Long read… If I were still Editor of The Cartographic Journal this would be the next Editorial. If I had some upcoming presentations and a keynote then here’s where my thoughts are at the moment.

I’m going to start this essay by harking back to my student days. It’s 1989. I’m sat in a large room in Oxford with around 30 other students at the beginning of a Bachelors degree in something called cartography. We were sat on stools around large tables that had opaque Perspex tops, and lights that shone from within. Literally light tables (albeit they were heavy beasts! They were to become one of the defining tools of our learning, along with the adjacent darkroom, and the store room stocked with peelcoat, scribecoat, film, snopake, scalpels, and all sorts of other critical tools for making maps (scalpels!!! – yep – couldn’t make maps of the time without them). Upstairs on a mezzanine floor were half a dozen small Macintosh computers. We wouldn’t get to play with these for at least another year. Next door was the photogrammetry lab with all sorts of incredible devices. Pretty much my entire degree programme was analog apart from a few dalliances with computer assisted technologies.

And so, sat around those light tables, my friends and I were set our very first practical assignment. The task – draw a map. I mean, it was a cartography course after all but we thought we might get a little tuition before being asked to create a masterpiece. I guess there were all manner of reasons our professor started us off this way. Having spent many years as an academic I now understand the motive, the pedagocial approach, and the idea of using the exercise as a way of establishing what sort of raw materials he was going to be working with for the next three years. But the exercise itself was intriguing. We were given a passage of text that described a geography. And our task was to interpret the text and draw a map. And that’s what we did. Here’s the passage of text:

Draw, stating your scale, a contoured map to show an island 65 km long from SW to NE which varies in width from 48km in the SW to 16km in the NE. The SW coast is much dissected by long, narrow fjord-like inlets, and is fringed by 5 small rocky islands of varying sizes. From this coast the land rises sharply to a plateau some 600m above sea level and extending through about one-third of the island. The plateau descends to a low undulating plain about 25km long and 20km wide. From the plain a range of hills rises to the NE, flanked by a coastal plain about 8km wide. From these hills. rivers flow to both plains and also from the plateau to the larger plain. The plateau is gritstone. The hills which run down to the coast in the NE, to form cliffs, are chalk. Much of the smaller coastal plain is marsh, but the larger plain, from which two estuaries open, is of well drained alluvial land. ln addition to relief and topography, show drainage, possible sites of settlement and lines of communication. Name your island appropriately!

And here’s my interpretation of the text. I got a B+ which was a decent grade I thought.

I’m going to park this story for a while, but I’ll return to it later.

There’s been a lot of noise about AI over the last year or so, and it’s blown up in geo-circles too. Ian Muehlenhaus and I have been experimenting with AI in an exercise we’ve called #365DaysofMaps as part of our contribution to the research agenda of the International Cartographic Association. You can check out our daily blogs (with occasional longer form write-ups and thoughts) at mapdesign.icaci.org.

In outline, we began to recognise in late 2025 that AI was likely the next big paradigm shift in cartography. We’ve both seen a few of these shifts having trained with pens, French curves, darkroom technology, then desktop publishing, GIS, Illustration software, mainframes, terminals, desktops, the web, mobile, and cloud computing (phew!). We missed out on the shift from clay tablet to papyrus, and copperplate engraving to lithography, but we’ve seen enough to appreciate that every now and again technology rears its head and transforms the way we do our work (the how of what we do). It’s often transformative. Sometimes you don’t need to jump all in. Often, old skills die, but you develop new ways to work too. I once wrote an article that likened editors of journals to Dr Who. The editors regenerate from time to time but the overarching work remains. It’s just the stewardship remains as another person becomes responsible for upkeep for a moment in a bigger life of something far bigger.

I feel like that about cartography too. I am but a blip in the landscape of a domain that has been very good to me and given me a wonderful career. I’ve tried my best to pay it forward in what I do. I don’t always hit the mark but I’d like to think more often than not I’ve made a reasonable contribution during my time. And so the cycle continues. New technologies bring new players, and new approaches. And what I’ve generally noticed is that any time this happens it comes with a tremendous amount of fear, trepidation, sometimes panic, and occasional loathing.

Yes – loathing. I know many who genuinely hate how technology impacts their work. They get all comfy in their slippers. They know what they know. And they like what they know. Along comes change and it’s disruptive. Of course that can be challenging. It’s how you respond to such challenges that’s important.

I suspect all of us in cartography feel this to some extent or another wondering how AI will change our professional lives. If you work in technology that produces software that cartographers use you’ll have seen it already. AI is being imbued into the software we use on a daily basis. It doesn’t matter what your preferred choice of sword is, they’re all being sharpened with AI. Maybe you work in a mapping agency who will likely be looking for ways to further automate and streamline production. Or maybe you’re a freelancer concerned that AI will render your services redundant if people no longer deem your fees are economically prudent when they can have AI make the map they want in a fraction of the time for an even smaller fraction of the money. We go online and see it being used everywhere. And it’s potentially scary what it can do with apparent ease. And inevitably the fear is compounded when we see almost immediate use of the technology for nefarious purposes.

