Does Family Search have a concerns department?
Does Family Search have a concerns department?
I am concerned about the integrity of Family Search. I spend a lot of time attaching sources and I am finding mistakes which I think are generated by AI.
Concern 1: Example from FindAGrave
Sarah F Devoe: Female 1842 – 15 February 1914 • L28G-LHFThe source to be attached is:
Name | Sarah Devoe Eades |
|---|---|
Birth Date | 1842 |
Death Date | 15 February 1914 |
Event Type | Burial |
Event Place | Oak Hill Cemetery, O'Fallon, St. Clair, Illinois, United States |
Event Place (Original) | Herkimer, Herkimer, New York, United States of America |
Cemetery | Oak Hill Cemetery |
Photograph Included | Y |
Note | Contains Biography |
The burial site does not match the death place. FindAGrave gives the correct cemetery in the correct place. I ignored the suggestion and entered the correct burial place. This is but one of many wrong burial sites I have corrected.
Concern 2: Places from obituaries, census records, draft records
Many times places are listed as a town with no state. AI picks a place from the thesaurus and “fixes” the missing place. An obituary for a woman who lived in Louisiana with multiple sources attached to Louisiana was listed as having died in Alexandria. AI placed her death in Alexandria, Virginia not Alexandria, Louisiana. I made the correction.
Unless researchers are diligent many wrong things will be added by clicking and accepting the suggestion.
Sincerely,
S K. Sutton
Answers
-
@s.k.sutton
1— In the past month, several Find a Grave indexing problems have surfaced and are being worked on.
A listing of several threads regarding these issues can be found here:
2a— Please see the following thread regarding obituaries:
2b and 2c — Please provide examples of issues with the place names for census and draft registrations.3 -
@s.k.sutton
To add to @Nyx773's comments, the place name list is also something engineers have been working hard to resolve. The incorrect standardization apparently arose during the migration of records to a new storage platform.3 -
Be aware, though this is not a Records issue, that there are placename problems on Draft Registrations in Family Tree caused by a BYU Record Linking Lab automated Family Tree project, as reported here:
I analysed this and reported it as one of the many BYU RLL Data Quality issues discussed here:
1 -
I believe there are several "contributing factors" but I have noticed the same "trend" with "less than stellar accuracy" in the Family Tree data. I call it "Junk Genealogy". I fear that our dependence on AI will exacerbate the problem even more. Even though this problem does not have an "immediate solution", I would hope that everyone will "belly up to the bar" and commit to making contributions to improve things (whether they be engineering, consulting, training, education, management, leadership, etc). …. just sayin.
1 -
@fnulnu1999 - the impact of AI is not, in my personal opinion, the issue, because the AI is confined to the indexing and transcribing. (Almost) Every single bit of garbage on a profile is put there by a human who has not thought through what they are doing or refuses to check the image (assuming that there is one). If the humans would think and evaluate at even a basic level, much of the junk would be caught.
6 -
The fundamental problem here is imo the battle in Family Tree between quality and quantity, where FS' messages (outside the Data Quality team, I hasten to add) seem to me to be muddled at best, particularly in relation to BYU RLL's activities.
3 -
We also must take care not to confuse tree data with record data.
FamilySearch takes problems with data integrity of records seriously. The work that Serra Nola and her team have done over recent months is evidence of that.
The tree is fully open-edit. The data there is only as good as each contributor's efforts. Many of us check our Following list every day to monitor and repair errors.
Edit to add: I know Serra Nola is no longer part of the Community. My comment did not need to be edited inaccurately.6 -
The known issues with whatever system FS started using a few years ago regarding location is very discouraging. Many concerns I raised 2-3 years ago remain incorrect. Now seems to have moved to Find A Grave and is totally messing that up too. Not sure why one would report as never is addressed. Example 1881 Ontario Census for Clinton, Huron, Ontario which I raised over 2 years ago - still incorrect. Bummer. What a mess as people attaching records often do not look at original, they assume the indexed location is correct, well, it was correct 3+ years ago, not any more! All 1881 records for Huron County incorrect. Example Clinton now shows as Huron Township in Bruce or Point Edward, Lambton. And so it goes. I still use FS but assume EVERY record with a location assignment is incorrect.
