top of page
Search

AI Doesn't Think. It Predicts. That's Why Genealogy Is So Interesting.

One of the questions I have been asked increasingly often over the past year is whether artificial intelligence can replace genealogists.

My answer is usually the same.

"No."

Then I pause for a moment.

"And also... yes."


That probably sounds like I'm trying to avoid the question, but let me explain.

Artificial intelligence is remarkably good at recognising patterns.


It can compare thousands of records in seconds.

It can suggest spelling variants.

It can summarise documents, translate languages and point me towards new collections I wasn't aware existed. It can assist me with managing my practice so I can be fully invested in the research and not, for example, in accounting. It can throw a good joke, here and there, so long, archival hours aren't as lonely. It can help with marketing and development - I have personally grown very fond of my AI.


I use those capabilities.

Happily.

But genealogy is an awkward discipline.

Because sooner or later every investigation reaches the point where the records stop agreeing with one another.


One document says 1872.

Another says 1874.

One census says Warsaw.

Another says Łódź.

The marriage record says Roman Catholic.

The death record says Lutheran.


At that point the question is no longer:

"Which record exists?"

When supposedly the same person exists in a different country, when she should have been already deceased: "How does one reconcile that?" when the court is asking daily for updates.

The question becomes:

"Why do these records disagree?"

That is a very different problem.


Artificial intelligence works with probabilities.

Forensic genealogy works with hypotheses.

Those two things sound similar.

They are not.


An AI model looks at millions of examples and predicts which answer is statistically the most likely.

A forensic genealogist asks whether the underlying assumption is correct in the first place.


Have the engraving on the matzeva been read correctly? Has the name of the person changed upon migration? What does the migration wave tell us about where the person can be located?


Sometimes those approaches arrive at the same conclusion.

Sometimes they do not.

One of the concepts I have written about elsewhere is what I call convergent identities.

Two different people gradually become one historical person because enough details appear to overlap.

Artificial intelligence is particularly susceptible to this type of problem.

Not because it is "wrong."

But because it does exactly what it was designed to do.

It predicts. If two identities appear sufficiently similar, the model naturally leans towards treating them as the same individual.


As genealogists, we have to be much more cautious.

Similarity is evidence, but it is not confirmation of identity.


Some of my most interesting cases have been solved not because I found another record, but because I stopped believing the obvious explanation.

I began asking different questions.

Could there be another person with the same name?

Has the jurisdiction changed?

Was the family moving between religious communities?

Is everyone assuming these documents belong to the same individual simply because no one has challenged that assumption?


Artificial intelligence is already changing genealogy.

I have no doubt about that.

In fact, one of my recent clients found me after asking Copilot to recommend a genealogist specialising in research across present-day Poland, Germany and the Czech Republic.


That alone tells us that AI is becoming an important gateway between specialists and the people looking for them.

I find that exciting.

Not threatening.

Because I don't see AI as replacing genealogists.

I see it as changing the questions we are able to ask.


The future, in my opinion, belongs neither to artificial intelligence nor to traditional genealogy alone.

It belongs to the thoughtful collaboration between the two.

After all...

Artificial intelligence helps us find information.

A genealogist still has to decide what that information actually means.


Related Services



 
 
 

Comments


bottom of page