Discussion
OK, this is genuinely scary. It's a picture of me, aged 5. It was taken in 1960. It was scanned and uploaded to Google Photos in 2002. No one has told Google where it was taken. But the map, bottom right, is correct.
Obviously, in 1960, my father's pre-war Exakta camera did not have GPS. Could you identify, from that picture, exactly where it was taken?
Google can.
@simon_brooke be careful now with any images of kids. Google flags things and people are being dragged into shit unnecessarily. Do not store images longterm with their or any of the other main cloud services.
@simon_brooke Google street view, modern GPSed photos and GeoGuesser are a very rich data set 🤔
@simon_brooke Their ability to pinpoint an image’s location from minimal clues is going to be very handy for police states. And with Google’s new motto apparently being “Be Evil,” you know they will not miss an opportunity to make a buck by handing over the locations of dissidents.
@simon_brooke Google Lens has been able to geolocate old photographs for a few years now. This article from three years ago gives an indication of some of their information sources from that period. Google will have vastly more information now, including other old photos, contemporary photos (Google Maps Street View includes 360° images from hikers), old maps, satellite photos, etc. The amount of data they have must be truly mind-boggling. Combined? Very powerful.
https://www.bellingcat.com/resources/2022/08/09/using-new-tech-to-investigate-old-photographs/
@simon_brooke I bet some experienced geoguesser players could 
@simon_brooke
That is all scary indeed.
This is genuinely astonishing and scary. I will try this with some scans of old photos, I have lying around.
@simon_brooke Geolocation of photos by visible elements on the background has been a thing for a while, that's why e.g. military photos undergo strict check by specialists, eventually background blurring, before being released, and failure to do so really costs lives.
It is because of AI. It obviously recognizes location according to landscape. You really cannot escape from their AI bullshit anymore. Feels like privacy intrusion.
@simon_brooke Google can reverse engineer images to find matches. If a similar background photo is available online it's possible for Google to match it.
Friends have done similar matches in identifying social media trolls from countryside snapshots used in profile pictures.
@simon_brooke Our local cycling community has this game where you should identify where a photo of a bike with background is taken, cycle there, take a pic and post a new pic. You'd me amazed at how good people are identifying locations, but they are taken now, not tens of years ago and the most frightening is that a computer can do it.
@simon_brooke Scary, indeed. Imagine, a fascist system using #Google.
BTW, they don't need GPS data from a camera. The have the largest photo data base with street/landscape photos, trained #AI #machineLearning to identify structures and compare #photos. If trees or walls are still there, it's easy. They used users for trainings like "click all parts with a bus". And they have your faces for #faceRecognition in the future ...
Eerie… but then again context is everything. Google has access to a huge amount of information in the images and exif information if available. Correlating all of this across its huge user base provides possibilities we cannot even imagine.
These companies and their tools already know more of us than we know about ourselves. We are the product.
Ever realized why we need rules and regulations around privacy?
If that isn't startling enough, try this one. This one shows a cat crossing a dirt track, with a dry-stone dike behind it. Above the dike, trees of a small spruce plantation can be seen.
The picture was taken in about 1966. I think by me with a box brownie -- a very cheap, primitive camera.
Again, the location shown is correct within 50 metres. HOW?
@simon_brooke with data and inference a lot is possible, i'm mostly surprised that they're allocating resources to do it automatically
could be interesting to upload pictures of places that changed dramatically since the 60. I know a lot of fields are now urban projects now. can google trim its reasoning on data set from a particular decade ?
@simon_brooke I wonder if other people have geotagged similar (enough) pictures of the same place.
Are there any that are obviously wrong? Or just some that don’t get geolocated?
@drnoble there are one or two that are very wrong. But not many.
@simon_brooke interesting, so not just lucky guesses.
@simon_brooke I see a telegraph pole with wires crossing a track. Is it still a dirt track?
Plus what I thought was another telegraph pole, maybe an end pole, near the left of the photo, although I don't see cables from there to the central pole.
Second guess: a pub sign, no, it's something with a junction box on it. Utilities?
@grahamperrin It is still a dirt track (and the dike is still there), but the electricity poles are now different, and the spruces in that wood have been felled (although there is still a wood there).
This is where Google thinks the photo was taken. The actual location was outside the white cottage about 30 metres north of the location Google has guessed.
Interestingly, this picture of my little sister and her cat -- taken only a few minutes before or after the other cat photo, with the same camera, on the same roll of film -- is not geolocated. There's nothing visible over the dike in this one.
@simon_brooke i wonder if the telegraph poll helps, in the UK every single one of them has a unique number so the location is known.
Or algorithmic calculation using your meta familial data - publicly available - and their Google earth data. Equally scary.
@barfilfarm Well, presumably. The cat photo is just outside the cottage we had in the 1960s... but where did Google get that information? It could be recognised in other photos taken at the same time, but there's nothing of the cottage in the picture.
Certainly they're mashing together huge amounts of data -- but I had not conceived that it would be economically viable to do that for random photos taken by unimportant people.
@simon_brooke every photo helps to train their machine learning. The more, the better. @barfilfarm
@simon_brooke
Maybe they trained their AIs long enough by letting people play geoguesser?
@simon_brooke
Is it using the upload location?
@Alternatecelt I scanned them all when I was living in Auchencairn, and almost certainly uploaded them all from there. The beach photo is from Carrick shore, by Gatehouse. The cat photo was taken on a farm about two miles from Auchencairn, so fairly local to where it was uploaded.
Pictures from Islay and from Skye in the same set are not geolocated despite having far more identifiable detail.
@simon_brooke @Alternatecelt Can you rule out that they were located from associated text? You have precisely located them in your posts here; presumably you have not mentioned them online previously?
@kbm0 @Alternatecelt there is no associated text. Just, after my mother died, I scanned all the pictures in her family album and uploaded them to Google Photos for the rest of the family to see.
@simon_brooke @kbm0 @Alternatecelt They also used the pictures and location data’s from your phone and the phones of your relatives and friends (contacts).
@simon_brooke @Alternatecelt I wasn't thinking about directly associated text, more likely that you or someone else had discussed the photos in a social media thread or similar. That would be enough to pinpoint them.
@simon_brooke How's it done that, I wonder? Very unpleasant
@janeishly They do have a huge library of pictures; presumably they can compare them. And the beach photo -- well, there's not a lot of detail in that photo, but nevertheless thousands of people must have taken pictures of that beach.
But the cat on the track? That's on a farm the present owner of which is very hostile to strangers. It's *really* unlikely that there are many photos of it.
BT Free is a non-profit organization founded by @ozoned@btfree.social . It's goal is for digital privacy rights, advocacy and consulting. This goal will be attained by hosting open platforms to allow others to seamlessly join the Fediverse on moderated instances or by helping others join the Fediverse.
