The desktop is missing crucial quality-of-life features

It was after I asked myself “why does setting the second bit of the keyboard controller’s port to high enable 21-bit physical addressability in x86 real mode?” that I realized my academic education had sunk too deep into the details. I know we need the nitty-gritty architects to construct a solid foundation of knowledge and best practices, but more so we need fresh dreamers to drive innovation beyond the conventional boundaries and find better ways to solve old problems. It’s the reason Nintendo banked on Miyamoto in 1977, a complete outsider to the gaming industry, and now the world has Donkey Kong, Mario, Zelda, and Pikmin.

Perhaps the most conventional computing field is systems, where its historical roots yielded ubiquitous, Unixy solutions that perservered today. Resource constraints explain the uniformity in low-level design in the kernel, but middleware is ripe for innovation. For example, the Wayland protocol is finally making tides in window compositing, increasing both isolation and protection. But I’m thinking much bigger than that and with the end user in mind, like taking advantage of spiffy intelligence systems to solve problems. So I thought to myself, “what do I dream for my laptop to do for me?” After some long sci-fi deliberations, I landed on a list of realistic quality-of-life improvements. But let’s talk about some of those deliberations.

To start, I believe people’s routines are very predictable, especially compared to predicting whether Schodinger’s cat is still alive or whether Overwatch 2 will ever have PvE (the answer to both is yes, but no). Everyone’s daily problems can be accurately summarized as the following:

These features are already achievable with today’s AI models of varying complexity, and I feel like they would significantly benefit a majority of users by making our software fit into our lives seamlessly. Now I am aware of the countless third party solutions, but after the initial excitement wears off, they often become background annoyances that you try to force into your routine; software should be simple enough that you don’t even know its there, or at least that its not a default. Google can stop marketing that they value our privacy and don’t harvest every piece of data imaginable and sell said data to any interested company. They could be doing so much more to make our lives seamless, so why do we not have space age prediction systems yet? Let’s investigate a few reasons.

• Complexity of tasks: Some tasks are simple to compute (int am_hungry() { return 1; }), while others are extremely difficult. In many applications, having a model that guesses what the user wants with acceptable accuracy, with low enough latency that the user doesn’t notice, and with minimal input from the user, is like wielding the whole Triforce; be ready for some compromise, unless you’re Ganon at the end of A Link to the Past.
• Privacy concerns: The truth is there are two types of people in the world: the ones who deep down are ok with the fact that the Chinese Communist Party knows about their new Pilates routine as long as they can use their social media, and the small minority who use Tor browser for things that aren’t video game piracy or hard drugs. Maybe don’t pretend to care about Big Data while also publishing your life.
• Hardware constraints: Every computer is in fact not running an Nvidia GPU and thus cannot parallelize with CUDA. But there is a silver lining… Even though the “in” thing in AI is to use unimaginably large transformers that eat 16GB memory for breakfast, there are now itsy bitsy “edge-computing friendly” architectures that use clever optimizations and pruning to run on even your decades old ThinkPad you use as a home computer.
• Diverse user needs and environments: The way I use my computer is indeed far removed from the average user. But many problems are still universal to everyone who owns a computer. And we can have different models that are tailored to different groups. We do not need a one-size-fits-all solution for most things.
• Economic considerations: Ahhhh now here’s the kicker. The problems I’m describing are probably too niche for anyone to make money off of them.

So while keeping in mind that I know nothing about privacy research, I feels like most people would prefer convenience over privacy after the initial creepiness wears off. Especially because the technical issues are subduing in the face of innovation in computer hardware and data science. I’m very interested in what professionals in this field think, about whether the productivity boost of intelligent software outweighs the long-term risks of automating our lives and potentially becoming dependent on big companies. Personally, I believe an open source solution sidesteps these risks, but at the cost of slower development and adoption. I would have no problem using OSS that implements decentralized machine learning, like federated learning, which I could easily configure to my liking. But why don’t we start smaller for now and make some quality-of-life improvements that could realistically help everyone?

How about the fact that my OLED laptop screen can display true blacks, but a sudden white image at night feels like watching a supernova with front row tickets. As display technologies converge to the infinite-contrast screen I’m looking at, the battle between dark mode and light mode users will only get more violent. Though my screen can move between complete darkness and lightness at 1/60th of a second like an Olympic sprinter, my eyes can only do so slowly like a lethargic, obese dog. Or how about the fact that battery doesn’t even last till the end of the day because of my power-hungry 4k OLED screen that is only used to render a terminal emulator and ChatGPT session, like using a Japanese chef’s knife to open a bag of cheetos. Alas, no one will make money off of software that dims your screen when you’re not around so the void can continue looking at your ostensibly white websites, but while saving battery. So until the government gives everyone a UBI and programmers can live happy Marxist lives under the tutelage of Richard Stallman, I will continue to use TLP and Dark Reader in the hopes that I can make it to the nearest charger before my poor XPS rests in the spirit world (seriously though, Dark Reader is an amazing plugin).

But somehow, I am glad that somewhere in the world, AI researchers are investigating topological neural networks and their implications on nudging me to purchase one brand of wireless earbuds over another. But Sony’s market capitalization will not capitulate to Bose if a handful of AI engineers solve my less profitable problems instead of making bank and taking vacations to Cancun where they snidely remark about the importance of their work while eating caviar out of gold chalices. I may be using a pair of QuietComfort Ultras instead of WF-1000XM5, but the world will not be reduced to ashes. Although my laptop battery may last a couple hours longer…