EasyGenie and AI: how we use AI images

Readers may recall the image shown above from the newsletter about “Destroy if Dead” boxes. That image is one of several that were created by AI. This week, I am going to talk about how EasyGenie is experimenting with AI images … and where we draw the line.

In the past few years, a flavor of artificial intelligence known as generative AI has taken the genealogy world by storm. We’ve raved about FamilySearch’s free full-text search, which uses AI to quickly find ancestors who are mentioned in old documents and creates high-quality transcriptions from the cursive handwriting.

But we have mixed feelings about other AI applications, especially those geared toward imagery. This came up when an early AI video simulator, OpenAI’s Sora, released a video based on a prompt of “historical footage from the California Gold Rush.”

That AI video clip was riveting, but contained many historical inaccuracies:

And while another video tool — MyHeritage’s LiveMemory — looked like a fun way to animate old photos of ancestors, Nicole saw things differently.

She had a visceral reaction after witnessing the tool make changes to a 1990s group photo with friends. Her face was swapped out with someone else’s, and a mysterious boy (circled in red below) suddenly appeared from nowhere and ran to the front of the group. The boy was not in the original photo.

Nicole said:

“In the future, you won't know what's real and what's not real … Why would someone want make a video to sabotage my memory? Why create those fake feelings?”

Clearly there are problems with AI images. So why did I use Canva’s AI image generator to make a picture of an overflowing box of old papers and photos? Or a picture of a DNA kit lying on a sunlit table, in the newsletter about 23andme going bankrupt?

The short answer: I didn’t have any real photos. When I use images in the newsletter, I typically source from the Library of Congress, licensed stock photos, scans of public records, or my own photo collection.

Sometimes, though, there is no existing photo I can turn to, and I don’t have the time or the objects required to stage a photo on my own.

In years past, I might have skipped the inclusion of a photo. However, I think newsletters that only contain text are not as interesting for readers. This is one reason I may opt for an AI-generated image.

Here’s another example, also from the “Destroy if Dead” issue, depicting confidential documents being fed into an old-fashioned paper shredder. I just didn't have any other image I could use:

Importantly, the AI images that I used in the newsletter are not based on specific people. They do not purport to be accurate representations of historical events. I am merely using them to illustrate a point, make the newsletter easier to understand, and make it more inviting to read. 

I think those are valid use cases, but if you have a different perspective, please let me know.

AI is a big topic, and I have more to share. Later this summer, I will discuss how AI has helped with newsletter research, as well as a failed experiment with AI copy editing. Stay tuned!