No, an AI map builder won’t steal your job

In the last couple of years, AI has taken over every aspect of our lives. I use it to build shopping lists, routines to ignore for my annual visits to the gym and - more than anything else - I use it in my work in GIS. 

The GIS world is ripe for an AI revolution, with a lot of companies taking the “map builder” approach to this, i.e. creating and using AI tools to automate the production of maps. I’d say my top skillset in geo is cartography and data visualization: I make (hopefully) cool maps. So - these developments should have me quaking in my boots, right? If companies can just use AI to build maps, what on earth do they need me for? 

But… I’m not quaking. Here’s why.

Map building is about 10% of the work

If your work touches data, you’ll probably hear the often-cited statistic: 60-80% of every data professional’s time is spent sourcing, cleaning, formatting and importing data. Give that statistic to a GIS professional, and they’ll respond: “80%? That sounds low!”

If you’re automating map building, you’re likely only automating about 10% of a GIS analyst’s work (source: my life).

If I think back to every GIS project I’ve ever worked on, the other 90% of that time is spent on…

Data management

Location data is incredibly rich in context, making it an absolute goldmine for training AI models. It’s also messy, chaotic and full of inconsistencies which can quickly turn that goldmine into… I don’t know, what’s the opposite of a goldmine? A rubbish dump?

I worked on a project once where I had to take development data from around 15 different public sector bodies and make a simple map showing projected housing and employment numbers. So easy, right? Definitely something an AI could do.

So, after weeks of back-and-forthing with data custodians at these organisations about licenses and access, the data started to trickle in with the velocity of golden syrup. I think I received the data in the requested format from maybe two of the organisations. The rest sent it through via a host of different formats: geojsons, CSVs, Microsoft Access tables, single-feature shapefiles, detailed CAD designs and individual PDF site plans. My personal favourites were forwarded email chains with textual descriptions of sites, with highlights including “land near station” and “somewhere west of *insert town here.” Meanwhile, housing and employment projections also landed in various formats: 529, 600-800, 3 acres, 2 floors, large (?), yes and - my personal favourite - “probably.”

So while the map took me all of 30 minutes to make, the time spent wrangling data - and data custodians - took days… perhaps weeks, not to mention the time I've lost in therapy.

The thing with GIS is most people only see the output - a map. They don’t see the hours, weeks or - in this case - LIFETIMES of work that went into getting the data map-ready. That includes managing data sourcing, prep, metadata, lineage, versioning… Your AI map builder isn’t doing that.

Business processes & procedures

If it’s difficult to train an AI to manage data, it’s even harder to teach it to work like an employee. I don’t mean making sarcastic comments during town halls or drinking your annual coffee budget by February (but I’ll stop just repeating my CV here). I mean teaching it to understand business and project structures, processes and expectations.

Sure, you can give your AI a stack of PDFs on business procedures to read, but that doesn’t mean it can understand how things actually work. For GIS work - like all work, really - the processes don’t live entirely in a handbook. They live in your team’s brains, collated through years of experiences, successes and failures. You can’t prompt a model to “remember that last time we did this, we had a big issue with stakeholder sign-off” - or “double-check if this map is going to the external version of the board or the internal one, because we may need to anonymise the data.”

We talk about AI as if it’s a replacement for human action, but often, what you need is a decision. Someone who can weigh up the formal process against the actual situation. Someone who remembers that last time we tried to merge two datasets like this, it broke everything - so maybe we should loop in the data architect before pushing ahead.

AI can follow rules, sure. But they can’t interpret or bend them - or write new ones. 

Stakeholder management

I think in my whole life as a map-maker, I have made one single map for myself. Every other map has been for someone else - typically a colleague or a client. Making that map means understanding their context, requirements and tastes, and aligning that with my expertise and skills. These are things an AI will struggle to understand, address or compromise on.

Sure you might be able to prompt “I need a map showing population density in London - and make it blue (please).” But what is the map going to be used to decide.. and what do they want the outcome to be? This would impact everything from design choices to the data used to convey population density (Time period? Granularity? Source?). Any GIS analyst would know the right questions to ask to make the best possible map, an AI would guess, and the results... differ 👇

AI is only good if you know what good looks like

I say this so often, I may as well get it tattooed. AI is only good if you know what good looks like. I can’t use chatGPT to make me a brilliant fantasy football team, because I don’t even know what a bog standard one looks like - let alone a brilliant one. I can use it to write a pretty decent blog post - because that’s much more in my wheelhouse, but crucially I never treat it like the end product, more like something written by an assistant or intern, needing my skills and expertise to turn it into something that hits.

