Were you ever a science fiction reader? As a kid I devoured sci-fi, rarely of the laser beams and alien sort, but usually the strange and societal challenging type, like Clarke, Bradbury, Gibson, Brin, Card, and Le Guin. A bit of Asimov and Heinlein because you were supposed to read them, more than anything.
Because I’d read this stuff, I believed the idea of thinking machines was something that could happen, but not necessarily in my lifetime. And when the thinking machines came, they’d be all powerful, and probably in the shape of a human, like robots. They’d be guided by Asimov’s Three Laws of Robotics (“First, A robot may not injure a human being…”) and we’d trust them to manage major parts of our lives and society.
That was kid stuff. Until one day I interviewed for a job at a company where the founders told me they had created an “AI, an algorithm” to produce news that we’d use to fill dozens of newspapers.
Based on what we know now, the claim seems fatuous from the start. I mean, AI is technically an algorithm (like most digital stuff), but today we understand that AI is so unbelievably complicated, to think that some little startup would have created an algorithm to behave as AI is laughable.
Unfortunately, fourteen years ago my understanding of AI and algorithms was poor, informed only by teenaged science fiction consumption.
Despite this, and for many other reasons, I was lured into the job, thinking it would be an adventure. The day I started, I asked for a briefing on how the AI – referred internally as “the algorithm” – worked. The company leaders put me off and put me off. Until finally, someone on the tech team revealed to me that there wasn’t really much of an algorithm, and certainly not an AI. The company’s news reporting was literally powered by hundreds of Filipinos on the other side of the world.
Lots of things unraveled after that, and soon after I quit the job.
The experience made me reflexively suspect of the idea that AI could be truly powerful or replace the considered work of thinking humans. The company I joined and left was not attempting to accelerate human ability, but to replace creative activity wholesale, the actual fatuousness.
I’ve learned to ask not, “who or what is powering this?” but “what exactly is the expected outcome of this system?”
Technology seems to have made our society into a pot of slowly boiling frogs, getting more and more comfortable with the rising heat of the systems used to manage our lives, nudging our thinking one way or another. When are we actually thinking for ourselves?
As information technology has advanced, the way we ask questions has changed. Thirty-five years ago, as a college student, if I wanted to understand a concept, I first had to know its existence before I could even name it.
As a college junior, I was assigned to write a capstone thesis. I knew I was interested in how institutions worked, so I started by pulling drawers of the library card catalog to look up “institutions”. Eventually I stumbled on books about organizations. I checked out a pile of them and spend a weekend reading until I came upon the idea of “organizational theory”, which studies how organizations work, and then I focused again and took out more books. Finally, I read about Ichak Adizes’ idea that organizations naturally move through three phases: Growth, stability, then either ossification or transformation.
Since then, that is the foundational idea for how I work within every organization I’ve ever been in. Once you think about the capabilities of the organization you’re encountering, you change your expectations and how to proceed with the people you work with.
The process of coming to understanding those phases involved deep reading, hours of thinking about examples, and lots of internal debate about how organizations work. Searching card catalogs, reading stacks of books, I had to choice but to be mired in the concept of organisational theory just so I could learn to ask the right question.
The arrival of search changed all that. You could skip the process of searching card catalogs and instead flit through search terms until you stumbled upon the idea that fit. Answers came quicker, but you missed a part for your brain to adapt to the concept and maybe develop your own solutions.
Today, AI searches bring a whole new paradigm. Never mind figuring out the right question. AI will sort through your idiocy and anticipate a solution for you, then produce a bespoke answer pulling in multiple lines of thinking. And, even more incredible, AI will suggest further lines of research (that it could help with) on organizational theory or any other topic you can dream of.
Despite this, I enjoy the slow, plodding work of reading non-fiction. Some of my favorite reading experiences are those that require me to stop every five to ten pages for my brain to cool down and be sure that it really understood everything I just read. It’s not the best way to get to immediate answers but I believe the method forges new mental connections, providing new ways to perceive the world I live in.
As our society moves forward, I expect fewer and fewer will choose this slow method, because knowing rewards you faster than understanding. And it’s hard to know when understanding will ever be important, since if you don’t understand, you can’t know how much you missed.
Still, I’ve never been much of a Luddite. Plunge right in! And let’s find out what technology can bring.
For the last few months I’ve gradually dipped my toe into AI. Last summer, when ChatGPT 4 was “a thing”, I checked it out. It was fun to talk to it but it was hard to see the utility, other than as a toy. So, I shelved it.
But all through this year, my job – the British Foreign Office – had been pushing staff to find ways to implement AI as part of work processes. We’re given a sandboxed version of Microsoft’s Copilot, which means the data we give it is kept only within the Foreign Office. But also, our version of Copilot can’t connect to the internet or to our files, so it is somewhat limited. Still – it’s useful! I started using it to generate summaries of large reports, to produce spreadsheets and analyse data. I’ve saved time.
Still, while that’s interesting, I hadn’t really thought much of AI, until I read Matt Schumer’s February 10 post on X. In case you haven’t heard about it, Schumer’s long screed became something of a phenomenon among tech heads. Here was an AI founder – but not one of the elite – laying out in clear terms how powerful the technology had become.
The models available today are unrecognizable from what existed even six months ago. The debate about whether AI is “really getting better” or “hitting a wall” — which has been going on for over a year — is over. It’s done…
If you haven’t tried AI in the last few months, what exists today would be unrecognizable to you.
That last bit convinced me. I signed up for a Pro version of Anthropic’s Claude Desktop, $20 a month, and started experimenting. The last three weeks have astonished me. Claude Pro is still gated from the web – which means it can’t conduct searches or access files that you don’t explicitly give it access. It also can’t control aspects of your computer unless you give it express control. But it produce code, develop regular processes, and the answers are more than rote, they have an element of taste.
