In the AI of the Beholder

by Soramimi Hanarejima

​

It’s just another one of those mid-afternoon lulls in the office when your beauty tracker app pops up a peripheral-vision notification informing you that you’ve only gotten 2 instances of beauty so far today. Puzzling, since you know there have been more than that. The jaunty golden retriever during your walk to the station; then in the subway car, that cute kid flashing a smile wide with delight and the open sweep of a teenager’s hand welcoming a gray-haired lady to the seat he had just vacated; the gorgeous billboard promoting tourism to Arcadia, its swath of lush meadow before a granite mountain range reigniting your desire to visit that distant land; the clump of furry moss vividly verdant upon the maroon of the brick wall you walked by on the way back from lunch. Maybe you didn’t look at these things long enough for your smartglasses’s sensors to register them. Or they were too far away. The kid was half a train car away from you.

​

You dismiss the notification and head out for a break, to find some beauty that will get you closer to the daily beauty count goal of 15, which is supposed to lock in health benefits of parasympathetic nervous system activation. To stack the odds in your favor, you visit the art museum a few blocks away.

​

As you wander familiar galleries with wooden floors that creak under every step, you make sure to look intently at the canvases with swirls of color and patterns of texture that delight you, giving the smartglasses ample time to pick up these works of art. This prolonged observation is especially easy to do while you’re in Quasars of the Mind, the special exhibit featuring a new series of massive pieces by FJ Trítrí. Each enthralls you with its depiction of a cognitive maelstrom churning luminous ideas, reconfiguring numerous constellations of concepts. After just half an hour in the museum, you’re confident you have a shot at today’s leaderboard for your department—if not the entire company.

​

On your way back to the office, you pull up your quantiselfie dashboard and find that your step count is now over 7000 yet your beauty count is only up to 5.

​

How is that possible? your thoughts balk. A bug?

​

You check for software updates and find none.

​

Is the machine vision hardware malfunctioning?

​

You run LookNLearn, and within moments of booting up, this AR language learning app has correctly labeled the elements of the sidewalk scene before you; the people, trees, bus stops and buildings in your field of view all have the correct Arcadian logograms hovering beside them.

​

Shortly after returning to the office, you take to your usual stairwell to make a customer support call.

​

“I’ve looked at twenty beautiful things today—at least. Only five were logged,” you explain after the agent’s boilerplate greeting.

​

“And you’re sure the app has been running continuously in foreground or background mode?” the agent asks in a tone both firm and warm that neatly avoids sounding patronizing.

​

“Quite sure.”

​

“Then you were probably looking at things that aren’t canonically beautiful—aren’t conventionally considered beautiful. The app has difficulty recognizing those sorts of cases.”

​

“That means the app isn’t doing its job properly. It’s not counting my experiences of beauty.”

​

“Yes, we’re sorry to hear that. The situation can be remedied by enabling training mode in the settings. Then you can manually log things as beautiful with a double tap on the left arm of your glasses. This will improve the scoring algorithm so the app can better identify things you feel to be beautiful. The data will also be sent to us for the purposes of making improvements to the app.”

​

“Great, I’ll give training mode a try.”

​

“Sounds good. Is there anything else I can assist you with today?”

​

“No, that’s it for the moment. Thanks.”

​

“Sure thing. Thanks for using BeauTrackUp, and have a great day!”

​

You linger in the stairwell, wondering if the quality of the call is being scored by some algorithm. If so, will it classify the interaction as smooth, effective or professional? Will this algorithm also be improved by comparisons of its evaluations with those made by humans?

​

When you get the customer service survey that will inevitably show up in your inbox soon, you will rate this call a 5 out of 5—regardless of whether you were talking with a person or bot. Now you can own the leaderboard.