When Nostalgia Becomes Training Data

“Mom/Dad, what were you like in the 90s?”

The newest social media trend has people happily uploading photos of themselves from 25 or 30 years ago. (A few weeks ago it was something similar: people posting photos of themselves from 2016.)

Most people aren’t really thinking about AI training data. They’re thinking about how cool – or hot 😉 – they were then. Fair enough.

But there’s another layer worth noticing: When someone posts an old photo under their real name, connected to their current profile, they’re creating a remarkably clean training datapoint for AI systems. Their face from decades ago, linked to the same face today. Verified identity. Clear labeling of the time period.

That kind of data is extremely valuable for training models that analyze faces, aging, and identity. Normally companies have to spend a lot of money assembling datasets like that. Social media trends turn it into a volunteer exercise.

For celebrities, this is less meaningful. Photos of them from decades ago are already widely available. For everyone else, though, those images often didn’t exist online until now.

To be clear, I’m not saying people shouldn’t share old photos. Looking back is fun, and many of us have images floating around online. But it’s a useful reminder of something about the AI era. We aren’t just using the machines; we’re constantly feeding them. Most of the time, we don’t even realize we’re part of the training set.