok, Meta, Grok, GenSpark and Perplexity) can provide much useful and nearly instant information about subjects in a concise format, especially when they have been pre-trained on a large volume of information about the subjects.

1)  I wondered if the FREE AI tools could provide me with a summary and a source citation (in EE format) for an 1851 England census record image for the John Richman household in Hilperton, Wiltshire, found on Ancestry.com.  Here is the attached record image:

I intentionally used a screen capture of the Ancestry page because it provides information that can be used for the source citation.  

2)  The first prompt was:  "Please create a summary of information and a source citation (in Evidence Explained format) for the attached record for the John Richman family found on Ancestry.com. Do not hallucinate."

The free ChatGPT4 provided this response:

Based on the image, here's the requested summary and citation:

Summary of Information

The 1851 England Census for the parish of Hilperton in Wiltshire lists the John Richman family residing in the village of Hilperton. The household members include:

  1. John Richman (Head): Male, age 40, occupation – Farmer of 35 acres, born in Hilperton, Wiltshire.
  2. Ann Richman (Wife): Female, age 39, born in Seend, Wiltshire.
  3. George Richman (Son): Male, age 12, born in Hilperton, Wiltshire.
  4. Elizabeth Richman (Daughter): Female, age 10, born in Hilperton, Wiltshire.
  5. Mary Richman (Daughter): Female, age 8, born in Hilperton, Wiltshire.
  6. Jane Richman (Daughter): Female, age 5, born in Hilperton, Wiltshire.
  7. Emma Richman (Daughter): Female, age 2, born in Hilperton, Wiltshire.

Citation (Evidence Explained Style)

"1851 England Census," Wiltshire, parish of Hilperton, village of Hilperton, John Richman household, HO 107/1840, folio 267, The National Archives, Kew; digital image, Ancestry.com (https://www.ancestry.com : accessed [date you accessed]): original data from The National Archives, London, England.

Let me know if you'd like to refine or expand this!

Well, there are many errors in the summary.  The handwriting is somewhat hard to read.  But ChatGPT4 invented three names and missed two names.  It got ages wrong, two occupations wrong, a birthplace wrong, and missed that two children are listed as grandchildren.  It completely missed the Residence column, and the Condition column.

3)  My second prompt was trying to be helpful:  Please try again. The ages of John, Ann and Elizabeth were wrong. The occupation of John is wrong. Ann's birthplace is wrong. Elizabeth has an occupation. there are no George, Jane or Emma persons in this family. There are two young children - Mary and Rosa, who are grandchildren of John and Ann.

The ChatGPT4 response was:

Thank you for catching those errors! Let me carefully reanalyze the record and provide an accurate summary and citation:


Summary of Information

The 1851 England Census for the parish of Hilperton, Wiltshire, lists the household of John Richman. The details are as follows:

  1. John Richman (Head): Male, age 68, occupation – Pauper, formerly a Mason, born in Hilperton, Wiltshire.
  2. Ann Richman (Wife): Female, age 70, born in Bradford-on-Avon, Wiltshire.
  3. Elizabeth Richman (Daughter): Female, age 33, occupation – Glover, born in Hilperton, Wiltshire.
  4. Mary Richman (Granddaughter): Female, age 5, born in Hilperton, Wiltshire.
  5. Rosa Richman (Granddaughter): Female, age 2, born in Hilperton, Wiltshire.

Citation (Evidence Explained Style)

"1851 England Census," Wiltshire, parish of Hilperton, village of Hilperton, John Richman household, HO 107/1840, folio 267, The National Archives, Kew; digital image, Ancestry.com (https://www.ancestry.com : accessed [date you accessed]): original data from The National Archives, London, England.

Well, it did somewhat better - it got the names right, but the ages of John, Ann, Elizabeth, Mary and Rosa are wrong, John's occupation is pretty clear on the record - a Butcher.  Ann's birthplace is hard to read - it should be Devizes.  Elizabewth's occupation is wrong - it should be Weaver,  Mary's occupation was missing - Errand girl, and Rosa's occupation was missing - Scholar.  Rosa's birthplace was wrong - it is listed as Somerset Bath.  

The citation in both tries was very good - almost perfect.  

4)  My hope was that the free ChatGPT4 was mature enough to read the handwriting and get the summary and citation correct.  If it was just one or two minor errors in the handwriting, I would feel a lot more confident in this type of Summarization request.

Unfortunately, it didn't provide enough correct information, especially on the second try, to use it to accurately capture information about a family in a census record.  

An alternative would be to provide the record without the Ancestry information for summarization - hoping it will do better.  I tried with the same first prompt, and it gave me essentially the same response, but with additional children listed.  

This census record had clear headings for each column, but it also had slash and cross-out marks on some entries, perhaps for counting purposes.  I chose it because it was relatively simple.  A U.S. 1940 census record might have to be a fairly large image so the AI tool can decipher the headings.

This exercise can be used as a benchmark.  If the summarization improves over time, this could be a real time-saver for researchers.  However, we will still have to review and correct any errors that are made by the AI tools.  

5)  What about other free AI tools?

I tried Google Gemini (it would not try); Claude (gave a poor summary with only 2 people with wrong information); CoPilot (it hallucinated like ChatGPT4 did, with more children); Grok 2 (it hallucinated like ChatGPT4 did, with more children); and Perplexity (it required a PDF, so I made one of the image, and it would not try).  

I also tried Steve Little's Fact Extractor custom prompt for ChatGPT4 and it worked better than the others, but was still not adequate.

Can the subscription Chatbots do a better job?  If so, which ones?

6)  Passenger lists, vital record certificates, military records, and other record types with handwriting on forms come to mind for further investigation.  

==============================================

My blog posts, and those of other genealogists, using Artificial Intelligence tools are on my AI and Genealogy Compendium page.

Copyright (c) 2025, Randall J. Seaver


Please comment on this post on the website by clicking the URL above and then the "Comments" link at the bottom of each post.  Share it on X, Facebook, or Pinterest using the icons below.  Or contact me by email at randy.seaver@gmail.com.  Note that all comments are moderated, so they may not appear immediately.

Subscribe to receive a free daily email from Genea-Musings using www.Blogtrottr.com.

No comments: