Through most of history and prehistory, daily existence required sweating. Humans mainly got to their fields or their shops on their own two legs. Over time we incorporated animals, then motors and big trains to move goods and each other, further and faster. More recently, we developed the technology to see and talk to someone else anywhere else in the planet at the speed of light traveling through a massive system of fiberoptic cables. Companies exist now without anyone ever meeting face to face. A rumor can start anywhere and spread to half the world in a matter of hours. We have solved the problem of distance.
Except as we have increased the speed of seemingly everything we could need, one speed has not changed in all of human history - the speed of our legs. I'm not talking about Usain Bolt's 100m record but rather our quotidian walking for sustenance. Even as our society has adapted to cars and the internet, our neighborhoods still have constraints related to the distance we are willing to walk. Sure, people drive all the time to reach great distances, but people don't live in cars. Bottlenecks with traffic and parking hinder car-driven societies and thus have not eliminated the value of walking. Though our society has evolved dramatically to incorporate new technologies, we cannot evolve away from the necessities of walking.
As diehard an urbanist as I am, this blog post is not about urban design. It's October 2026, and I am but a speck of dust in the pile of others blogging their thoughts on artificial intelligence. Given the nature of my work as a data scientist at a technology startup, AI has entirely upended my day-to-day work experience. I've had to constantly rewire my instincts and think, "would AI do this task more efficiently?" With the prospect of needing to compete with other businesses, I cannot afford to not use AI. As Claude code emerged this year as a better programmer than me, gradually my life has evolved to one where most of my day is spent waiting for AI to work and reading its output.
This shift has been exhausting. I have found mornings where I have begun work for hours and not fully woken up. Many times where I was posed with a problem that would have required me to think and research, I have instead redirected the task to Claude. Often its response is perfectly adequate and I'm able to continue my day without having activated the deeper neurons of my brain. Instead of writing, which requires active brain churn, I am reactively reading. The number of Pull Request reviews I am assigned has increased exponentially. Even worse, I am doing the same with my robot, and I'm constantly reading tons of generic, overly verbose code. Reading on a not-fully-activated frontal cortex is the source of the exhaustion, and eventually I realize I have to figure some way to snap out of the brain fog. As great portions of our code base become AI-written, I feel my grip on its structures and linkages slipping away.
I have enjoyed AI the most for my side hobbies and non-work training. Recently I was curious to understand OpenAI's proof of the Navier-Stokes existence problem. 15 years removed from my math master's program, I am not sharp on reading mathematics papers. But I had some exposure to Navier-Stokes equations previously via computational fluid dynamics, and by using AI tools to explain the problem and solution to me, I was able to understand the issue well enough within a week to present on it to my colleagues. I spent 15 minutes or so several nights during the week breaking down components of the equation and proof. I'd think about the problem during the day (if the prof says the equation breaks down, what does that mean physically?) and come back with followup questions. On my own, I would have spent much time figuring out even what to read. At first, I went to the Wikipedia article for this topic, but at first pass it was quite intimidating and I didn't have the foundations to make much sense of anything. Using AI reduced so much friction by adjusting to my initial level, and then scaling up as I wanted to know more.
AI accelerated my learning by generating 500 words on Navier-Stokes in a few seconds, but I had to do the hard work by reading and thinking at a steady pace. My brain is limited both in speed (https://pubmed.ncbi.nlm.nih.gov/39694032/ claims we can process ~ 10 bits/second) and in storage. I can't read that fast and I hit fatigue. Thus I was only able to get to my understanding of the complicated Millennium Problem over multiple days. AI is able to consolidate the information of the known internet incredibly quickly and effectively, but we are constrained in our ability to absorb this information by biology. Our brains are now the bottleneck, and technology has not meaningfully improved this.
Well what's wrong with that? We can still use AI to supercharge what we feed our brain like I just described. The problem is that this is a choice, and we are primed to choose not to do that. It's effortful to use our brain and deeply engage, and very easy to use AI to put stuff out there. As we've seen with many tech products, our default usage is impulsive, dopamine-feeding and short-term oriented. Tech companies saw this in their user data and optimized their products around it. Left to themselves, they will naturally cater to the lowest common denominator, and we will get products that enable us to slop their way through the day. We will get the digital equivalent of cookie-cutter suburbs full of dazed drivers.
Left to the free markets, suburbs have become the natural outcome. It is very easy for private developers to buy land and build houses, it's very hard for them to create their own public transportation network. Successfully orienting development around transit requires top level government planning. It requires well thought-out subway route planning, massive infrastructure buildout, prudent mixed-used zoning, and scale. Done well, it enables us to do more with our days.
It is similar for AI. Government can play a major role in imagining a prosperous AI-enabled world, and to implement the policies that encourage that. One lesson that I've learned from using tech over the last decade is that while usage and addiction may feel like a personal problem, it's very much influenced, even manipulated, by product design. It's the same with AI products - imagine if there wasn't a button to copy your output from Claude? A more friction-inducing interface would reduce slop and encourage us to slow down. Additionally, governments can vet models for critical thinking explanations and ask that counter arguments prevent too much affirmation bias.
However, without taking accountability away from the big tech builders, there are personal habits we can install. We can choose to center our brain in our workflow, to use AI less for its output and more as a learning tool. It might seem antiquated or even wasteful in the age of effiency-maxxing to do tasks "manually", but sometimes it's worth the mental engagement. AI output might always be faster in the short-term, but actual knowledge gain pays off in the long-term.
In this post, I've consciously resisted using AI, even as my mind wonders, in this extremely meta moment, if I could get it to rephrase this awkward sentence and cut out some commas. Partly I wanted to challenge myself, to keep my writing skills sharp. It's like intentionally walking to the store after driving around all day. The output, this blog post, isn't nearly as valuable as the exercise of fleshing out this analogy and thinking through how I really feel. It's taken me days, but I think it's been worth it.
No comments:
Post a Comment