I’m sure you’ve heard of (and probably used) a new phenomenon called AI. I’m curious about it, but also cautious. I apparently don’t like drastic change, and with all the movies, blogs, books and articles about how AI is going to take over the world one day, I feel we should all be cautious. Some say nothing will happen, and I hope they’re correct, but I will be watching.
There hasn’t been much use of AI on my computer, but my phone is a different story. There are many apps that use AI, and it can be great. Research can be much quicker, and with better results.
I recently read an edition of “American Ancestors,” a magazine from The New England Historic Genealogical Society. The highlight of the issue was “How AI is Changing Genealogy,” and had articles by James Heffernan, Melanie McComb and Maureen Taylor. The articles list the pros and cons of AI use in genealogy — and from what I read, much of the advice is the same as with our regular genealogical research: be specific in what we ask for, and watch for things that might be incorrect.
That includes the proper asking of questions. If we don’t understand the parameters of the program we’re using, we won’t be able to ask our questions in such a way that we’ll get the answers we need. Taylor says every prompt/question has three parts: a role, a goal and a task. We need to tell the program what to do, and what results we want to find. Otherwise, the system might “hallucinate,” or make up an answer that aren’t correct.
Taylor also says two people asking the same question of AI can receive different answers. That’s because the AI platform “learns” from us and tailors responses to what it “thinks” we want to know. Taylor says best practices include citing sources, telling our readers we used AI, and including the date we did the search.
A main issue with AI is often photographs. We can tell our AI to make a photo of someone, and then offer what we want to see in the photo. I’ve seen some photos where a person has six fingers on each hand, or one that’s fairly famous of a man kissing another man’s feet, but both feet are left. If the instructions aren’t specific enough, there can be errors.
This goes for historic photos, too. If you have an old cabinet card photo of Aunt Tillie that’s seen better days, you might want to use something to enhance the photo. That process could remove creases, missing parts of the photo, spots, discolorations and other problems — but the “fixed” photo may not have the details the original had.
For example, there was a photo of a woman in Heffernan’s article. After the enhancement and colorization was complete, the new photo looked very nice, but the AI program didn’t catch some of the details.
There was a curl on the woman’s forehead (part of her hairstyle), but the program changed that to a wisp of hair. There was a lace portion of her collar, which is now missing, and there was a faintly seen addition to her dress eliminated from the final product. The program also “softened” her features. All of these changes made the photo nice, but the woman looks younger in the final photo, and it’s much more difficult to tell when the original was taken.
We can also use AI to add people into a photo (missing siblings, etc.) in a way that’s almost undetectable, which can bring a family back together again, but is historically inaccurate.
AI can be used for writing reports and analyzing data, and that can be great, but it can also mislead. We’ll talk more about this next time.
— Jim Moses welcomes comments and suggestions at jmosesgen@gmail.com.