In the week I am writing this, Google released a new tool in Google Earth which allowed users to deploy their Nano Banana Generative AI tool to add imaginary places and spaces directly into the satellite imagery. Now, the intent was to enable people to use it for sensible purposes, perhaps as a planning tool, but within a day people were showing fabricated road traffic accidents in Amsterdam, and imaginary nuclear powerplants in Iran amongst other applications. And Google were forced to remove the tool. Quite a climbdown. But as with any innovation in human history, there’s always bad actors who find a way to utilize it for an unintended purpose.

Our mapping experiment was intended to just play in the sandpit of AI, outside of our daily jobs. We wanted to see what the fuss was, what potential it had for our cartographic workflows, and where the pain points were. In other words, to perhaps try and get a little ahead of the curve to be forewarned and forearmed.

I’m not much of a coder, or even a web mapping guy beyond using a GUI to publish maps to the web. But Ian is far more adept at those parts of cartographic practice. We broadly agreed that he was going to see how AI could help make web maps while I would focus more on Generative AI using image generators to make maps. We’d do 7 days on, 7 days off, and use lists of ‘Days of…’ as a way to generate the themes for the maps we’d make. That way the map topics effectively choose themselves and would, to an extent, limit any bias we might bring to the process by being too invested in any one particular map.

Crucially, this would be completed outside of our day jobs so we really needed the work to be a minimal time sink. We had to take a daily theme, think of a map, design and produce it (using AI), and write a short blog to accompany it every day. So off we went.

As I write we’re about 7 months into the project. We’ve written up various thoughts and experiences along the way via the blog. We’ve learnt a lot, and continue to do so but perhaps the overarching outcome so far is that the pace of change has been so rapid that the sort of maps we made in January and February pale into comparison with what AI is capable of offering now, just a few months later. It’s quite staggering. And along with that experience has been the increasing noise around AI in cartography in general, and of our project as it ended up bubbling to the surface in July.

We’ve had a lot of support from onlookers interested in following progress. But as with many pieces of work that play with new technology the critics soon emerged. I’m not going to get into the details here but being called “professionally negligent”, “unethical”, “slopmerchants”, “lazy”, and a whole host of other epithets is disheartening. We believe it reflects very poorly on the moralistic posturing some have taken which is deeply ironic given some of the same folks have explicitly railed against similar public commentary before (for instance at the onset of web mapping when many, me included, were less than positive about the quality of the maps being published).

Those who have labeled us may call their comments ‘critique’ if they like, but if you begin by saying “I hate maps made using AI”, then use various slurs to talk about colleagues and their  efforts, you’re picking a fight and simply trying to provoke a reaction.

It’s not helpful. It’s unnecessary noise. We’ve tried to ignore it and not respond on social media. I’ve commented on these issues elsewhere so there’s no point giving more air to it, other than to say it’s been an interesting side issue to try and face scrutiny in this fashion from a community we set out to offer assistance to via our experiment.

So, in summarizing my personal thoughts at this point in time, I’ve been looking for a way to better frame my own thinking on the use of AI in cartography. Many comments borne out of my daily experiences are written up in the daily blogs, but what’s the overarching narrative?

I’ve been accused of being an “AI proponent” which is laughable just because I am making maps using AI and the glove seems to fit the negative narrative some wish to portray me in. I also make maps using GIS, as well as pens and pencils. But it’s as if the AI badge is used solely as derogatory framing at the moment. I apparently make untrustworthy slop using AI, but trusted, and beautiful maps using GIS. I simply don’t believe I can be both people at the same time. So I’ve been exploring how I can summarise my thoughts.

A good friend of mine (thanks Steven!) sent me an article by Jeremy Theocharis who has clearly been thinking along similar lines. I don’t know Mr Theocharis at all. He appears to be a developer who specializes in open-source infrastructure in support of manufacturing, but not of anything cartographic. His article, I thought, was a decent stab at framing commonly expressed critique about AI but while also being clear of where he saw its value to his work. I saw many parallels between his experience and mine (and Ian’s). It made me think of how I can adapt the sentiment in much of what he was saying into the cartographic realm.

And before someone claims I’m plagiarizing, I’m not. Any decent student was always taught to do exactly this – use books and other people’s ideas as a framework for expressing their own perspective, and cite your sources. Straight up copying is clearly plagiarism but building on what others have said becomes a useful contribution.

Anyway, here goes…

I’ve read plenty of people’s comments and some more thoughtful critiques of using AI in creative work, and many of them are hard to disagree with from the perspective they take. They point out issues like homogenization of creative output, loss of cartographic craft, overreliance on tools, and the erosion of deep domain understanding. These seem to be well worn criticisms of AI but there’s every chance we heard very much the same criticism when GIS killed traditional design and production methods, or when print cartography was existentially threatened by web mapping. In the context of cartography, these concerns are obviously resonant but are they simply parroting the obvious claims made by anyone who fears major change, especially if it has the real possibility of impacting their livelihood.

In my experience, domain knowledge and understanding simply do not die. New people start making maps, and they are usually pretty sub-par but people learn, and standards normally improve. We’ve seen that in web mapping over the last decade for certain. Are maps any more homogenized than they once were? I would say not. There is a remarkable breadth of design on show right across the board. Maps have never just been made by cartographers either. Some of the best, and possibly most famous, maps in cartographic history both of decades ago, and in the last few years have been made by people with no formal cartographic training. What matters is the map is the output of human input, and people use a wide array of tools to achieve their aims.

