Reorganizing Data with ChatGPT

As you can see from other posts on this blog I have been studying and commenting on the history of Stevensville, my home town. Among other things this involves looking at old documents collected and saved by friends and family members. In some cases this can be tedious and time consuming.

The type of documents I’m talking about are mostly old clippings of newspaper articles, photographs, and other material found in places like local library archives. Often I am looking for general information about things like “When did this fire take place?” and often more specific information like “Who was the father of X?”

Is there a way that AI can help speed up the process of looking through these documents?

Discovering a Step Forward
The other day I was searching through information about a particular cemetery in Stevensville. We had visited this cemetery many years ago and I was pretty sure I had seen the graves of at least one family member.

But confirming this, and identifying the person buried there is difficult when you are 2 hours away.

As luck would have it I found a useful document from the 1970s in the local library archive listing all the grave sites. The list included the names of the people buried there and some family information about them.

Manually searching through this document was certainly possible (I had a pdf of it.) But I thought it might be an interesting exercise to ask ChatGPT to analyze the document and reorder it in a searchable format that made finding things in it much easier and quicker.

Sure enough, after I copied the data from the original, uploaded it to ChatGPT, and instructed it to create a “searchable document” I got what I was looking for.

It also gave me additional information about how to turn this original doc into “a proper genealogical database-style document”. In other words there are standards for this type of document, and apparently, I had enough data to create one.

And yes indeed it created the “proper” document and I was able to find what I was looking for in a matter of seconds. There they were at grave sites 275, 276 and 277

Another Step Forward
But that wasn’t the end. Because a few days earlier we had actually been to the cemetery and physically looked for those grave sites – and didn’t find them.

It turned out that when I looked more closely at the original document (where I got the original data), it included a hand-drawn map of the grave locations. And sure enough, I could see where 275, 276 and 277 were supposed to be.

That was fine and dandy and now finding the graves should be relatively easy. We could either visit the cemetery again, or possibly send a drone down to check it out. (That was a joke.)

In the meantime I thought it would be interesting to see if ChatGPT could create a more readable version of the map. This was clearly a slightly different ballgame. Because I would be asking ChatGPT to interpret an old poorly written map with numbers that in many cases were barely legible. These numbers were not typed. They were hand written with very specific quirks (like wierd 2s.)

In about 1 minute I got a new map. ChatGPT also gave me interesting comments indicating that it knew what it was doing. For example:
“The chart is a plot/location map of the cemetery, not another name index. It appears to show the physical arrangement of the numbered tombstones, with the numbers corresponding to the numbered inscriptions on pages 1–22. The large rectangle labelled “CHURCH” is a major reference point, and “ROAD” is marked along the right/lower portion of the plan.

I can distinguish long sequences of stone numbers—particularly the earlier stones across the upper/left portion and the later numbers in the lower portion. There are also groupings and spacing that appear to represent actual rows of graves. Some numbers are crowded together or faint, so I would want to mark those as uncertain rather than guess.”

My Conclusion
Just to be clear, the map was not perfect. Some important numbers (grave stones) were left off the map, and the positioning could have been better. But on the whole it was pretty good.

This exercise confirmed my previous conclusion that the value of AI is easy to under-estimate. If we don’t have the imagination to test its abilities in new areas we will not discover what it is capable of doing, and how it can actually help us.

Unlike many negative comments and evaluations of AI, this one seems pretty positive. AI can do stuff for us if we have the imagination to ask it.