The same thing applies to maps - but it’s not just about making a “good” map.  It applies to the underlying data & analytics - as well as to whether it actually does that job the map needs to do. It’s not just aesthetics. A map can look great and be terrible… Does AI know the difference?

The price of accessibility?

One of my favourite stories working in GIS was when a manager called me up angry at 4pm on a Friday. My map was wrong. The data on it looked totally different to the data they were looking at in QGIS. We did some investigating and… it turned out they had downloaded the shapefile from our centralized database to their local machine and made a huge amount of edits from there. Seriously... hundreds.

This is an extreme example - and could obviously happen with any sort of data or file. But GIS has an awful lot of people who aren’t GIS experts working in it. I want to be super clear - I am not gatekeeping. I want GIS to be open and accessible for anyone who wants to use it, and I am 100% up for AI enabling that - but this should be for the parts of GIS where that makes sense.

In my opinion, every GIS project still needs an expert in the backseat, setting up processes, guardrails and best practices. Otherwise you’re letting AI loose on your data, processes, solutions and platforms - and you’d better be damn sure they’re water-tight.*

*They aren’t. No - not even yours.

The best maps are by humans, for humans

Have a think about what the most common map you’ve made is.

For me, it’s 100% a red line boundary map. This is a really simple map featuring a basemap, a boundary showing a site or area of interest and your standard marginalia. Easy peasy, I could do it in my sleep (once I’d got past all the data and stakeholder management). Could an AI do that? Yes - and please, be my guest. Let them be the one addressing a project manager’s urgent request to make the red 10% darker (no, not that dark!) because they “just like it that way.”

LLMs, don't give up your day job

What’s not easy? Creating a map where the stakeholder is adamant that it must have 48 layers on it, a map that needs to tell a really tight and sensitive story, or one that just needs to be so stunning that it makes a potential customer sign off a project there and then. Those are the types of project that need really careful consideration, iteration and creative flair - with a focus on outcomes, not just outputs. 

AI in geo - what is it good for?

So so much! Here are some of my favourite examples of how AI is changing geo: 

• Guardrails-up Agents: I love the concept of AI agents in GIS - almost like smart assistants that can take a goal and figure out how to get there. The thing is, this only works when they’re designed with guardrails: clear boundaries, defined data sources, and expert oversight. That’s why - and this isn’t an ad, I just work for them - I love the approach CARTO is taking to AI Agents; created by experts, designed for decision-makers. 

• Supporting code creation: I swear I was going to get good at Python one day. Guess I don’t have to now!

• Data creation: particularly dummy data! I work in product marketing and I have a huge need for “looks real, is fake” spatial data - and AI just makes this so easy to create. 

• Feature detection on imagery: deep learning for tasks like building footprint extraction, change detection over time and land cover classification.

• Foundation models: training machine learning models on massive, diverse datasets is generating incredible results at a huge scale. Shout out to my amazing CARTO colleagues who have been doing some great work with Google’s PDFM, which you can read about here. 

  
  

So no, I don’t see the future of AI in GIS as just “automated map making.” I see it as a powerful tool that amplifies the expertise of GIS analysts - helping us work faster, smarter, and at a whole new scale.

Closing thoughts

A friend said to me recently that they think it will be AI that breaks the internet, not Kim Kardashian like we all thought. They theorised that we’ll get to a point where it’ll feel like everything is created and consumed by AI - not humans. It’s happening already - 70% of AI-generated music is just streamed back by AI. 

And we’ll get sick of it. I love AI, but even I’m tired of scrolling through LinkedIn knowing I’m reading stuff created by LLMs, not humans. We’ll crave small-scale interactions with actual humans, driven by empathy, connections and storytelling.

And what are maps, if not just a way of storytelling with data? The best stories are the ones told by people, and the best maps are too. 

So - sure - buy that fancy-looking tool that has a built-in AI map builder. Let it take some of the simple map-building tasks away from your non-experts. Let it build you hundreds of red line boundary maps… But if you think that an AI map builder can replace even 10% of what your GIS analyst is doing for you, you couldn’t be more wrong.

(And - sidebar - if all you’re using your GIS analyst for is making boundary maps… please ask them what they COULD be doing for you! It could literally change your organisation)

The job of the GIS analyst is safe - even if it is changing. 

…All this being said, I’d look like such a loser if I lost my job to AI right now.