Maps are not just technical artifacts; they are interpretive, aesthetic, and communicative acts. They encode decisions about what is important, what is legible, and what story is being told. So when people say that AI will dilute the intent or flatten the uniqueness of cartographic design, there is possibly some truth in that, if we let that happen. And it’s that last part that I feel is crucial.

If we simply ask an AI agent to “make a map showing…”, and we leave the entire process of design to how the LLM interprets the request then guess what, the map is likely going to be utter crap.

Now, let me caveat that. I sometimes make purely illustrative maps such as those you might see by Jo Mora, Heinrich Berann, Anton Thomas and many others. I might not have weeks or months, or even the required skill to make that kind of map. In this particular circumstance I, the human in the loop, am making a conscious decision, to leave the design work to the LLM and its image generating capability. I want to see what it comes up with. If I do not like it, I don’t have to use it, and I can ask again for a different map. Some call this the very worst of the use of Generative AI in Cartography – “low ambition” resulting in “AI slop”. Sure, I am handing over the task of coming up with, and producing a design that is highly illustrative, but the outcome only exists if I let it, and if I subsequently approve it and use it. I see it as one form of using AI for a very specific form of cartographic output. I get that these type of maps are not to everyone’s taste. Just the same as any artwork.

And so, via our experiment, I’m using AI. Not because the (sensible) critiques are wrong, but because they’re incomplete, and because shouting down people for trying things out just backs you into a corner anyway. I am trying to be open-minded even though my inclination is to remain skeptical.

As someone who graduated from a cartography bachelor’s programme that had taught me manual photogrammetry and projection plotting, darkroom print technology, and the skill of a ruling pen and a scribing tool straight into a workplace that demanded we start using GIS, I think I’m reasonably well placed to have some sense of the importance of not getting too wedded to how things have always been done. Life moves pretty fast, as Ferris once said.

Let’s dig deeper and translate some common criticisms of AI into the mapping domain.

“AI erodes originality”

I’m not going to get into the issues of AI being trained on copyrighted materials too deeply. I have no control over it. Is it unethical? I can see that argument but on the other hand we all look at, absorb, and develop ideas from reading and seeing copyrighted materials every day of our lives. Has anyone ever made a map that hasn’t been, to some extent, informed by someone else’s work? Maybe the key aspect here is in being able to cite sources of inspiration. I have tried to ensure the maps using AI that I have made have cited their sources. It’s not always worked but it’s something we can consider as an ethical way of recognizing prior art.

“AI erodes the environment”

Is AI bad for the environment. Again, the answer is almost certainly ‘yes’ but most of all human activity is bad for the environment. We are parasites on this planet. I can take a principled stance on using AI or not, but then I have to consider absolutely everything I ever do in the same light. Products, food, travel, clothing etc. maybe a better approach is to at least be considerate in using AI? And hopefully the environmental cost will reduce over time because it likely isn’t currently sustainable.

“AI erodes quality”

Is AI guilty of producing what is commonly referred to as ‘slop’, or in other words it produces generic output instead of carefully curated quality? In cartography, this means AI-generated maps often default to familiar design patterns—commonly used colour ramps, conventional symbology, and standard framing. They tend to look like something you’ve seen before. But that’s because you probably have seen these maps before. Remember, LLMs have been ‘trained’, and that training has comprised of what’s available to whoever trained the LLM. But is this really any different to our ability to recognize a map made using all types of tools.

For many years the use of an orange gradient fill, and one of a number of north arrows from the Esri style libraries, easily marked a map out as having been produced using Esri technology. Despite the recognizable fingerprints, did we vilify the person who made that map, or decry the technology for having familiar and over-used defaults? Quality comes with thinking. That doesn’t disappear through using AI.

“AI erodes trust”

Perhaps ‘trust’ is a key dimension to this framing. In other presentations I’ve explored the way in which people fall generally into three types when creating maps – honest cartographers, liars, or bullshit artists (after many other authors who have explored creative deception). Honest people want you to be able to trust their map, and see you as a trustworthy source. Liars also want you to trust their map, but they deliberately tell you something that lies about the topic. BS artists don’t care what they map, but they still want you to trust them. So how do you build trust when using LLMs to help make maps, particularly if the default perception is that you are doing no more than offering up ‘slop’?

Well, personally speaking I don’t necessarily trust people. I take a healthy skepticism into my reading of other people’s maps. But I can normally sniff out the human in the loop. I can see where people have likely spent time, and that in turn, is often deserving of my time. But it’s not as easy anymore. And certainly not for the novice map reader.

Maybe the point here is that more senior cartographers will already have amassed a certain well of trust. Why should the assumption be that they will throw that away just because they’re using a different tool? I don’t think that will happen. And senior people have always taught junior people so that idea of developing trust through mapping truthfully should be maintained.

