“Practical Applications of Machine Learning” by Blaize Kaye

My tab beeps and vibrates. There is a message from my Fotobot, the subject: “New Slideshow: Good Times!”

The Fotobot is, according to the manual, a “state of the art machine learning driven photographic curation program.” It digs through your photos to remove red-eye, fix blurry images, and tag people whose faces it recognizes from your social networks.

It has recently learned to compile slideshows.

I tap on the message and my tab opens to a video. A simple electronic tune plays over a black screen which fades into a picture of Julie and I on a beach. New Year’s Eve, 1999, our first vacation together. It’s early in the night and we’re sitting near a driftwood fire. She has her feet buried in the sand.

Machine learning algorithms built into the Fotobot enable it to recognize the same face in a series of pictures taken over the course of many years. It can recognize that under the weight and the wrinkles, and despite the fact that I no longer have a full head of hair, I’m the same person as the young man in the picture. A young man, barely out of high school, sitting on a beach with the most beautiful woman he’d ever met.

After a moment, the image changes. Now it’s Julie and I at a party, our housewarming. She’s wearing a short yellow dress and is drinking cheap wine out of a champagne glass. I’m drinking beer and am in a black t-shirt and khakis.

Machine learning isn’t only able to recognize faces, it can also infer emotion and mood. This is how the Fotobot was able to analyse this picture of us and determine (perhaps from the arch in her eyebrows, or the contour of my lips) how happy we were in our small apartment. Despite the fact that all we owned was a hand-me-down mattress and busted couch, despite the fact we were broke, the Fotobot could still infer that we were happy.

Another image. Now we’re in front of the Colosseum. We saved up for more than a year to visit Rome. She’s wearing a purple skirt and white blouse, I’m in a white t-shirt and jeans. There’s the slightest touch of grey in my hair. My left arm is draped over her shoulder.

The picture changes. Now it’s Julie and I inside a mirrored elevator on the day we met Ashleigh. Julie’s holding her swollen belly and is taking a photo with her free hand. I’m standing behind her, grinning widely, arms laden with bags and blankets.

The soundtrack picks up the pace. The images change quicker now.

Julie and me seeing Ash off on her first day of school.

The three of us at lunch at a friend’s house.

Julie in a long red dress, her hair up.

Now a soccer match.

Now a graduation ceremony.

Now another wedding.

Julie. Me. Ash.

A whole life.

Oh, my girls.

Oh, my Julie.

The slideshow ends with Julie and I sitting together in the shade on a stone bench. It’s one of the last photos I have of us together. She’s wearing a wide brimmed hat and has a thick blanket over her legs. Ash is behind the camera while Julie’s wheelchair is just outside the frame, as are the tents of the makeshift neuroplague palliative care center.

I wonder about the machine learning algorithms that drove the Fotobot to compile this slideshow. I wonder about these algorithm’s practical limits, and the inadvertent algorithmic cruelty of those limits.

They can detect the smiles in the two faces sitting quietly together on a late autumn afternoon.

But, can they detect that I was trying to smile for her?

Can they detect how desperately I was trying to smile for her?

The music resolves to silence. The image of us on the bench blurs and the Fotobot icon appears, along with a message inviting me to “Start over from the beginning?”

I choose “Yes.”

Blaize Kaye‘s writing has appeared in Nature, Fantastic Stories, and New Contrast, among other fine venues.

He has been shortlisted for the Nommo award for best African SF short story, as well as the Short Story Day Africa prize.

Leave a Reply

Your email address will not be published. Required fields are marked *