Why the AI can read your mind but nothing works

A day does not go by without breathless headlines about advances in AI, chatbots, machine learning, or related technologies. Thanks to the so-called internet-of-things, your iPhone can sync with your dishwasher, and thanks to Neuralink, in the not-too-distant future will sync with your mind. You can summon a chatbot to solve business problems, or use Microsoft Copilot to help you with a coding project. GPT-4 can produce prose of a high enough quality that it threatens to make make journalists obsolete.

Every major tech company such as Alphabet, Apple, and Meta are in an arms race to launch the next great AI tool that will radically change and disrupt business in ways never before thought possible and unlocking untold windfalls of economic productivity in the process. Everyday hundreds of articles are uploaded to arXiv about progress in this field. It’s dizzying trying to keep up with it all.

Yet this is juxtaposed with an epidemic of Enshittification, “a term coined by writer Cory Doctorow in November 2022 to describe a pattern of decreasing quality observed in online services and products,” but the concept is just as applicable to the offline world, too. It’s not your imagination–in spite of all this progress in AI and other technologies–many things still seem shitty or do not work, whether it’s new appliances that break easily or USB connectors that fail, particularly USB-C cables, which thanks to the short connection profile are prone to breaking compared to the more sturdy older USB cables. Why can’t some of this brainpower that is being used to make better chatbots or device-syncing also be summoned to making products that are more reliable or better?

Like, in spite of increasingly advanced language models, why are we still stuck with leaky and useless cup lids, condiment packets that squirt on your clothes , or wobbly tables and chairs that need wadded-up napkins to stabilize. Or plastic utensils that are liable to snapping at even the slightest overapplication of force. Yes, in spite of GPT-4, the only solution to wobbly furniture is napkins, not better furniture. Why do stamps still exist? Why cannot anyone disrupt those industries? Why can’t the big brains in tech make better condiment packets or tables that do not wobble? Compared to seemingly quotidian developments in AI, the most notable recent innovation out of the condiments industry in decades, AFIK, are those ketchup bottles with the large flat tops, so the bottles can be stood upside-down to allow for the ketchup to collect without having to shake the bottle. Yes, this was a big enough deal that people are still talking about it.

Part of the problem is simple problems are not uncommonly surprisingly hard to fix. For developers, what is supposed to be a ‘simple coding problem’ can sometimes turn into a multi-hour slog–ask anyone who has tried to align CSS tables to work perfectly on all browsers. Regarding wobbly tables, it does not help that people are lazy: when they stand up or sit down, they lean or press their often obese or overweight bodies on the table to get leverage, causing the weaker parts to slightly warp or deform due to metal fatigue over many years by thousands of people. Even if the table is structurally strong, the hinges and soldering are the weak points. Trying to overengineer a table is not economically viable.

Companies make more money by consumers having to frequently replace breakage-prone products after some predictable amount of time has elapsed, known as planned obsolescence. If durable goods cost half as much but break three times as often, this is like another hidden tax or hidden form of inflation that may be overlooked by the ‘official’ CPI stats. It’s not an accident your stuff breaks: it’s a message that it’s time to replace your refrigerator with one that has an even bigger ice cube maker, or a phone with a slightly higher resolution camera and a few extra gigabytes of storage. Same for mandatory recurring subscriptions, such as a $20/month fee so your smart appliance can ‘locate itself’.

Also, blame economic incentives and consumer preferences for the persistence of crappy old technology. Door technology largely has not changed in a hundred years, in large part because there is little money in making better doors, and consumers don’t care that much either. The coffee shops I patronize are still full of people even with wobbly tables and chairs. Consumer preferences tend to be inelastic for small things, unlike agonizing over college choices or comparison shopping for a new car. Unlike AI, no company or VC firm is going to get a trillion-dollar valuation or a lucrative exit by designing better ketchup packets or thumbprint or iris-activated doors like in science fiction. Here’s to our AI-enhanced but still breakage-prone and wobbly-furniture future.