Of course, people will claim that credibility is on the line. I’m aware that is how our mapping experiment has been framed by some. I’ve heard comments that suggest I should hold higher standards, or I am risking the reputation I’ve built over my career. But this is precisely why it’s important to use my relatively prominent position in the cartographic community to experiment and explore from within the discipline. I don’t feel I’m risking my credibility – or maybe I am, and actually that’s perfectly OK, because taking risks is part and parcel of doing good work and moving a discipline forwards. Sure, I could kick back, earn my salary and simply eek things out until I retire. But trying to make a difference is a critical part of helping shape a discipline. I am using LLMs but it’s my thinking, based on my experience, that I am using. The LLM is simply implementing my thoughts in new ways that widen the scope of my cartographic toolkit.

But there’s also a fascinating aspect to this question of trust. I have a sense that some people, at least, hold maps produced by AI to a higher standard than maps produced by more traditional means (is Adobe Illustrator or GIS traditional?). There seems to be a hang-up over so-called ‘image-maps’ because they’re seen as inaccurate and only formed of pixels. Yet even if i use vector data when creating a map using a GIS I will inevitably use heavily processed raster data within it, and may maps all end up as images anyway. The same pixels. So, some are hating on image-generators when we’ve been generating maps using pixels for decades anyway. For example, my map of the Total Eclipse was roundly loved. It won awards. Yet there’s some serious license taken with the way I depicted shade across the map, and frankly it’s utter nonsense. One expert in solar eclipses kindly pointed out that if you are in the track of 100% occlusion you will experience totality, and darkness. But even in an area close by at 98% occlusion you’ll actually not experience 98% darkness. It’ll actually seem more like full daylight because you really don’t need much sunlight for it to appear like daylight. So I used imagery on my map to create an illusion. This is a creative deception that I created to give the map emphasis, and for artistic effect. It’s a lie. Yet the map was made by my hand so it was entirely trusted. The point being, we’ve used images, and raster dats as part-or whole in map-making for decades to create effects that ‘work’ for the map being made. Hillshades? Hachures? Contours? All fabricated. Yet now, some seem to expect absolute perfection from an image generator else it’s ‘slop’. Maybe this is holding AI to different standards. Or maybe the real skill of a cartographer is that we’ve mastered the art of creative deception to the extent people can neither spot it or care enough to hunt it out even when it’s right in front of their noses.

“AI erodes learning”

If an AI generates a choropleth map, chooses a specific projection, or suggests a layout, it can bypass the mental steps that actually teach you cartographic principles. Yes. But many maps are already made in a vacuum of people not knowing thematic mapping techniques (non-normalised data), which projection to use (equal area is essential) and layout (awkward, misaligned elements, missing or poorly sized elements etc). Perhaps a better way of thinking about this is that expertise remains essential specifically to act as a human litmus test for what an AI might produce. Accept what works, technically and aesthetically, and reject, or ask for ineffective or inappropriate work to be redone.

Of course, the novice who isn’t particularly interested in making maps using cartographic wisdom will continue. These people are not interested in learning anyway. This problem plagued early web mapping, but yet is not in evidence to anywhere near the same degree anymore in that realm. I suspect a similar pattern will play out with AI maps. The early adopters and experimenters will make mistakes, and will evidence a poor ability to make sound decisions (I’ve certainly made some pretty poor maps, though my intent was still to be truthful). But this will improve. Learning about cartography is not in and of itself a function of the tool used to make the map anyway.

What is learning based upon anyway? It’s usually based on what has gone before; the structure, frameworks, arguments, agreement, best practices of the discipline. Rules (such as they are) form our understanding but are there to be broken meaningfully, and justifiably. LLMs are amplifying all of this because they can provide the history of cartography rapidly. They are sharp and to the point, fast, and good at looking things up and checking. If you give an LLM very little then don’t expect much in return. Maybe quality is actually something that AI can improve across the board (though maybe that’s still a work in progress).

“AI erodes craftsmanship.”

The careful tuning of label placement, the subtle balancing of visual hierarchy, the selection of typefaces—these are the things that distinguish a thoughtful map from a merely functional one. The argument here is that maps somehow have to attain an almost mythical level of craftsmanship to be considered valuable. Very few made by any means, pass this almost unattainable test. Many more remain useful, and functional, if not necessarily hitting the highs of a masterful map.

But this is like any design. A chair is designed. Every chair is designed. Would I prefer a chair designed by Charles, and Ray Eames or one from IKEA? Well, they likely both perform the same function so what’s the problem? (And actually I have both an Eames chair and chairs from IKEA so both can sit side-by-side anyway.). But despite them being functionally the same one is held up as exhibiting much higher design standards. That in itself is largely subjective even if we agree.

Everything is designed. But not everything is necessarily designed equally. True craftsmanship will persist. It always does. And I suspect it’ll simply become easier to identify the best maps from the rest. And if you stand by your work, whatever tool you used to create it, then you are continuing your craft. You are engaged in it. And, maybe you also help generate a little more credibility in the process.

