When AI Puts a Face on Our Ancestors

Did you see Mark Thompson’s recent Facebook post about genealogy and AI enhancement of photographs? His post raises an important question about the potential for AI to change how we perceive our ancestors.

Many of us have experimented with colorizing old photographs or using AI to sharpen, restore, or enhance them. Those tools can produce amazing results. But as I’ve begun using AI to help tell my ancestors’ stories, I’ve encountered another issue—one that has made me increasingly cautious about the images AI adds to those stories.

What happens when AI creates an image that looks as if it belongs to our family history, but doesn’t?

When Illustration Starts Looking Like Evidence

Google NotebookLM does a fantastic job of creating videos from the source material it is given. I’ve experimented with giving it narrative reports, military records, and other genealogical material and asking it to create video overviews.

The narration can be very effective.

The visuals are another matter.

NotebookLM may supplement the story with photographs, historical images, or line drawings that I did not provide. I may recognize immediately that those images are merely illustrative. But will someone watching the finished video know that?

One of the first times I recognized this problem was while working with the Civil War pension record of my great-great-grandfather, Richmond Fisk Hammond.

I asked NotebookLM to create a cinematic video from his pension record. I liked the audio discussion, but when the video discussed Richmond’s involvement with the Grand Army of the Republic, it showed a photograph of a large GAR group standing on the steps of an impressive building.

There was just one problem.

It wasn't Richmond's GAR group.

In fact, the building was far too large and ornate to have been the setting I would associate with Richmond’s GAR activities in Dodge City.

I knew that.

But would a viewer?

Someone watching the video could easily come away believing that Richmond was one of the men in that photograph. An image intended simply to illustrate the GAR had suddenly become something that looked like documentation of his story.

Because of that visual misrepresentation, I discarded the video and started over.

But Photographs Aren't the Only Problem

Since then, I’ve tried prompting AI tools to use only photographs that I provide. I thought that would solve much of the problem.

Recently, however, I realized that even line drawings can create the same confusion.

I encountered this while creating a video about my great-great-grandmother, Rachel Elmeda Christy Ricketts.

This case is particularly troubling because I do not have—and have never seen—a photograph of Rachel, her husband, or their twin daughters. I have no visual reference against which I can compare an AI-generated representation of any member of this family.

I gave NotebookLM a narrative report about Rachel and asked it to create a cinematic video. The video opened with a line drawing of a young woman.

Is that what Rachel looked like?

I have absolutely no idea.

I know that the drawing is simply an illustration selected or generated for the video. But someone watching Rachel's story might reasonably assume that the young woman is intended to represent Rachel. Since I don't have an authentic photograph of Rachel, there is nothing within the video to counter that impression.

In effect, AI has given Rachel a face when I don't actually know what her face looked like.

So I tried again, this time requesting an explainer video.

That version included line drawings of twin girls.

Rachel and her husband did have twins. But again, I have never seen photographs of them. I have no idea whether these drawings bear the slightest resemblance to Rachel's daughters.

Yet when drawings of twin girls appear while a video is telling the story of a woman who had twins, it is very easy for the viewer to make that connection.

Then, toward the end of the video, another image appeared—one that looked like a collection of old family photographs.

Those images present an additional problem. Not only are they not photographs of Rachel's family, but to my eye they appear too modern to represent Rachel, her husband, and their children.

Once again, the visual creates an impression that the evidence doesn't support.

The Problem Is Context

None of these images necessarily claims, by itself, to depict Rachel or her family. The problem is what happens when the images are placed inside a video specifically telling Rachel's story.

Context gives them meaning.

A viewer sees a young woman while hearing about Rachel and may naturally associate that face with Rachel. A drawing of twins shown while discussing her children can become her twins in the viewer's mind. A collection of vintage-looking photographs can appear to be actual photographs of the family—even when the photographs are from the wrong time period.

Before long, AI hasn't just illustrated the story.

It has unintentionally created a visual memory of people for whom I have no known photographs at all.

And that concerns me.

Why This Matters in Genealogy

As genealogists, we spend a great deal of time distinguishing between what we know, what we think, and what we suspect.

We cite records. We question family stories. We identify photographs. We try not to attach the wrong person to the wrong family.

Shouldn't we apply that same standard to AI-generated storytelling?

An AI-generated image may make a story more visually interesting, but if viewers interpret that image as evidence, we have introduced something into the family story that wasn't actually there.

In Rachel's case, the danger seems particularly significant. Since I don't have an authentic photograph of her, an AI-generated drawing could easily become the image that I—or someone watching the video—begins to associate with her.

Images are powerful. Long after someone forgets the details of a narration, they may remember the face they saw on the screen.

Even when we have no idea what that ancestor actually looked like.

My Decision

For that reason, I have decided not to share the NotebookLM videos about Rachel Christy Ricketts.

The narration may be useful and the videos may be attractive, but I don't want an AI-generated young woman, drawings of twin girls, or unrelated family photographs to become associated with Rachel and her family.

This experience also explains why I increasingly tell AI image tools:

No photos of people.

For my genealogy work, I'd rather use documents, maps, handwriting, newspapers, gravestones, landscapes, objects, timelines, and other imagery that supports the story without inventing a face for an ancestor.

AI can be an extraordinary tool for helping us tell family stories. But perhaps part of using it responsibly is knowing when an illustration crosses the line from helping us imagine the past to unintentionally rewriting it.

What do you think?

When we have no known photograph of an ancestor, should AI-generated drawings of people be used in telling that person's story? Is a disclaimer enough? Or does simply showing an invented face risk creating a visual "memory" that future viewers may mistake for the real person?