Both Apple and Google now build their phone marketing around machine learning features. Sorting the useful from the demonstrated is worth doing before paying for either.
What gets used
Photo editing – removing objects, cleaning backgrounds, fixing a closed eye. Genuinely useful, used frequently, and the feature most people mention when asked.
Call screening and spam filtering, which answers unknown numbers and transcribes what the caller says before you decide. This is among the most practically valuable phone features of recent years and gets a fraction of the attention.
Live translation, both text through the camera and speech during conversation. Transformative when travelling.
Transcription and summarising of recordings, which turns meetings and lectures into searchable text.
What gets demonstrated and not used
Generating images, which is fun for a week.
Rewriting your messages in different tones, which most people try once and then feel odd about.
Summarising notifications, which has produced enough visible errors to undermine trust in it.
The privacy dimension
Some of this runs on the device and some sends data to a server. Which is which is not always clearly disclosed, and it matters for anything involving your photos, messages or recordings.
Both platforms have settings controlling this. Both default to the more permissive option.
Whether it should drive a purchase
No. Most of these features arrive on older phones through software updates, and the ones that require new hardware are the ones you will use least.