Craft changes too. Once it was how you engraved a letterform into copperplate. Now it’s developing into how you direct the LLM. And yes, I do think of this as a form of craft. Accepting what an LLM gives you is literally the worst thing you can do in most circumstances. You are absolving yourself of the responsibility of crafting a piece of work. But checking the output and then asking for changes, wholesale, or minor tweaks, is engaging with the work, and crafting the output. To do this successfully you bring to the fore your cartographic expertise – and once again, emphasise its value rather than letting the LLM take over the work in an ad hoc manner. Because the human is capable of discerning something that is ‘good’ from something that is patently ‘bad’, we remain critical in the cartographic process. Of course, sometimes the response ‘hallucinates’ and conjures up something that is patently incorrect. So, correct it. And on the rare occasion the LLM gets stuck in a loop of not being able to deliver what you need, start again, or use a different LLM and AI agent… or train your own agent based on your more specific criteria.

“AI erodes heterogeneity”

When many designers use similar LLMs, we risk a convergence toward a narrow visual language of maps. Yes, quite possibly, but you simply can’t argue that AI will produce generic outputs (bad), erodes craftsmanship (bad), and then claim that homogenization is also bad. If more maps are constructed to better standards, then whether they begin to look similar or not is almost immaterial. There will always be maps that buck the trends, and if the human in the loop specifically directs the AI to make a map that veers from the  common visual languages it is most familiar with, then other map designs can and will be built.

So we’re back to craftsmanship again, and those who develop their craft using AI and are better able to marshal LLMs to produce what they are thinking about, then they will be seen as those who produce the better work. Has it ever been any different in cartography? It’s always those that think the most who produce the best work. The ones who are merely technically competent can produce good work, but it rarely shines quite as brightly or for as long.

These critiques are valid. They describe real risks I’ve observed in early experiments with AI-assisted mapping workflows. Even some experiments I’ve been publishing could be used as examples to support the assertions. (I’m sure some folks will revel in showing me a map I’ve made that contradicts my thinking.)

But there’s levels and nuance to this question of validity. In terms of the experiment Ian Muehlenhaus and I are currently involved in, we committed to make a map a day during 2026 simply to see what AI offered. As noted elsewhere, we’ll report on the detail of our experiment when we’re at the year end, and in appropriate places (i.e. not social media), but it’s been fascinating to see how the technology has developed even over such a short amount of time. AI has certainly proved to be incredibly useful to meet the central aim of making a map a day. How else would that have been possible given this is a side-project and unfunded. By the end of the year, we’ll have made 365 maps. These are typically quick (and often non-QA-ed generative AI maps). Sometimes, given our schedules, we have time to iterate more on a given map but not always. These may be built using different LLM image creators, and designed into full-functioning web maps or even interactive online games (honestly, some of the interactive maps and games Ian has built in a few hours are killer!.)

But we have also become keenly aware of where AI hasn’t necessarily supported the ideas we’ve wanted to execute. Maybe the assumption it is a solution to everything is currently its biggest fallacy. What tool is a true panacea? None! But as part of the cartographer’s wider tool kit, we feel it is going to have a major role to play going forward.

I find AI useful—not as a replacement for cartographic thinking, but as a force multiplier for it. The key is to treat AI not as the map-maker, but as a cartographic assistant that operates within constraints defined by the human in the loop. The human will need cartographic expertise to direct AI to make what they envisage. Along the way the AI will translate the directions into a product that meets the requirements which can then be honed, or even simply rejected. Think of AI like this – it’s a new UI for making maps, but it doesn’t replace you in the making of the map. It’s a new UX for sure too, and you may not prefer it but I’ll bet there’s even now aspects of your current workflow and toolkit you don’t exactly enjoy either.

Following on from my brief assessment of some of the main crtiques as I currently see them, here’s my ten point plan of where I feel AI is currently able to help us make maps:

1. From blank canvas to rapid exploration

In the early stages of map design, I often don’t know what the final product should be. I may have data, a topic, and some vague intent, but not a visual direction.

AI can help me explore:

• Alternative visual encodings
• Different thematic treatments of the same dataset
• Variations in design logic
• Layout sketches

This isn’t about accepting the first suggestion. It’s about quickly generating many starting points so I can react to them critically. The value isn’t in the LLM’s initial output; it’s in the acceleration of iteration.

2. Automating the repetitive, preserving the interpretive

Cartographic production includes a lot of necessary but non-creative work:

• Cleaning attribute tables
• Generating standardized symbology sets
• Drafting metadata
• Creating multiple versions for different scales or audiences

AI can help with these tasks, freeing up time for decisions that actually matter:

• What should this map emphasize?
• How should the spatial relationships be interpreted?
• What is the narrative?

The more I delegate the mechanical parts, the more attention I can give to the interpretive core of cartography. Making a map is essentially an intellectual activity, not just a bunch of skills. You can’t just put skills together and expect a map to emerge. You have to do the thinking first, and work through the intellectual issues first.

3. AI as a critique engine

One of the most interesting uses of AI is not to generate maps, but to evaluate them.

I’ve started using AI to ask questions like:

• “What might confuse a reader about this map?”
• “Is the visual hierarchy aligned with the intended message?”
• “Does the colour scheme support accessibility?”

The responses are not authoritative, but they are surprisingly useful as a second opinion. They often surface obvious issues I’ve possibly become blind to. In this sense, AI functions less like a designer and more like a junior reviewer who never gets tired of looking at drafts. Ian’s Map Doctor provides a really valuable tool to give me a check on my work, and even suggests alternatives which, of course, I can either take on board or reject. In many ways this is akin to having an automated checklist to provide you with an extra set of eyes on your work. And who doesn’t value that feedback? Some of the best feedback I’ve ever received on my maps has been the harshest. Sometimes it hurts but I don’t recall a piece of sound advice ever not making the map better.

