Making Algorithmic Art with ChatGPT

On progress

We're living in an interesting time in history regarding our society's personal relationship with work. Folks have feared automation for generations, and we have forever functioned with the understanding that the more uniquely human a job is, the safer it was. Fine manual labor graduated from massive industrial bots cutting and welding steel, to the nimblest of delta arm and pick and place technologies assembling ever shrinking SMD components. Each of these is faster than any ten salary and benefit-requiring humans for any appropriate task. But, we've never viewed the capacity to grunt, sweat, and be bored as uniquely human, nor do most envy these lost opportunities - so it goes.

Numeric intuition was the next ability to fall. Where previous innovations simply augmented the capacity of humans to shuffle numbers for profit, the competition was not fierce, nor the consumers sophisticated. If you could make the line go up on Excel, your life was probably quite good for a time. While human creativity, intuition, and pattern recognition are beautiful in their own right, they do not scale. Starting ten years ago my friends in finance inexorably began to retrain or lose their jobs if they did not function in chiefly charismatic or legal capacities. You cannot broaden the span of your quantitative vision for a singular task beyond the reach of a CUDA-enabled A100 GPU running the latest dissertation-cum-high-frequency-trading algorithm. That GPU will never ask for a raise, a promotion, or PTO, and you can fill entire warehouses with them. Still, charm and argument alike remained uniquely human for a time, so they were.

With each generational re-appropriation of labour there have been reactions and there have been Luddites. Techies and historians alike defend this shift by parroting the argument that technology has no True morality. As no one has yet patented kitten-soul-powered kitten-seeking-drones, we can certainly maintain hope for a while longer.

On chat bots

Today though, we have ChatGPT and several closely related LLM APIs that may function as inconsistently reliable oracles, ready and willing to answer any question 10 times a second at a nominal cost. These algorithms are as charming, argumentative, or as hallucinatory as you like. All they require is for you to tune your parameters appropriately and ask them in the right way. They are not particularly unique in construction, the technology has been open-sourced for a while. Your phone uses much the same concept to suggest the next word in every text you send to a loved one. What's different is the scope, the scope is insane, and the emergent behavior and intuition it contains is far greater than the vision of singular humans on countless tasks.

As with every once-human talent bourn from a machine, there will be layoffs and instability in the short term. The least moral practitioners of technology are often the most enthusiastic. If you've noticed your search results reduced to nonsense over the past few months, it's because of a raft of data practitioners like me gold-rushing to devour the first page of Google with banal, procedurally generated, ad-enabled content on every topic you might possibly dream about. The Turing test is dead and the bots are cheap - we are presently sleepwalking over the (final) trust-cliff of the internet.

On utility

Right, so that's the bad things. Lets look at what's good, or more gooder at least? ChatGPT and it's ilk share one trait that none of these prior innovations truly possessed: access. You do not need massive equipment capital nor a heady investment in technical education to use them. For the time being they are simply there for everyone regardless of age, language, education, or wealth.

The layoffs will be temporary as well. Sure, the average worker puts in a solid 14 productive hours per week, and most every trial of the four day work week and UBI has been an amazing success . Nevertheless, America remains smitten with its derriere in chair time - our cult of presenteeism. With the aid of technology, the productivity of the average American worker has skyrocketed for generations. Rather than liberating ourselves to more freely experience our moment of kinship with the universe, we have simply invented more work that needs doing. There is always something more to be produced, always something more to extract, to refine, and for as long as these views hold there will always be more work to be done. This new technology presents but another chapter in American workers yielding far more productivity for the same old price , and the jobs will come back in sorts as old habits take hold.

But lets go back to that point on access. There's an idiom I've long followed that any new technology that is good enough to use today should be considered an immediate replacement for anything that preceded it. Barring the early hype-cycle, each technology should only get better, cheaper, and more accessible as time goes on. ChatGPT is good enough for everyone to use today, and you should be using it. You wouldn't brand yourself an incapable cook for the inability to strike and bellow a wooden fire. Neither does the presence nor absence of a bike define your ability to ride a bike. If you can write python, perform computer vision tasks, or craft SQL queries using ChatGPT, then you can do those things, and that is sufficient. Either that, or your competition will first, and so we go. We are essentially living through the invention of the steel plow, looking on with our muddied hands and questioning where it will take us.

On art

I've been mulling over this new capacity to perform now, and what is either possible or easy today that was neither yesterday. I'd long wanted to code a system to capture and 3D print significant locations from my life. Unfortunately, I just never had the time nor the money to make it happen as a graduate student. So, I asked ChatGPT bluntly, how do I do this thing with python. We agreed that OSMNX was the way to pull building and road geometries, that lined up with my past dabbling. The terrain though, was a mess. We spent hours chasing dependency rabbit-holes over deprecated queries and refactored APIs. Four hours in, I visited our old friend Google and quickly found some snippets to extract the satellite data as a raster which I could align and interpolate adequately. With that, a couple more trips to StackOverflow and ChatGPT for those matplotlib functions I'm always forgetting, and my png image a la digital elevation model (DEM) was ready, I just needed to turn it into polygons:

I took a moonshot and simply asked ChatGPT to do the work, and it almost did. I passed along the off-by-one error code, received a short apology and a corrected snippet, and I had a working 3D mesh of one of my favorite places a minute later. It wasn't printable yet, DEMs are made for hydrology wonks, not sentimental techies with specialized equipment. So, I asked ChatGPT to fill in the bottom gap with a flat plane. It understood my desire for a flat plane, but it mirrored the 1000x1000x2 polygons capturing my memories above flatly down below, in turn exceeding the software's memory limits. A quick scolding and a few more tries later and it reached the limit of its capabilities on this task: two overlain triangles spanning 50% of the bottom plane which I had to manually code an adjustment for.

So, is this art, and was I an artist? I know my way around a pen and a brush, and I captured the original vision of the design. The code took a blend of experience and intuition to yield aesthetically pleasing results. However, the most challenging albeit dull task was executed by an inhuman intelligence of sorts. Meanwhile, my friends who sacrificed everything for a life of expression are seeing their greatest works stolen, chewed up, and spit out by paid apps and crude queries for social media cred. They are certainly artists, and they are suffering for it.

Today, we can do unimaginable things with once unimaginable tools, but what does that make us? I certainly remember how to ride a bike, and a potter is still a potter if a wheel turns their clay, but I don't think my new tchotchke can answer any of these questions for me: