[gojomo]

Potentially, it might help find errors in the existing year-labeling - either via recognizing cases where guessers have wide divergences with a certain bias, or the explicit avenue via the about (email with errors).

Or, help identify features of a photo that definitively place it at certain times, by human perception. That might help tune autogenerated content intended to reflect certain eras to avoid anachronisms, or more strongly hint specific years or vague ranges.

Now, I don't think these are likely motivations for the game – the mere fun of playing seems enough to me. But it seems they could be.

[system2]

Pretty far fetched. The date of the photos are known so ai's only intention would be to find out if humans suck at guessing or not. Which would be a useless data to improve the systems.

[gojomo]

But some of the dates have errors, which one might, in some applications like to find.

And alternatively, how well a scene matches the audience's clear indicators of era is valuable in entertainment – even moreso than accuracy, in some cases.

Again, I don't think these are likely motivations for this particular system. But in a wide world of many aims, it's within the realm of possibiity that someone might try a system like this for such purposes.