I’m opposed to generative AI and all its works. According to some leading names in the cartographic community, that makes me a gatekeeper, a Luddite, and an irrational AI hater.
So, generative AI has come to cartography. The ICA’s Commission on Map Design has been posting examples of maps produced using generative AI tools, and there’s a website highlighting the ways you can use ChatGPT to generate maps using natural language prompts. The proponents of this stuff have been making some fairly inflammatory statements in its defence, such that I can only imagine they’ve been getting some grief about using it and are pushing back against it. Hence the stuff about gatekeepers, Luddites and haters. These are epithets I’ve seen thrown around before by pro-AI advocates in other fields where the use of generative AI has proven even more controversial. They’re not new. They’re also not worth taking seriously.
Let’s start with “gatekeeper”: the argument is that generative AI tools open the field to non-professionals who might not otherwise be able to make maps. The gatekeeper argument has often had merit: there have been many instances where tools that made things easier for mere mortals (iMovie for video editing, for example, or affordable digital SLRs) were dismissed or frowned upon by professionals whose livelihoods might be impacted by people empowered to DIY it instead. But the people touting AI-generated maps appear to be professionals, not people embracing a new tool to make maps they’d otherwise lack the skills to produce.
People like me, in other words. In the 23 years I’ve been blogging about maps, I’ve made no bones about the fact that I’m not a mapmaker. I have no cartographic skills whatsoever. I’d dearly love to be able to make maps of my own, and I’m deeply envious of people who have that skill. You’d think that an AI mapmaking tool would be a dream come true for me. But you’d be wrong: I wouldn’t touch it with a ten-foot pole.
Partly that’s because I wear more than one hat: I’m also a science fiction/fantasy writer and critic, and in that field generative AI, which was trained on the work of so many of my colleagues without their consent, and which is has been used to inundate fiction markets with auto-generated slop, is professional poison. But mostly it’s because I wouldn’t be the one making the map: the AI is doing the work. I couldn’t claim it as mine, nor would I be able to enjoy the satisfaction—the joy—of having created something.
At first blush, using generative AI to make a map seems to fall somewhere between hiring a contractor to make a map for me and a rather too upscaled version of Google MyMaps or MapChart, and I’m not sure what the end result would be good for beyond personal, internal or limited-audience professional use (e.g. illustrating a slide deck). Maybe posting as a meme to Reddit, although subreddits are increasingly banning AI-generated content for some inexplicable reason.
That the people promoting AI-generated maps appear to be professionals is confusing. It’s one thing to experiment with an algorithmic tool that generates maps from natural language prompts, quite another to propose using the output of those prompts in actual work. One key problem is that AI-generated work is not copyrightable. Legally speaking, you don’t own the work. So why should anyone pay you money for an AI-generated map based on your prompts, or employ you to make one? Anyone can use the same prompts and get the same results. And then what? Would an AI-generated map be accepted for publication in the NACIS Atlas of Design or the GeoHipster calendar? (This question will come up. Have your popcorn ready.) Would anyone buy, on purpose, an entire atlas of AI-generated maps? (Amazon, Apple et al. are already flooded with AI-generated fake books designed to fool buyers.)
And in the event that AI-generated maps were widely adopted by the industry, it would not go well for professional mapmakers. Crow all you like about the time saved by using AI tools: it shows you haven’t been paying attention. AI adoption in other fields has not meant more leisure time for professionals. It’s led to mass layoffs, and precarious employment for the survivors whose work has shifted to checking and correcting AI outputs, which sounds very fulfilling. The savings, in other words, are not passed on to you. If anyone should be Luddites, it’s professionals—and please read up on the actual Luddite movement (preferably without asking ChatGPT) before deploying “Luddite” as an epithet. Luddites were opposed to low pay, poor working conditions and bad quality of work—and you should be too.
And we do need to talk about the bad quality of work. Large language models like ChatGPT are stochastic parrots: what they produce is not necessarily true. That is to say, they make stuff up. AI hallucinations in the form of fictitious citations have proliferated in academic publications and legal filings (for which lawyers have been sanctioned). The whole point of using the tool to save time and effort is undermined if you have to double-check its work.
If you thought, as I initially did, that this would be less of a problem in AI-generated cartography for some reason, sadly, no.
In a recent blog post, Daniel Huffman noted that some of the AI-generated maps shared by ICA’s Commission on Map Design contain “serious inaccuracies” (as in, a U.S. map with two Ohios). He does not mince words.
The people behind these maps have acknowledged that the maps contain errors, but have continued to post them on social media without including any sort of caution. While geography enthusiasts such as you and I will spot some of the errors, I do not doubt that some lay readers are going to take the maps at face value and be misled.
Personally, I think it’s unethical to widely share maps that have known mistakes, without alerting your audience to them. Sure, errors creep into everyone’s work, and we do our best to avoid and correct them. But it’s one thing to let a mistake slip past, and quite another to knowingly share false information. I do not think that the folks behind the Commission on Map Design’s efforts have an intent to mislead, and part of their brief is to explore and demonstrate the errors that AI makes. But, it’s baffling that they’re sharing those maps with the wider public, without first warning the audience about the quality of the information that they’re consuming. To my mind, they are acting with professional negligence.
Maps, Daniel argues, should not misrepresent their subject, nor should they leave their readers with “a significantly inaccurate understanding” of their subject. Dylan Moriarty thinks AI-generated maps should come with a disclaimer—all the more because an inaccurate map shared outside of its context could be taken for real.
(Not that maps generated by ChatGPT are hard to spot: they have the same graphic design language shared by a lot of AI-generated artwork, in terms of text and colour choices and general over-sharpening. People are pushing back against flyers that use these design defaults, so don’t be surprised if they have the same reaction to AI maps.)