4. Expanding cartographic vocabulary

There’s a subtle but important benefit to AI having been exposed to previous work: exposure to unfamiliar approaches. AI models have seen vast quantities of visual material. Despite my years in the industry I can’t possibly have seen everything, but AI has likely seen more than many of us put together. When prompted carefully, they can suggest:

• Less common technical or design choices
• Unusual but valid thematic representations
• Alternative labeling or symbology strategies
• Non-traditional map forms (e.g., isochrone maps, cartograms, hybrid narrative layouts)

Not all of these suggestions are good or will be applicable for a specific mapping task. But they can break me out of habitual design patterns and stretch me. In cartography, habit is both a strength and a limitation. I can recognize the work of many of my peers through subtle design signatures that become their trademark. Great! Until they then suggest they were trying to be different.

5. The risks are real, but worth facing honestly

The criticisms don’t disappear just because the tool is useful in some contexts. If I rely too heavily on AI, I risk weakening my own cartographic judgment, and possibly the trust others may place in my work because of their own views of the value of AI (whether rational or not). My judgment is built through repetition, failure, and refinement. Other’s judgment of me is built through my body of cartographic work and previous writings, and I hope in understanding that we don’t always have to agree on everything. Different views count.

So, as I experiment, I intentionally use AI in a way that tries to keep me in the loop rather than replacing me. Sure, that means I have inevitably faced the ire of peers who are devoutly anti-AI (c’mon, I work for a proprietary GIS company and have faced probably more open source zealots over the years than anything the AI-haters can throw at me) but pushing myself and my craft with new, and emerging technology is a path I’m keen to tread. It won’t all work. But it’ll make my practice all the richer. I can promise you now, I’ve made sloppier maps without AI than I have using AI.

6. Widening technical capabilities and freeing up creative opportunities

As I’ve noted, if many mapmakers use similar prompts and models, we may see a narrowing of visual styles. This is especially problematic in cartography, where diversity of representation can reflect different cultural, analytical, and communicative priorities.

The antidote is not to avoid AI and assume it’s a scourge that cannot be reined in, but to assert stronger design intent than the model defaults to. AI is not a substitute for effort in the cartographic design and production process, but how we expend effort is shifting. We are in a position to effectively outsource the purely mechanical aspects of our jobs, releasing more time to focus on the creative and thoughtful dimensions. That, I find emancipating! Where once people freed themselves from the scribing tool, and scripting languages, we’re potentially free from the shackles of the GUI, and the mouse too.

AI often encodes implicit assumptions about what a “good map” looks like: North-up orientation, certain colour conventions, familiar symbology. If I accept those without critique, I inherit them. So I try to treat AI outputs as proposals, not solutions. My acid test before I release a map remains the same whether it’s built using AI or not – do I like what I see? Am I happy to put my name on it? Does it stack up? Many maps I’ve made have never seen the light of day and live in perpetuity on my hard drive. The same is true of maps I’ve made using AI. Many don’t exist publicly.

7. Being realistic about AI

Rather than being hamstrung by some existential crisis about AI, I’ve found myself taking a far more practical stance. Here’s how I think about AI in cartographic design today, though this may change in the coming months or years of course:

• It is not a cartographer
• It is not a source of truth
• It is not a substitute for spatial reasoning
• It does not necessarily exhibit good cartographic taste

In many respects I find myself at that crossroads again. The same one I experienced when I graduated from a bachelors degree that had taught me what we might now refer to as ‘traditional cartography’, and which had zero GIS in the curriculum. I was becoming aware of GIS but little did I know that as I collected my degree certificate, and my ‘qualification’ in cartography that most of the mechanics of what I had learned was obsolete. But the theories, concepts, and critical thinking was immediately transferable. Having to re-tool or die on graduation taught me that the career I was entering was one heavily impacted by technological change. That was the life I was going to lead. I still rely heavily on much of the bedrock of cartographic thinking I was exposed to at University, and from which I was encouraged to express through forming my own thinking and practice. But the mechanics of making maps in the late 1980s and early 1990s were left behind long ago.

I’m therefore realistic enough to accept the cycle happening again. But not so unrealistic to expect it to reinvent everything. It may, though, be kind of useful for helping do what I do. It’s worth exploring.

8. Own your use of AI

Given I’ve started using AI, it is a tool that I’ve found can:

• speed up exploration
• build and process data (including conversion and formatting)
• handle repetitive production tasks
• suggest alternatives
• act as a reflective mirror
• create maps under close direction
• iterate

But the responsibility for clarity, accuracy, and meaning still sits firmly with the cartographer… with me. I cannot absolve responsibility to AI, and nor should I. Any mistake on the map is ultimately my mistake whether I put it there using a scribing tool, a pen, a poorly executed geoprocessing tool, an errant mouse click, or using an AI agent and an LLM with a cruddy prompt.

