It’s the 182nd day of the year. We’re at the half-way point in 2026. When Ian and I had this crazy idea to make a map a day using AI as an experiment in “prompt cartography” we thought we might run out of steam after a few weeks but here we are. Prompt cartography is essentially setting a tool (in this case an AI agent) to work using natural language via a Large Language Model (LLM). Between us we’ve made 182 maps (actually a few more given the extras every now and again). We’re still going. But the half-way stage is usually a time to pause, take stock and reassess. It’s a time to reflect on opportunities and to see whether what you started out doing is in any way coming to fruition.
Well, we’ve faced a fair bit of criticism and some vitriol for our efforts. Quite often this comes from people who seem to be lashing out against our perceived selling-out. How can we be cartographers if we peddle “AI slop” is the accusation, and the maps we’ve made are dogsh*t”. We could have just carried on making maps using GIS software, or code or whatever, but we’re both intrigued by new things, and especially those that have the potential to impact our work. What is the potential of this new technology? Is it going to have as profound an impact as the printing press…the computer…the internet, on cartography? The jury is still out of course, but unless you explore and see what opportunities are possible we’ll never know and as Ferris once said “Life moves pretty fast. If you don’t stop and look around once in a while, you could miss it.”
Cartography has always been an art, science AND technology and during my professional career I’ve seen very little movement in the art and science but massive development in the technology. I trained using pens and scribing tools. I used darkrooms to build composite images on photographic film. I had to rapidly pivot to desktop publishing, then GIS. I learned programming with a Commodore 64, then Turbo Pascal and moved onto scripting languages. I went from Paint to Freehand to Illustrator. I’ve dabbled with Natural Scene Designer. I’ve made web maps. But in all of that not a lot has changed in terms of the theory, or the concepts. It’s merely the way in which I’ve expended elbow grease, or, more accurately, how I’ve used technology to do a lot of the heavy lifting in my craft. And I’ve noticed with many cartographers over the decades that they’re adverse to change. And yet they are prepared to have shifts in technology kill their career if they refuse to re-tool. So, getting ahead of the curve is always a useful approach when technology brings change to your work.
Well, rather than being scared of what’s coming, or trying to ignore it, or feeling threatened, we are stopping to take a look around. And it’s been quite eye-opening. Even in the first half of the year various AI platforms have developed remarkably. Labelling on maps (images mostly) was atrocious at the start of the year. It’s by no means the finished article but improvements are impressive. The ability for AI to collate, curate, and organise data is incredible. It is taking away a major pain-point in cartographic work. And while, stylistically, there’s more to be done, many maps we’ve created are pretty pleasing on the eye.
Do they contain errors? YES! And we’d be the first to admit that. But telling us the maps are crap is missing the point. We are not trying to create finished, finessed products. We’re trying to see what AI can help us make as a first pass, with very minimal (if any) editing. Tell me a map anyone creates which is correct first time round. They don’t exist and anyone who tells you they do is lying. Maps take ages to make to be as close to perfect as we can make them. But we’d challenge anyone to make the maps we’ve made in the same amount of time and get to where we’ve gotten. Imagine if the first pass maps we’ve created were fully editable. Maybe all we’d need is some tweaking of labels and it’d be good to go. Are we that far away from fully editable maps? Maybe we’ll see that by the end of the year.
We’ve also faced criticism from those who see AI as simply plagiarism. Well, everything people do is informed by everything else. We read, we learn, we put into practice, we repeat etc. It’s exactly what an LLM is doing. It is a tool we harness to deliver something we ask it to do based on the knowledge it has accrued. It doesn’t scrape the internet. It has literally read and assimilated everything published by humankind. It’s a different mode of production for sure, and learning how to converse with it in a meaningful way using proper language is vital but it isn’t replacing creativity, it’s harnessing it.
