RootsChat.Com
General => The Common Room => Topic started by: Zaphod99 on Wednesday 04 February 26 09:31 GMT (UK)
-
I didn't know where to put this article because it's not really a technical one, and it certainly isn't totally off topic. It's about a fascinating new development in artificial intelligence, not quite available yet, but can you imagine how this will change research using websites like Ancestry, My Heritage, and Find My Past. The concept is basically one of letting your browser have your details to log into these websites, which possibly won't make the website happy, but you then let it run amok doing its own research.
https://www.ghacks.net/2026/01/30/chrome-gets-smarter-gemini-can-now-browse-the-web-for-you/
Like with any new technology there is going to be a period of teething troubles but just think what it could mean.
Zaph
-
The problem with AI like that is that it chooses the most available information rather than understand the reasoning. I recently posted some research on Facebook about a specific topic to my local area. Further research has led me to change some of it: despite that AI tells me the incorrect info from the FB post because that is the first and easiest evidence it has found. Ancestry is great but Geneweb is poor as it repeats errors made by people copying errors. I have been transcribing C18 wills in Somerset from Ancestry scans. The AI translation is only 80% correct and I need to amend every will because AI cannot read it or translates it out of context based on its own incorrect predictive text.
When I here politicians that AI are going to sort out the NHS treatmentof cancer on the news this moring I seriously wonder what planet these idiots are on to rely on a useful but flawed tool.
-
but can you imagine how this will change research using websites like Ancestry [...] just think what it could mean.
Lots of rubbish created by AI instead of humans on Ancestry, I suspect.
If you could keep the AI away from people's personal trees, and make it stick to the source information, maybe you would get something useful. But to me, the key to successful genealogy is not thinking in straight lines, but using lateral thinking and investigating hunches, neither of which AI is any good at currently. Even if we get AGI (https://en.wikipedia.org/wiki/Artificial_general_intelligence) in the next 5-10 years, I doubt that it will think or reason in the way (most) humans do.
-
I've read the above comments and totally agree, but this is leading-edge technology, and it is going to steadily get better. Just think how much AI has changed in the three and a bit years since Chat GPT first appeared. It's got to the stage where it's a lot more accurate than some of the semi-literate, ill-informed responses on Facebook.
Just imagine AI, learning to behave like an experienced researcher, able to tell the difference between 20 copied trees and one carefully researched one.
And just imagine what could be possible with AI analysing somebody's DNA match results. I spend a lot of time looking at mine but I actually wonder whether I should stop, find something else to do for the next five years, and then rely on AI to do in a couple of hours what I could spend the next five years doing.
Zaph
-
Genuine question...
And where would be the fun in that?
Where would that rush come from that you get when you resolve a mystery you've been chasing for months/years? That emotion when you see an ancestors record appear on the screen in front of you, after resolving that mis-spelling / pseudonym.
It might satisfy the "I want my family tree in a day" brigade. But....
-
"And where would be the fun in that?"
Precisely. The detective work is the fun. An AI bot is not fun. Equally, bots can crawl through the web and produce a classification of all living things, their geographic distribution, their life histories, their behavior and their morphology, but it is much more rewarding and enjoyable to actually look at animals and plants in person.
-
They are good points, I like a challenger as much as the next man, but there are certain gaps in my research that I would like solving, so that I can get on with the rest of my life.
We've all sat for hours, pondering over the last clue in a crossword, before wondering whether to go for the dictionary or wait for tomorrow's answers.
Zaph
-
My son-in-law makes a great deal of use of ChatGPT and will vigourously defend its capabilities. My own experience of AI relates principaly to its use in family history research, and it is in this context that I find that any reliance on it is profoundly disturbing.
Simple queries on the internet these days respond at the top with AI generated results. I find that without exception that these results are wrong. However, the way the information gained by AI is presented appears to be authoratitive and it is only by being knowledgable in the specific area of research that the diecrepancies and absolute errors in the response can be identified. A non-specialist may well accept the AI response to be accurate, and if the respose is repeated, then the mis-information becomes spread more widely, and worringly, it starts to attract a cloak of truthfulness.
The use of AI in places such as Ancestry, etc., I think is not entirley clear, but probably does lie behing the so-called "hints". When I use Ancestry in my research, I find that often the majority of the hints are completely wrong. However, it is unsurprising that these hints are sometimes seen in public member trees accepted without any validation whatsoever, leading in one case to a mother giving birth over one hundred years after she had died! The hints also lead to a reactive and lazy form of research rather than a proactive one where information obtained is forensically evaluated to determine whether or not it supports the research or otherwise. That is not to say, however, that Ancestry's hints do not sometimes provide valuable information from sources that might otherwise be overlooked, and other avenues for research that might not otherwise have been considered.
I general, I have not found AI to be of any particular benefit in my own research; effective and successful research requires, like most things in life, simple dedication and hard work, and not a little imagination.
Please feel free to disagree.
-
Should have saved this topic for 1st April.
Martin
-
Ancestry hints are atrocious, and generally the AI summary at the top of a search is equally bad, it's a tool, it is coming, it is here to stay, and if you haven't tried it recently you should persist, as month on month they are getting very much better. They won't do the work for us, but they are tools that need a craftsman to use. I'm not trying to sell it I'm just trying to help.
Zaph
-
AI is like something that Del Boy Trotter would sell you ;D, just not necessarily stolen, but may not work properly.
On a side note I do like the recently new FamilySearch text search which can help seek out scans of words in historic documents but it is still in infancy and not infallible. You type "Joe Bloggs" into the search feature, choose a country and date range and get some results.
-
" I do like the recently new FamilySearch text search"
Yes, me too. I've turned up several interesting items with the text search.
-
In this context, AI is reliant on the quality of the records it is searching. As we all know, records are incomplete or misleading and a lot of family trees on Ancestry are the product of wishful thinking and bad inferences (I've lost count of the number of times it has been obvious that someone has relied on secondary sources rather than primary ones and fallen into quite serious error.) In my case, I have one branch of the family in C19 Nottingham who consistently told untruths to officialdom, typically to cover up illegitimacies or, in one case, bigamy, and that's only really apparent when one looks at primary sources.
Another example I've run across too often is researchers forgetting that names were often reused in an era of significant perinatal and infant mortality and the Joe or Josephine Soap baptised in 1756 may not be the individual who married in 1782 despite the same name and the same parents.
So whatever the claimed capabilities of AI, and even if those claims are true, it is a problem if it does not follow the golden rule - never trust what you see on Ancestry and never forget that some records are missing or as yet untranscribed. Relying on it is, in that sense, as problematic as relying on other researchers. If it throws up hints and possible research avenues that's all well and good, but it is no substitute for doing the work of corroborating and verifying or, as I do on Ancestry, clearly marking some 'facts' as either provisional or unverified placeholders.
(edited for typos)
-
Hi,
An interesting question. Thechnically you have been able to do this kind of thing for a long while.
But it is against the T’s and C’s of your contract, so I think you would loose access pretty quickly.
It frustrates me greatly that all the sites don’t use the possibilities more. However that is probably down to privacy issues. I suspect that a well trained AI could these days more or less instantly come up with pretty convincing trees for 90% of our DNA matches.
You can do a lot by sheer slog, with say Pro-Tools, but it is against the slog, but then also where would be the fun in that!
MyHeritage and Ancestry both have there cut down versions with Theries and Thrulines but I am sure there are very impressive things behind the scenes, but then you get to privacy.
We all see this every day with private trees and people who don’t respond to contacts for whatever reason.
It is all very much a Pandora’s Box. Be careful of what you wish for!
Happy Hunting
-
I would say that all surviving baptism records (church, NC, workhouse baptisms etc) for, lets say Norfolk 1680 to 1720 probably only are about 70 to 75% of all baptisms that ever took place in that county in those 40 years. And even then, you have to wonder how many were never baptised in those 40 years. There will be many baptisms 1680-1720 in Norfolk that have not survived due to lost records, parish registers for a certain parish not surviving that far back, gaps in PR's, destroyed records etc.