I don’t believe in stamping maps ‘made with AI’ because it’s a hostage to fortune. People tune out. Bias gets in the way. The assumption remains, at least at the moment, that AI=slop. It doesn’t, but it’ll take a while to tune that out of the conversation in the same way that it took several years for the proliferation of web maps to yield more consistently solid work.

9. The Core Tradeoff

The tension I am finding, and one which seems to flare up among the cartographic community every now and again, is similar to other creative domains going through the same growing pains with AI:

• If you use AI, you claim to gain speed, breadth, and convenience.
• If you avoid AI, you claim to preserve depth, difficulty, and perhaps originality.

The question is not whether these critiques are correct, they likely are. The question is whether the tradeoffs are worthwhile, and more so, to acknowledge that the two are not mutually exclusive anyway. I can still quite easily avoid AI, or I can go all in. I prefer to simply acknowledge that a new tool has entered my cartographic toolkit, and I can and will use it when I see fit, and where the benefits outweigh the alternatives.

Even when I do make use of AI the goal is not to outsource thinking. It is to think better about maps, with alternative tools that do things I cannot execute using other tools that bring the same benefits.

And framing AI use as unethical is just such a lazy caricature. You cannot be both ethical and unethical by virtue of using one tool over another for any mapping task as long as you remain the human in the loop. Avoiding AI does not make you an ethical mapmaker by default. And now is not the place but I can give you chapter and verse on the lack of ethical behaviour a good number of cartographers exhibit despite their vacuous statements to the contrary. Just do good, open, and transparent work. Ethics is more about your general conduct in life, and in the way you carry yourself professionally, and not a function of one tool or another.

10. Seek joy in your work

I feel so indebted to the domain of cartography in so many ways. It’s brought me a good lifestyle, many good friends, travel, and experiences I would likely never have otherwise had. I have tried to pay it back and pay it forward through my work whether it’s maps, presentations, books, writing, or simply sharing what small piece of wisdom I can with someone. I’ve been fortunate to help many students launch their own careers which is an immense source of pride. And most of all, I’ve had fun. It’s a wonderful career and domain to be involved in.

I know not everyone agrees with my perspective and ideas but it would be terribly boring if that were the case anyway. Healthy debate is critical to the advancement of any scientific field and long may that continue (despite the current trend for codes of conduct that actually stifle debate). But ultimately, use AI or don’t use AI – and continue to seek joy in your mapping pursuits.

Where do we go from here…

Cartography has always evolved alongside new technologies, from hand-drawn plates, to GIS, to automated rendering systems. Each shift has been met with skepticism, and each has ultimately expanded what maps can be. These shifts have always engendered fear by those who fundamentally reject change. Some fall by the wayside. New folks emerge to develop the domain in new, and exciting ways, the cycle continues, and something else will undoubtedly emerge to challenge what we know, and hold dear and true.

The clarion call of ‘use AI on everything’ is to be avoided. You can be an early adopter, but it doesn’t mean you adopt it for anything and everything. Just ask yourself what the value of using it in a particular context is and proceed with caution.

For consumers, the last decade has been characterized by fake news, and fake maps and so on. And now, people are complaining that they can’t tell if something is made using AI. Are they being tricked? Are they watching a deepfake? Is it a deepfake map? Well, perhaps what we’re actually witnessing is that more than ever the needs of the general public to be even more critical of what they are seeing is paramount, and for that we need way more than printing ‘made with AI’ across a map. We need way better education. We need people to better understand geography, and to be able to properly interpret graphical work. Yes – we need graphicacy! We always have. Maybe now we need it more than ever. That said, if cartographers use AI with intent then we won’t ever get to the situation of AI interpreting or imagining the map, and maps will continue to be the paragon of trustworthiness (I can hear someone in the back saying “but…but…all maps lie”, and yes, you’re right but let’s not go into that here).

Maybe what we’re actually seeing is a reimagination of what it means to be a cartographer. For decades people have made maps, and you never had to be a cartographer to do so. Web mapping brought computer scientists and coders to the domain, and the move toward web-based storytelling brought even more. AI is likely to see more try out map-making. That simply means people no longer need to hire an actual cartographer to help them make a map. All those years learning about projections, metadata, points, lines, and polygons…and for what? Is spatial special? Not really. But cartographers (and geospatial folks generally) have for far too long relied on what they believed was a superpower – the ability to drive the machinery to make the tool of the trade – the map. Possibly because no-one else wanted to. Sure, we learnt about the art and science, but it was our technical proficiency that gave us the skills that were in demand – the ability to actually make maps, and often good maps. Maybe AI has finally destroyed that which set us apart. If AI can handle the technical aspects of map-making, and we can fob off the mechanics of map-making then others are bound to want to become involved.

People no longer have to stare at a blank piece of paper (or screen) and a folder of datasets and figure out what ways the data needs manipulating, what analyses need applying, and how to render the results in a meaningful cartographic manner. No. They can simply ask AI a logical question, and tell it to make a map that shows the results. The AI will figure out what the query means, how it needs to parse the data, perform the analytical steps, and show the answer on a map. In simple terms, a machine is now capable of performing all the parts except defining the initial question (plus any further iterations required by the human).

