The AI-Assisted Geographer: From Deep Learning Research to Everyday GIS Automation
A Brief Introduction to “AI”
Prompt: “Generate a blog post about AI/ML, include my research experience and how I integrate AI into my workflow, make no mistakes.”
When we look back at prompts like the sentence above in a few years, we may reminisce. The idea of AI as a search bar-esqe tool or a way to generate text is rapidly fading, and the horizon of AI generation is looking very different. When we say “AI” we are talking about large language models (LLMs).
LLMs are part of a family called “neural networks” which have been around for a while. Only recently have we obtained computers which can process the data they require. When OpenAI released GPT-3 in 2020, capability and hype would only grow over the next few years. I was an early adopter, initially having it write sonnets about my dog (and watching it struggle with multiplication).
Fast forward six years and you’ll find me using it to make myself a better geographer. Through this blog post, I hope to share how AI has enhanced my research, before exploring the ways that I have utilized AI (in my research and beyond).
Presenting my work at AAG!
AI in my Research
My first experience with AI and geography was in my sophomore year at Penn State, where I used an online deep learning website to analyze imagery for a research project. I spent weeks annotating lakes by hand, only to feed it all into the website as “true” data—data it used to train itself to annotate more lakes for me.
After my time as an annotator, I was taken on to work with a lab doing deep learning research on lakes located on Antarctic ice shelves, as an evaluator of how the methods worked. I was able to see behind the scenes on the latest advances in data processing and how machine learning/deep learning (the same family as LLMs) are revolutionizing geography.
With the speed of modern hardware, it is reasonable to process the entirety of Antarctic lakes within a few weeks once the neural network is trained, something that would have been unthinkable before. My current research focuses on comparatively analyzing Antarctic lake detection algorithms, meaning I get to see how traditional detection and annotation methods you might see in ArcGIS (i.e. random forest, K-means) stack up with newer, deep learning based approaches. I’m still finishing up my work, but the results are looking promising.
Maps involved in my work, the colors represent lake detections by each method. DeepLakes is the deep learning one.
Recently, I had the opportunity to present my work at the American Association of Geographers (AAG) Annual Meeting in San Francisco, California. AAG is a once a year, geography centric meeting where a diverse group of people from all over the world come to share their work.
Presenting my work was very rewarding; as an undergraduate, it was a great opportunity to develop an elevator pitch on the fly (who am I, what am I doing, why does it matter?) and practice selling my research. Additionally, there are a litany of talks (from info sessions on the YPN to discipline specific business meetings) and posters open all day long you can attend.
I learned so much about other applications of GeoAI; one poster used a social media website’s in-built AI to create a database of user’s posts with specific location data and image properties to track flooding. Keeping in the know about how others are using GeoAI is as important as using it yourself.
AI in ArcGIS
What I’ve been able to accomplish in my research would not have been possible without AI. My experience in Python was moderate, but I had never handled something as complicated as requesting data from (sometimes outdated) government websites, processing it into a usable format and doing the intense data analysis a comparative study needs. I would take what I knew about the data I needed, the code that I could find on stack overflow and some ideas I had, coalesce them into a large text prompt (simplifying it for the AI) and send it over to one service, get a chunk of code, and send that to a different service to see if I could get a better response.
Companies design their models to be responsive to competitors, so if you tell one AI “your competitor made this code” it will be sure to give you a better response. I like to think of the services as “Undergraduate-horsepower,” as in there will be bugs, but it is offloading work that you don’t want to do. In the end, this makes you more efficient at accomplishing a task. The frontiers of what is achievable with AI generated code and advice is growing in recent years (and months).
Credit Esri; how the AI workflow looks!
At the most basic level, AI is now woven into the fabric of our digital environment—from Google Workspace to our daily calendars. However, the most transformative shift for GIS professionals is the integration of AI directly into the technical workspace.
Modern AI models can now interface with your computer's terminal, automating tasks like downloading datasets, organizing file structures, and even launching ArcGIS through simple natural language commands. Within the ArcGIS environment specifically, AI Assistants are beginning to bridge the gap between complex geoprocessing and intuitive interaction. Instead of navigating through nested toolboxes, a user can simply prompt the software to, say, explore the pickleball market (Use an AI assistant to explore the pickleball market | Documentation).
The AI then interprets the spatial logic, selects the appropriate analysis tools, and configures the workflow automatically. This intelligence extends into field operations through applications like ArcGIS Survey123, where the assistant can help you to create surveys that integrate into your workflow with a couple of clicks.
Final Thoughts
AI is not at the stage where it can replace the decisions a human can make, but it can help you accomplish things you never could have imagined. For anyone trying to use it in a geographic context, try replicating papers or studies from major open source publications, tackle a project you would otherwise ascribe to the coding department, or try to automate a task you do everyday.
Understanding that AI is not able to replace most jobs, you should realize that it can improve your ability to do yours; better to try it out than be left behind.
I hope you enjoyed reading this blog! If you want to connect with me, feel free to reach out to jrowley429@gmail.com or connect with me James Rowley | LinkedIn.