I’ve done some amazing things with Claude, I created a webapp that imports US electricity generator data to map out current and proposed projects, asked it to read all my blog posts and giving me a critique of what I need to do to improve my writing (in great detail!), and I used it to produce transcripts of a set of audio interviews and then create a cut list to edit them into episodes.
Some of these tasks are more complicated than others. And some of them required me to refine the request, or debug the result many times over multiple hours (the webapp took about 10 hours). But the fact is almost all of these things I could not do on my own. In addition, the most powerful results came from analysis of data I provided, like the webapp, resume, blog posts, and audio transcripts.
While AI can answer questions like, “What is organizational theory?” it is at its most powerful when you review existing information or create a system to gather information to be analysed.
While search is a question answering machine, today’s AI is a perspective changing machine. It still lacks creative problem solving. And when you ask it for ideas, it gives you something from the muddling middle, never an explosive or thrilling concept. For that, you need a human.
But still: When used by an inquisitive person, AI gathers and manipulates information in unexpected, unanticipated ways so that human minds can see the world in a new way.
If Steve Jobs thought of computers as bicycles for the mind, AI turns them into motorcycles.
Perhaps you know someone who has a motorcycle in their garage, or maybe just a bicycle, that they never take out. They use it on really sunny days, but they don’t think of it as a tool for transport. But then there’s a guy who uses their bike for delivery, maybe road trips. This second guy understands the power of their motorcycle.
Now, maybe you know someone who has a really jacked up computer. They just use it for Microsoft Word. Or maybe to play Rainbow Six. But they’ve never coded. Never edited video. Never really put the computer through its paces.
I’m learning that AI is going to be the same way: Incredibly powerful for the inquisitive, but of limited utility for lazy thinkers.
It’s this reason that I’m doubtful that AI will become a giant job-killing machine. As much as CEOs imagine their empires as efficiency engines, 90% of the world’s organizations are actually a mirror of human civilization, set up as a series of social interactions to carry out a certain goal. Thus, the famous saying, “Culture eats strategy for breakfast.”
Schumer, in his post, raises the job-killing spectre:
AI isn’t replacing one specific skill. It’s a general substitute for cognitive work. It gets better at everything simultaneously. When factories automated, a displaced worker could retrain as an office worker. When the internet disrupted retail, workers moved into logistics or services. But AI doesn’t leave a convenient gap to move into. Whatever you retrain for, it’s improving at that too.
AI is coming for your white collar job, a claim AI protagonists love, since it allows CEOs to imagine giant productivity gains! And in tech, from what I hear from people who code, it certainly is taking up a huge portion of programming work.
Another example: My wife used to work at Skidmore, Owings & Merrill, one of the biggest architecture firms around. Forty years ago, she says they used to employ hundreds of architects just to make drawings. Whole office floors full of drafting tables and men (always white men!) in starched shirts. And then AutoCAD came. Soon after, Skidmore started letting people go by the hundreds.
I think if your white collar job is largely technical – you’re compiling data to put into the TPS report every Monday – then you’re in trouble. But many white collar jobs (like mine!) are mostly about people and convincing people to do things. Until AI can understand cultural social cues, I’m not sure it’ll replace all those humans.
AI’s motorcycle of the mind provides you with power to create new systems of data gathering, manipulation, and management. For instance, have you ever had an idea for an app? Now you can use AI to make one! Maybe not Microsoft Word, but maybe a webapp that helps you manage your car rental business. Just tell it what you want! The real challenge with AI now is threefold: How creative an idea can you dream up? How do you ask the right question? How do you get the data? If you solve those three challenges, then AI can help you do just about anything that doesn’t involve physical labor – or maybe if you ask the right questions, it can!
Really, the challenge is to come up with a good enough place to take your AI motorcycle. What kind of app do I want to make? And can I do it before a thousand other guys think of the same thing?
AI isn’t going to unleash a wave of economy-destabilizing white collar layoffs. While some people will lose their jobs, white collar workers have endured because of their social fealty to the bosses. Instead, we should expect AI to empower a new wave of crap and a new type of institutional enshittification we never imagined.
Instead of a wave of job destruction, I predict a future of endless people coming up with endless solutions for every problem. Soon we will be awash in programs, apps, algorithms. Since the first dot-com boom every entrepreneur with an idea was limited by their ability to find a tech co-founder and the techie’s ability to produce the vision. Then, the development process was limited by the speed at which coders were able to iterate, which could take months. Now, we are closing in on a world where a lone entrepreneur with an idea can develop and iterate a solution within days.
Zoom, goes the motorcycle.
AI is going to hypercharge the speed of institutional change. We won’t get robots, but our lived experience will be a lot weirder than anything science fiction ever dreamed up.
What also I’ve been thinking about…
Big for-profit news sites are looking wobbly. As an alternative, here’s a great list of indie news pubs. Also, a longer list of local pubs – maybe one in your town.
Practical typography is guaranteed to up your design game and make your documents and website better.
Luke Igel and Riley Walz made jmail.com, a site that turns Jeffrey Epstein’s emails into a gMail account. It’s fun!
Four years ago Senegal built a commuter rail line for Dakar. It’s a huge success. For me, this is less of a “how cool!” than “why the fuck not for every developing city?”
Meanwhile, in East Africa, Kenya is launching a program to promote EV car and scooter purchases because locally produced electricity is cheaper than gasoline and buying foreign petrol creates a currency balance of payments problem. Expect to see more of this around the world.
The Wall Street Journal story on Kristi Noem at Homeland Security is even more bonkers than you thought. (cached version)