At one of my workplaces, I was required to install an application called Timedoctor. This application tracked all my behavior on my laptop in real-time: it could sense whether I was active or idle, what programs I was using, what file or tab I had open. While it’s presented as a way to ensure employee productivity and accurate track working time for paychecks, it felt really dystopian and invasive. I remember spending so much energy stressing about appearing productive by switching tabs and clicking random things that I didn’t even end up being any more productive than without it.

When it comes to digital privacy, especially concerning more serious matters like crime, some may argue: if you’re not doing anything wrong, you have no reason to mind being watched. I disagree to an extent. Privacy is essential to a feeling of comfort and security. Invading it for the sake of preventing “bad” behaviours will only discomfort everyone, and those who intended to commit such acts will find ways to circumvent it anyway.

Even now, the line between too much and too little surveillance continues to shift. Facial expression trackers in public spaces, mouse movement trackers, eye trackers, etc.—these are all existing inventions currently in use. While they have supposed benefits for worker productivity, I argue that they contribute to the toxic “hustle culture” aspect of capitalism. They’re more efficient at making people put on a facade of ideal behavior than addressing root issues causing misbehavior. In my personal experience, overwork and the resulting exhaustion are the biggest obstacles to my productivity. Forcing myself awake during work hours only temporarily addresses the issue, and exacerbates it in the long term.

I wanted to raise the question: what are the consequences of these new AI/ML technologies being made—who are they really benefitting? When they solve problems, are they treating symptoms or the cause—or even contributing to them?

Then, I asked: how can I build an application using AI/ML technology that makes things…objectively worse?

‍And thus, Snoozer was born.

Version 1: Teachable Machine and Glitch

I took photos of myself in various "asleep" and "awake" states and uploaded them to Teachable Machine. Given the scope of the project, I was unable to amass a giant dataset of photos of faces and label them all. Instead, I curated a very clean dataset of my own facial expressions. While it was functional in certain contexts, it was not universal enough for practical use.

I deployed this version on Glitch, using p5.js for my webcam processing and interface. This setup was unable to be overlayed on other websites, however.

I developed a popup for the Chrome Extension that would inject the Snoozer interface as an element directly within the current webpage's DOM. I wanted a persistent overlay with a custom shape, not a popup that would close when clicked out of, and I wanted the overlay draggable around the page.

This was the final development structure and UX flow to accommodate Snoozer's features. A major challenge was programming tab detection to automatically activate/deactivate the Snoozer interface when tabs were switched. I used the Chrome Tabs API and a background script to handle listening to tab events.