by Dan Moren
Apple Photos’s concert identification seems to play more misses than hits
Technology professional Chris Devers1 has taken a close look at Photos’s concert feature, where it tries to tag pictures you take at musical events with the name of the show. Unfortunately, it’s a feature that’s rife with inaccuracies. Here’s just one class of example:
Apple’s software struggles with understanding who the headline act is in a multiple-band lineup.
I’m sure it doesn’t help that the listings for these shows are a metadata mess, with the names listed in seemingly any order: the headliner might be at the top of a sign, at the bottom of a poster, or in a big font on the middle of a web page.
In any case, mixing up an opener for the headliner is a common mismatch in Apple Photos concert event tagging.
For the first example, in what will become a theme, getting the tagging right for Pixies concerts seems to be a chronic problem in my Photos library. In this case, Franz Ferdinand opened for them, but FF gets top billing according to Apple Photos.
Chris has the receipts, including plenty of pictures with incorrect captions. The problem isn’t limited to confusing headliners and openers: music festivals are incorrectly labelled as a single artist, concerts are confused with shows at nearby locations or even venues with multiple rooms, and photos taken on the same date are assumed to all be at the same event, even when they’re not.
I confess this is a feature I rarely think about because I don’t go to that many live concerts. Searching my Photos library (which you can do specifically for “concerts” to see the images it’s tagged) did find a correctly identified Guster concert from June 2023—though it doesn’t mention that they were playing with the Boston Symphony Orchestra—but the vast majority were wrong, including several pictures of the Relay 10 celebration in London in July 2024, which were identified as a “Liang Lawrence Concert.” (That concert appears to have taken place the same night at a nearby venue).
All of this certainly feels like a machine learning feature just making its best guesses based on the information available with no way to determine whether something is true or not, an all-too-common occurrence. Ultimately it’s just not doing it well enough to be—stop me if you’ve heard this one before—reliable or useful. To my mind, though, the real failing—as Chris points out—is that Apple doesn’t provide any way for you to fix this. You can’t manually re-tag or even simply remove the incorrect tag. That feels like a real oversight and turns this feature from half-baked to totally uncooked.
- A fellow Somervillain! ↩