-
I have believed for quite a while that the younger generation is being dumbed down by all the dross that is fed to them and by an education system that does not teach them to think.
I am an Engineer and worked in the Building industry for years and the standard today is nothing like it was, it is the same with every other Profession.
AI is only going to make the young rely even more on a computer programme than their own brain.
It will not bode well for the future.
-
An 'historical society' in an area where I have some genealogical dealings, found a reference to a particular woman. They had no idea who she was, so one of their number evidently used ChatGPT or equivalent and prompted it to find a connection between the name and the locality, which it did. It contrived a match between a one-time landowner and a distant relative but caveated it that it was speculative. The caveat went out the window and they deemed this as a proclamation of the Oracle.
I did warn them this identification was highly flawed (for a number of easily verifiable reasons and also pointed to previous research that someone else had done 30 years ago) but they weren't having it and politely told me to shove it. These are adult people that should know better but they are beholden to the new AI overlords without understanding how it works and what the limitations are. I don't know if they ever did retract or caveat this identification but I've left them to it.
-
In a previous job in the NHS Primary care, there was an initiative to replace all paper patient records. The ides was to pay a company millions of pounds to scan and transcribe them all onto the NHS network. Having seen what AI does to will transcripts where is makes things up or fits the closest thing it thinks to be right, the idea the NHS will do that to very faded writing by a profession known for their bad handwriting in medical jargon and bespoke abbreviations (NFB mean anything to anyone in Somerset?) fills me with angst.
-
Passenger lists before 1900 can contain pretty basic info and not enough to ID a person even with a less common name. So you could descend from a "William Taylor" who emigrated to Australia in about 1855, and the AI could come up with a likely candidate for 1855, aged 25. And all is given is William Taylor, aged 25, from England, so could be any William Taylor. If the person's ancestor was called Marmaduke Bracegirdle, then it probably is the right person. ;D
Assisted immigration records can contain more info, and of course convicts.
-
It was me who started this particular topic and I have put a calendar reminder for February 2029 to have another read of it, and I implore anyone who has contributed to do the same.
I particularly note the comment from somebody saying that machines accessing subscription databases will be restricted, but I think by then it will be very difficult for systems like ancestry and my heritage and others to actually tell the difference. Read about the concept of Turing machines.
Zaph
-
Passenger lists before 1900 can contain pretty basic info and not enough to ID a person even with a less common name. So you could descend from a "William Taylor" who emigrated to Australia in about 1855, and the AI could come up with a likely candidate for 1855, aged 25. And all is given is William Taylor, aged 25, from England, so could be any William Taylor. If the person's ancestor was called Marmaduke Bracegirdle, then it probably is the right person. ;D
A name and age, as you say, are not unique pairings but people happily adopt those records when in need. On top of this is "survivor bias" which afflicts genealogy more generally, i.e. your research tends to be biased by the things you can find. AI has that baked-in in spades.
I often operate close to the 'Marmaduke Bracegirdle' rarity of names but that brings with it another bias. As people perceive a name to be exceedingly rare, they don't question it at all when they find it - it must be the same person. So what we might find is that Marmaduke was a notable family figure in the 18th century and that his name has been propagated down the generations sporadically, such that you'll find distant cousins of the same name, of similar age and similar birth location. The naïve genealogist wades in and conflates these cousins as a singular identity.
I have encountered a couple of instances where people have conflated distinct people with relatively rare names, where one was a victim of WW1 and the other not. A date-of-death is entered and that line extinguished - job done. Whereas in reality the chap is alive and well in Tooting in the 1950s.
