I’ve been following this thread for the last 3 weeks and was hoping for an answer. I think I’ve finally sorted my issue out (hasn’t finished yet, but increased a lot last night).
So results would return in Photos search for me, but it always said 10% optimised, and never budged. I didn’t have iCloud Photos synced since I moved onto the 15 Pro (no specific reason, just hadn’t done it).
Firstly I tried a restore from iCloud (rather than a nearby phone which I used when I first set up the phone) - nothing happened. Then I decided to move to the public ios 18.2 beta - still nothing, even after a few days. Then I noticed something peculiar. Even though the app said that the work to tag People was still going (so it didn’t show any), if I went on some of my older photos and swiped up, there were people actually tagged (I saw the circular image of the person in the lower left). After clicking on that, I saw that it came up with references to images that weren’t even on my phone (they were in iCloud however), so somehow there were references to old photos in the backup I had used to restore my phone the first time I got it, suggesting it might be breaking because of dependencies between the tags that were on my phone and photos in iCloud.
Simply turning on iCloud photo sync then worked for me. So it was a simple fix for myself, though something that I really shouldn’t have needed to do.
Yesterday, I went searching for a photo in my massive library where I knew there was braided hair. I searched for “braid”, but I got the predictable:
“Some results may not appear ”
message, and got almost no hits. I began to think “I think you mean ‘some results WILL NEVER appear!’”
But today, I decided to try it again, and I got dozens of accurate hits of people with braided hair, including the one pic from 5 years back that I was looking for. And the message went away. So this has me speculating that searching for “braid” began a background job searching for hits, a job that didn’t complete until some time later. And instead of telling me there’s a job running, Apple just tells me not expect “some results” at the moment.
The only problem with this hypothesis is that today I went back to my previous search term “helicopter”, and it still says “Some Results May Not Appear”. I do get 37 hits, though; and I’m not sure if that’s more than before. But apparently, Photos doesn’t think it’s done searching? Perhaps a few more months of time should do it?
I have a new hypothesis… Maybe Photos knows that it’s not good at identifying helicopters? And so “Some results may not appear” really means “I can spot some helicopters, but I know I’m not very good at it, so you might want to scan through your 120,000 photo library yourself looking for them.”
I have some good/bad news. The seemingly strange results you all have been seeing are probably Photos working as it should. I do not know the details, but I am pretty sure the method by which Photos creates an index of people, things or features, is some form of ANN processing. This training takes a long time. It is partially a bootstrap process. Photos knows in general what a face is, and on a first analysis pass can detect some faces. This database of local faces is then sorted into groups of similar faces, those faces are analyzed for why they are similar, and these refined concepts of patterns become name people to search for, and are reapplied to the original data for a more targeted search. This is only a speculative example of the sort of thing the ANN processing capacity is being used for. The point is the patterns searched for are derived from refined analysis of the data, in a recursively iterative process. This is how Artificial Intelligence, or Machine Learning, whatever the current buzz word is, works.
Years ago, not so long after the new photos application had been released, there was a similar discussion of why everything takes so long. At that time, based on reasoning similar to that expressed above, I suggested it might take months for analysis of a large existing library of photos. Apple support was following that discussion, and confirmed my estimate of the time which might be needed.
Today’s Macs have much more powerful processors, but the data they are asked to process is much more complex. When this much greater task is completed (if it ever is), the result is more reliable. As discussed by others above, the analysis takes a long time, and the intermediate results can seem a little flakey.
That is the way this form of machine learning works. Apple has been working on the hardware which enables machine learning, and how the data is fed in and the results interpreted, for many years.