0 -
@Lori9999 I feel there has genuinely been much better engagement between users and FS internal teams in the last few months, via some much more proactive moderating and, I suspect, improved internal processes. Try the new Records Issues form, available via the Search Ask a Question screen. N.B. it's not particularly user friendly and you will at present need to keep a local copy of what you submit, but it does seem to be a step forward. I filled one in about cemetery names from Find a Grave, the relevant thread is here:
4 -
AI is being misused for source record place name corrections. It has created many indexing mistakes on the sources in FamilySearch, and source record edits are not allowed! Both problems are very frustrating when using FamilySearch.
Can FamilySearch fix its AI software? If not, a lot of time spent by users has been and is being wasted, and you will lose users.
Dear FamilySearch, please acknowledge the scope and seriousness of the problem. Please give an estimated fix date. 🙂
-3 -
Again, "AI" is being blamed for any and everything, even when AI is not remotely involved.
3 -
An obvious example of a place name mistake from AI and/or Get Involved/Verify Places is:
Lawrence Howard Smalley, L2XD-3KR, with a birthplace of Osage, Mitchell, Iowa, United States. The following source record is incorrectly indexed as Osage, Becker, Minnesota, United States. Record indexing Edit is unavailable!
"United States, Obituary Records, 2014-2023", FamilySearch (https://www.familysearch.org/ark:/61903/1:1:XMPK-8M26 : Fri Nov 29 21:06:22 UTC 2024), Entry for Lawrence Howard Smalley and Rev Marilyn Sargent, 23 March 2024."
0 -
@DavidESmalley - It is absolutely legitimate to raise our concerns with FamilySearch about accuracy of things like placenames. However, it does the complainant's credibility no good if AI is blamed when it's been nowhere near the proceedings. Quite apart from anything, if FS Support follow a script driven process, then complains about AI, risk being prematurely closed "because AI is not involved".
The issue of the wrong Osage (or whatever) being chosen is one that has been around for, quite literally, years. FS has been working on the fixes for years - because they are tight lipped about what they are doing (which is normally understandable) we have no idea about how they are progressing. But what is certain is that the issues came up long before FS started using AI. The issues can, and presumably have, arise from the purely algorithm driven approach that FS used for many years. I have no visibility of anything to say whether the original deterministic approach has been supplanted to any degree by AI (which is generally driven by statistics at root) but I'd suggest not referring to the issues as being AI linked because of the risk of premature closure.
5 -
@fnulnu1999 @DavidESmalley
Please read the following post (including the post linked in one of the replies) for a fairly thorough discussion of how:
— Artificial Intelligence is actually used by FamilySearch
— Automated Computer Scripts are used by Brigham Young University's Record Linking Lab1 -
Thank you to everyone who has taken the time to share your comments, experiences, and concerns. We genuinely appreciate the thought and care that many of you have put into explaining the issues you’re seeing.
This discussion has grown far beyond the original question and now touches on several separate topics, so we will be closing the thread for now. Before doing so, we want to acknowledge the feelings behind these posts. Many of you are trying to protect the quality of the shared tree, and it’s clear how deeply you care about accuracy, integrity, and the experience of other researchers. That commitment is valued.
AI‑assisted features and related technologies are still being developed, evaluated, and improved. When something isn’t working as expected, specific examples and well‑documented feedback are especially helpful. They give the responsible teams the clarity they need to understand what’s happening and where adjustments may be needed.
We also recognize that several detailed suggestions and supporting materials related to the BYU Record Linking Lab have already been submitted through the Ideas process. We appreciate the time, effort, and patience involved in documenting those concerns. Submitting an idea ensures that the appropriate teams can review it, but it does not guarantee a particular outcome or immediate change.
The Community team cannot investigate individual projects, make development decisions, or provide updates on the status of submitted ideas. Reposting the same concerns in additional discussion threads will not create a separate review or speed up the process, and it can unintentionally create stress for other community members.
Please remember that FamilySearch Community is meant to be a place where everyone can come together, share experiences, and discuss issues without feeling threatened, discouraged, or overwhelmed. Keeping conversations respectful and focused helps maintain a welcoming environment for all.
For new or substantially different suggestions for improving FamilySearch, please continue to use Suggest an Idea. Including clear examples, Person IDs, screenshots, links, dates, usernames or project names, and a description of the improvement being requested gives the reviewing teams the information they need to properly evaluate a suggestion.
We appreciate everyone’s dedication to protecting the accuracy and integrity of FamilySearch. Because the main concerns raised here have already been formally submitted and the discussion has moved well beyond the original topic, we will now close this thread.
2
