I'm sure AI-based ageing can do a good enough job to convince many people that a fake image of someone they haven't seen for years is an older version of the person they remember; but how often would it succeed in ageing an old photo in such a way that it looks like a person I have seen recently and therefore have knowledge rather than guesses about exactly what the years have changed about them?

(Not a rhetorical question to disagree with you, I genuinely have no idea if ageing is predictable enough for a high % result or if it would only fool people with poor visual memory and/or who haven't seen the person in over a decade.)

I feel like even ignoring the big unknowns (at what age, if any, will a person start going bald, or choose to grow a beard or to die their hair, or get a scar on their face, etc.) there must be a lot of more subtle but still important aspects from skin tone to makeup style to hair to...

I've looked up photos of some school classmates that I haven't seen since we were teens (a couple of decades ago), and while nearly all of them I think "ah yes I can still recognise them", I don't feel I would have accurately guessed how they would look now from my memories of how they used to look. Even looking at old photos of family members I see regularly still to this day, even for example comparing old photos of me and old photos of my siblings, it's surprising how hard it would be for a human to predict the exact course of ageing - and my instinct is that this is more down to randomness that can't be predicted than down to precise logic that an AI could learn to predict rather than guess at. But I could be wrong.