-
Passenger lists before 1900 can contain pretty basic info and not enough to ID a person even with a less common name. So you could descend from a "William Taylor" who emigrated to Australia in about 1855, and the AI could come up with a likely candidate for 1855, aged 25. And all is given is William Taylor, aged 25, from England, so could be any William Taylor. If the person's ancestor was called Marmaduke Bracegirdle, then it probably is the right person. ;D
A name and age, as you say, are not unique pairings but people happily adopt those records when in need. On top of this is "survivor bias" which afflicts genealogy more generally, i.e. your research tends to be biased by the things you can find. AI has that baked-in in spades.
I often operate close to the 'Marmaduke Bracegirdle' rarity of names but that brings with it another bias. As people perceive a name to be exceedingly rare, they don't question it at all when they find it - it must be the same person. So what we might find is that Marmaduke was a notable family figure in the 18th century and that his name has been propagated down the generations sporadically, such that you'll find distant cousins of the same name, of similar age and similar birth location. The naïve genealogist wades in and conflates these cousins as a singular identity.
I have encountered a couple of instances where people have conflated distinct people with relatively rare names, where one was a victim of WW1 and the other not. A date-of-death is entered and that line extinguished - job done. Whereas in reality the chap is alive and well in Tooting in the 1950s.
I have often found 2 cousins of the same name, and similar ages. Can lead to confusion.
But if you did find a Marmaduke Bracegirdle was not the same person you thought, as you say they may have been cousins of similar ages, and if the father of one of them is as yet unknown, it may have been a brother or cousin of the other Marmaduke Bracegirdle whose dad is know, so may help further.
I see you are researching the surname Pawling and its many, many variants. I once had an eye on a Rebecca Palding born in 1704 in Norwich, married name Harbord, until I found their daughter Elizabeth born 1733 was a namesake cousin of the correct Elizabeth Harbord who was my ancestor born 1731. I had found Rebecca may have descended from a Christopher Pawling/Palding of Hickling whose son was apprenticed in Norwich in the 1660s, whose forename was the same as Rebecca's paternal grandfather.
-
I have often found 2 cousins of the same name, and similar ages. Can lead to confusion.
Sure, that happens all the time. The distinction that I'm pointing out here is that a very rare name has been seen to lead the unwary (or AI) to rule out any consideration for it being the wrong person. It is judged a virtual impossibility for it to be anything other than one person. The naïve assessment of rarity is flawed because they are blind to the prospect of correlation with prior (unseen or undocumented) generations.
-
Ancestry’s Thrulines is pretty weird without AI.
I fail to see how a flawed system can be corrected simply by the use of AI.
1st Law of Computing:- Rubbish in = Rubbish out.
-
Ancestry’s Thrulines is pretty weird without AI.
Aye, it effectively treats family trees as accurate and thereby suggests 'hints' that are incorrect. Easy to control for if you recognise what is happening but it looks like quite a few people don't.
-
Lots of rubbish created by AI instead of humans on Ancestry, I suspect.
The clue is in the title: Artificial Intelligence generates artificial intelligence (in the military sense).
However we have very recently been introduced by our daughter to the colourising ability of ChatGPT (for one - others can do it). If you can provide it with a decent sharp b&w photograph it can make it more lifelike, as we are used to seeing now. It's not much good for retrieving indistinct pictures, the result can look like a different set of people.
I think I have posted before our 1865 four-generations picture printed from the original glass plate, so very sharp, although the baby moved its head slightly. AI has turned it into colour which I find convincing, tho it is obviously guesswork. I suspect it may make deductions from the shades of grey, knowing how panchromatic film would have treated them. It has also unfuzzed the baby's head. The old lady was born in 1782.
-
Having been a restorer and colourist on here for 20 years, I am in danger of losing my hobby to Aartificial Iintelligence I'm afraid. However, AI won't allow you to change the colours of the outfits and won't know if Granny's eyes were Blue, Green, Grey or Brown.
It is affecting the Photo Restoration Board as the number of requests has drastically reduced, there may still be posters out there who prefer the "Hands on approach" but I won't hold my breath.