This is not the end of geography, or of cartography. As my good friend Ed Parsons has recently expressed, it actually emphasizes the continued, and extended importance of it. We’re literally in a situation where all of that mechanical friction to making a map (hours, days, weeks) has been taken care of. That means we have way more time to dedicate to the fundamental questions themselves. By allowing us to focus more on these questions, we as the human in the loop, can reapportion our time to the more critical aspects of map-making, and not merely the days of mechanically placing individual labels by hand.

What is expertise anyway? What does it mean to be an expert cartographer? By shunning the potential of AI in our domain, are we really claiming that our ability to drive software is what makes us special? Is pushing buttons what makes us expert cartographers? Or is it actually more about how to understand often complex spatial patterns and develop ways of communicating the same information to different audiences? I know where I stand on this.

AI continues the trend of allowing us to break free from how we make the map. A while ago lithography replaced copperplate engraving and cartographers rejoiced. More recently, the ability to use a drafting pen to draw the map was replaced by the computer which could draw a far more consistent line width than any hand. More rejoicing as automation sped up our processes. And now that change is upon us again as we no longer have to use a mouse to drive a GUI, or write code – we can replace our focus from simply operating software to what is truly important in cartography – making a map that shows what, where, and when, to other humans. We can think more clearly about what AI cannot, namely the questions that need asking. AI isn’t a cartographer. It has no innate interest in being a cartographer. We do.

AI is simply a layer of evolution in cartography, as it is in many aspects of life currently. But we’re moving beyond simply executing the mechanics of making maps to being able to focus more on the critical aspects of map communication. The challenge, I feel, is not to resist it, nor to surrender to it, but to use it in a way that strengthens the cartographic craft instead of diluting it. And that, like good map design itself, is a matter of judgment. Human judgment. Alberto Cairo recently asserted a personal heuristic about using AI in information design/visualization which has relevance to this discussion. He suggested you “might consider using AI

1. If you can clearly envision your desired design and the decisions and process that would lead to it if you were to use non-AI tools instead.
2. Even better: If you – alone or part of a team with sufficient time and resources – could recreate the same design with non-AI tools.” [original emphasis]

I like this sentiment. It backs up the essential idea that AI is merely a tool that can be deployed but not to absolve the designer (cartographer) of the thinking behind the work. I can see exceptions, such as using the artistic capabilities to produce a piece of work that your talent is perhaps not sufficiently capable of achieving (e.g. pictorial or artistic maps) but even then, collaborating with an artist that can bring to bear their abilities is always possible.

And so. Back to the original exercise of creating a map by hand that I began this essay sharing. What my professor had unwittingly given us as our first assignment was a prompt.

It was a written expression that directed us, as students to make a map, using whatever tools we were comfortable with. We all used pens, and coloured pencils because that was the technology at our disposal not having learnt anything about proper cartographic tools. And so I thought, some 37 years later, that I would feed the exact same prompt into today’s AI agents and see what LLM image generators would create.

ChatGPT created ‘Skerrychalk Island’

Google’s Nano Banana 2 created ‘Aethelred’s Isle’

I can attest to the fact that both these maps were made in a fraction of the time it took me to make my original hand-drawn map. Would you grade either as a B+ or better?

But as with the original prompt and 30 novice cartographic students who each created something different to one another, the two maps above are also very different. But they do contain clear similarities that are explicitly interpreted from the detailed direction of the prompt. These similarities are also in evidence on my map. All three came from the same prompt.

The beauty of this exercise is that the imaginary island was in fact a description of the geography of a footprint on a bath mat. My recollection is that no-one figured out they were drawing a footprint. Maybe I’d have got an A+ if that had been my interpretation.

So why share this and bookend this essay? Well, I think my Professor, Roger Anson, was onto something back in 1989 (and no doubt earlier with previous student cohorts). He wrote a prompt. And he asked his tools of choice (us, as his students) to draw a map following that prompt. Over the years I’ve learnt many different ways of drawing that prompt, and in fact have used the same exercise on my own students. Here’s a later version of the result I made using GIS.

Obviously a footprint right? Well yes, if you direct your tool of choice with intent, and given I knew it was supposed to be more foot-shaped when I used GIS to create this version the output was always going to look like a footprint. But it’s only a footprint because we, as humans, recognise it as such.

What I believe this proves is that the cartographer remains in complete control of the output whether they draw with pens, a GIS, and even when using an LLM. In the case of this example maybe my Professor should have slowed down and thought a little more about the prompt. It probably needs a little more work if he truly expected our outputs to look like a foot. Slowing down is probably not a bad approach when using AI and LLMs. Process your thinking. Express your direction with intent. Be detail oriented. I’ve long-held the belief that one of the drawbacks of all sorts of automation in cartographic design and production has been the erosion of the importance of time in the process. This has always been magnified by automation though the provision of an ‘undo’ button is always a help. Probably my over-riding criticism of AI in cartography is that it will erode thinking time further, for those who don’t especially care about thinking anyway. This issue will become magnified with maps being made by AI in minutes.

Slow down. You move too fast. You gotta let the morning last…

Take time to think. The cartographer is still in control. Until the next dissonance in cartographic design and production technology at least, which will undoubtedly bring more fear and loathing.

If you got this far, then thanks for spending the time. Happy mapping (however you make your maps)!