Some have demanded we stamp our maps with a ‘Made with AI’ sticker presumably to let them know it was built using methods they disapprove of, and to give them an easy way to explain away any error or problem they see with it. Why did these claims never get made for other technology? It’s as if people view the use of AI as somehow sinister, and that it is deliberately out to catch you out. I would maintain the truth in a map is still in the hands of the cartographer. But whereas expertise was once in handling a pen with skill, not it’s in handling language as a way to control the output. Yes, we’re having to re-learn sentence structure, grammar, and conversation all over again. We’re having to be explicit, and avoid the potential for misunderstanding and mis-communications. Being precise with words helps the quality enormously. So no, we’ve not branded the maps as if they’re already apologising. because no-one ever branded their maps with a sticker that explained the technology they used.
Which brings me onto the accusation that AI is unethical, and that Ian and I are unethical in what we are doing. Seriously? Tools are only unethical in the way in which they are used. And there is nothing about our intent that is unethical. We’ve been very open, and clear about our intent with this experiment. Why is making maps with words worse and a disgrace than making them with code, or a GUI? It’s all just automation of the mechanical means of committing an idea to paper, or screen. Cartographers have been doing this for centuries in different ways. There is no cartographic oath that we took that upholds the use of certain traditions or methods of production, and to hell with progress (whether we like the progress we’re seeing or not is largely immaterial). If that were the case we’d still all be outsourcing our work to copperplate engravers…but wait….COPPER – too valuable to waste on map-making surely?
And what of the environmental cost? Well yes, it’s not particularly environmentally friendly. But then again, not much is. A lot of new and emerging technology has huge initial costs of some form or another. It’s not to belittle that impact but plenty of other methods of making maps have impacts too, and we’re fairly convinced the impact of LLMs and AI agents will dramatically improve over time. Sure, it’s something to keep in mind, but is it really the stick you want to beat it with? Maybe look at everything else in the world around you.
Perhaps those who we’ve seen who seem most annoyed are those that have relatively recently ‘found’ cartography. They’re predominantly computer scientists who found out that maps were driven by numbers and their ability to code interactive maps made them pretty useful in the age of the internet. But vibe coding, and using AI to actually write code is also making them redundant, and they’re getting annoyed. It turns out that being creative is of more value than simply being able to code because creativity is much harder to replace.
And what of academia in all of this? It’s widely known I spent 20+ years in academia developing and running successful GIS programmes but I was recently pondering how technology has shifted the practice of teaching students the ‘how-to’ of map design, or GIS more generally. And I thought of it like this…when I did my undergraduate degree we took a course called computer assisted cartography and we spent 10 weeks (yes, weeks) writing code (using GIMMS if anyone can remember that – fun!). 11 pages of code and I created ONE single choropleth map. For our final year project we each took two whole semesters to make ONE page in a published atlas. When i started teaching a few years later I got a similar computer assisted cartography class down to maybe 3 weeks worth of effort required by the students. By the time I was considering leaving the world of academia that same map could easily be made in a 1-hour student practical. Sure, the concepts still took time but the compression of time the advances in technology brought meant iteration and the undo button were as important when experimenting.
So what if I were still teaching? Well, I’m pretty convinced a student sat in a lecture could have completed the practical assignment using natural language before I’ve even finished the lecture. And yes, we can discuss ‘learning’ etc but purely in practical terms the compression of time has made the ‘doing’ component of our discipline fundamentally different to teach.
So, here we are, still experimenting. We didn’t know we’d be doing this this time last year because life moves pretty fast. But we’ve made 182 or so maps this year already and there’s more to make. Let’s see where we are in another six months time. And if you’re one of those who continually criticises what we’re doing, then I wish you well as technology eventually overtakes your skillset. I hope it’s a hill you’re prepared to professionally die upon. Or, you could learn a little, and give it a go. Contribute to this remarkable change that’s coming at us, and shape our understanding of how to best make use of it. That would be useful.
Plenty more thoughts but please be reminded this is an experiment. We’re going to reflect and put together proper summaries after we complete the maps for the year. In the meantime, we also recorded a special Geomob Podcast with Steven Feldman where we talk about a lot of the background to what we’re doing. Feel free to take a listen: https://thegeomob.com/podcast/episode-340
So how do you map the half way point? I’ve no idea. let’s ask AI.
Original Prompt (ChatGPT)
For the half-way point of the year can you make a map that somehow signifies the date?