I may have to dust off my paints, Brushes and Easel :-\
Carol
-
I didn't know where to put this article because it's not really a technical one, and it certainly isn't totally off topic. It's about a fascinating new development in artificial intelligence, not quite available yet, but can you imagine how this will change research using websites like Ancestry, My Heritage, and Find My Past. The concept is basically one of letting your browser have your details to log into these websites, which possibly won't make the website happy, but you then let it run amok doing its own research.
https://www.ghacks.net/2026/01/30/chrome-gets-smarter-gemini-can-now-browse-the-web-for-you/
Like with any new technology there is going to be a period of teething troubles but just think what it could mean.
Zaph
You're joking right? You want to give over all your personal information so who ever owns your browser or the AI tech can do a deep search to collect as much information on you that they can use for their own benefit.
I'm confused of why you would suggest such a thing on a forum comprised of genealogist enthusiasts. If you find genealogical research so tedious then hire a human genealogist.
-
Carol, as long as you override the automatic settings and give AI suitable prompts you can get it to do anything with any part of the garment or physical features.
Zaph
Having been a restorer and colourist on here for 20 years, I am in danger of losing my hobby to Aartificial Iintelligence I'm afraid. However, AI won't allow you to change the colours of the outfits and won't know if Granny's eyes were Blue, Green, Grey or Brown.
It is affecting the Photo Restoration Board as the number of requests has drastically reduced, there may still be posters out there who prefer the "Hands on approach" but I won't hold my breath.
I may have to dust off my paints, Brushes and Easel :-\
Carol
-
I'm ambivalent on AI in genealogy. There are legitimate deep research areas in which high quality AI can be exceptionally useful and assistive (although I'm not sure we are there yet functionally).
The normalisation of AI in photo "restoration" on the other hand is absolutely awful. It goes beyond colourisation because it's a fundamental reinvention of the original image - no one seems to care (or think about?) the fact that the people in the photograph have had all the defining nuances of their face overwritten in the process of upscaling/"restoring" it. There's going to be a huge problem with basic authenticity of old photographs for future researchers.
-
AI restoration on close-up faces makes them all look like featureless Thunderbird puppets. It's okay for restoring damaged photographs, but restoring photo facial features is atrocious.
Zaph
-
I agree with Emerald City and Zaphod......my concern is that future generations won't recognise their Ancestors as any likeness in the original is essentially lost.
Carol
-
An interesting development
https://www.whodoyouthinkyouaremagazine.com/news/gemini-artificial-intelligence-the-national-archives-fake-records
Not only did the AI (large language model) get part of the initial inquiry wrong, when asked to provide proof of its methodology it lied - I guess the machines are getting more human ::)
-
Interesting. It underlines that we always need to do our own research, not rely on others.
Martin
-
I am a big fan of artificial intelligence services and tools, but so many people don't even bother giving a prompt, they just state what their problem is. We'll have to wait until 2030 for that. Artificial intelligence is like a hammer or a screwdriver, they are designed for specific tasks, and yes, you can open a tin with a hammer, and you can bang things in with a screwdriver but there are better ways of doing things.
However, I will recount a recent disaster with artificial intelligence. In a news article I saw a small excerpt or extract from a painting, and I uploaded that image to two different AI sites, asking for information about the source painting. Not only did they both get it wrong, they took me through several iterations of trying to tell me where in the suggested painting the extract actually appeared, and like a fool, I was zooming in where they were telling me to look, different in both cases, before finally I decided that I was being told a load of rubbish.
Zaph
-
AI smacks of SKYNET.
-
It's a tool that's here to stay and the sooner we all get used to it, the more confident we will become. People moaned about registering births, people moaned about fingerprints, people said that anyone travelling on a train at more than 20 miles an hour would die.
Zaph
-
I won't get used to it on the Photo Restoration...It's robbing me of my 20s of restoring and colouring on there. The AI versions that I springing up on there are atrocious. They look plastic and artificially produced. The colours are hit and miss and leave parts uncoloured. It searches the internet for similar shaped facial features to produced a Photofit. You can't be the personal touch.
Rant Over.
Carol