In other words, take responsibility for the work, because a computer can’t. And blaming the user for the hallucinations of the tool—the tool isn’t at fault, the user should have given a better prompt—goes against the notion that this is a tool that anyone can use. If doing the prompt the right way is a special skill, and debugging and error-checking the outputs still requires specialist knowledge, then the gatekeeper argument is disingenuous. Because if even cartographers can publish a map with two Ohios, the average Joe on whose behalf this tool is ostensibly made is going to have a rougher time of it. To say nothing of professionals in other fields who resort to AI-generated maps rather than hiring a cartographer (again, why are cartographers promoting this?) who might not appreciate the trap they’re falling into.
One last Luddite argument about generative AI is to compare it to the typewriter—an analogy a lot of AI apologists make, such that I have to wonder if ChatGPT is giving out the same talking points to all its users. Generative AI, the argument goes, is a tool like a typewriter: you wouldn’t oblige someone to declare that they wrote something with a typewriter instead of a pen, nor would you discriminate against work just because it was written with a typewriter.
This analogy is specious horseshit on so many levels. For one thing, publishers actually insisted on typewritten manuscripts on legibility grounds, to the point of specifying one-inch margins, double spacing and Pica pitch: they would discriminate against you for not using a typewriter, in other words. (Such gatekeeping!) Also, writers as a group will not shut up about their tools: they swear by Scrivener and swear at Word in public; George R. R. Martin’s preference for the vintage word processing program WordStar is well-known; Larry McMurtry thanked his typewriter at the Golden Globes. The typewriter preferences of so many writers are public knowledge. In general, whether it’s software or camera lenses or brushes or paints, no artist has ever felt the need to hide what they were using.
Are there differences between writing with a pen, writing with a typewriter and writing with a word processor? Absolutely. But in each case you are in complete control of what appears on the page: no one is going to question your authorship if you use an IBM Wheelwriter instead of an Olympia SM9. And while I myself enjoy messing around with typewriters, I also know why they went obsolete: for actual work, computers are objectively better. No one needed to strong-arm writers into making the switch, nor was it necessary to get defensive or insulting toward the stubborn few who stuck to their ribbons and paper. People by and large adopt new tech when the benefits are obvious and they are able (can afford) to do so. (It’s not like professional cartographers are pining for the days of scribing and waxed type.)
The fact that people are actively, volubly resisting generative AI should tell you something. So should the fact that AI boosters feel the need to attack the people resisting AI.
The angry defensiveness at play makes me think that the writing tool that best resembles generative AI isn’t the typewriter. It’s cocaine.
Look, I’m serious. Cocaine is totally a writer’s tool: Stephen King wouldn’t have written The Tommyknockers without a small mountain of it. Writers integrating liberal amounts of drugs and alcohol into their writing process is well documented: they fall into the catch-22 of needing it to create, but having it progressively impair their ability to do so. Writers’ substance abuse is a great metaphor for using generative AI: use it too much and it strangles your ability to think for yourself. There are plenty of stories of people who can’t stop using generative AI for everything—every creative work, every executive decision, they ask ChatGPT, even when told not to, even when they’re facing professional sanctions for the last time they used it. It’s fucking addictive.
Now, not a lot of what I’ve said so far has been specifically about maps, which is a problem for a post on a map blog, but that’s because the problem of generative AI can’t be seen in isolation. Those who spend most of their time in the mapmaking community might not be aware that the arguments the proponents use are not new, and not limited to cartography. It’s in the proponents’ interest to be laser-focused on what it can do for their specific field, what it can do and how easy it is, because the problems we’ve heard about haven’t really struck home yet. But as I said, I wear a lot of hats, and I’ve seen what it’s done in other fields, and what it’s doing at a macro level.
I can’t ignore the spike in RAM and storage prices that have sent hard drive, SSD and computer prices skyrocketing; data centres spiking electricity usage and consuming scarce water supplies and being generally bad for climate change and the environment; large language models being basically built on plagiarizing their training materials, such that class-action lawsuits are yielding payouts in the billions; AI crawlers hammering web sites and increasing hosting costs; the entire industry showing every appearance of a bubble, with no clear way of recouping the hundreds of billions in investment unless the AI industry essentially swallows every last dollar on the planet, making AI a paperclip maximizer that converts every computer chip, watt, dollar and neuron into bad medical advice, made-up legal citations and maps with two Ohios; the artificially low prices relative to AI company spending that make it clear that we’re in stage one of enshittification, and once everyone’s job has been replaced by chatbots and agents the prices will be jacked up in a hopeless attempt to pay for all this; the increasing public fury at having to deal with AI chatbots instead of live human beings whenever they need customer support.
They’d rather talk about gatekeepers and haters and Luddites and typewriters because they don’t want you to notice what this is going to cost.
They want to tell you that AI is inevitable because they know you won’t adopt it willingly. They know it’s widely hated. They can’t allow you to think that you have a choice.
But we do.
They want to tell you that a mapmaker’s work has no value. That understanding geography, learning skills, developing an eye and discernment and the all-important attention to detail—these are no longer necessary. That you can just learn to write the prompts to make the machine give you the map you want.
Apparently, those who spend years making beautiful and informative maps—the ones that over the past 23 years, have sometimes literally taken my breath away—are just a bunch of suckers.
The hell with that.
I love the work mapmakers do. And I know they do too.
Generative AI completely misses that point.
I don’t hate AI-generated maps because I’m a gatekeeper or a Luddite. I hate AI-generated maps because they’re lazy.