In the case of the latest Samsung flagships, however, those impressions are at times skewed. The Galaxy S23 Ultra recently involved itself in a rather interesting conspiracy, revolving around the much marketed “moon photos”. The gist of it is that Galaxy phones are using AI in order to recognize and pretty much place the moon where they think it should be.
Now that some time has passed since that debate, we can afford becoming witness to a brand new fiasco: the Samsung photo remaster AI adding teeth to photos of babies. On the one hand, it is absolutely disturbing, while on the other, it is oddly accurate and we’d congratulate the algorithm, if it wasn’t absolutely unnecessary. But hold up, let’s elaborate on the photo remaster feature. It is tucked away in the Gallery app on Samsung smartphones running on One UI 5.1 (or later). It’s meant to automatically improve user photos through removing unwanted lens distortion, reflections and other generally bothersome imperfections. That being said, it is not auto-applied upon taking a snap, so you can rest assured that no weirdness is imminent when just taking snaps with your brand new Galaxy phone.
The feature is operated entirely by an AI, which seems to have an odd fascination with dental features, as some online users have come to believe. This is because the AI basically equipped a 7 month old baby with a full set of pearly whites. This was discovered and documented by Twitter user @earcity in the form of several videos.
The results haven’t been that easy to replicate by others online, but the Twitter user that shared the video, claims that this has happened on numerous occasions. As of now, no one on Samsung’s end has commented on this.
This is one of those cases where it is impressive to see an AI do such a fine job with adding a previously non-existent element, but at the cost of feeling like you are taking a stroll through the uncanny valley. We are still in that turbulent evolutionary step, where AI is still learning, albeit rapidly developing. In other words: this will likely become addressed in a future tweak to the algorithm.
