The Process

Around 4 months ago I started a project to create a geopolitical risk report for every country in the world. The concept is something I’d considered before—about a year ago I thought about using my newfound free time following graduating university to create a history and foreign policy-centered database (think a more focused wikipedia); but, each time the reality of the hours of research and writing it would take to complete the project set in, my inspiration would wither. For someone plagued by a preinclination towards laziness, even the thought of it was too much to bear.

Fast forward to a few months into my new job, something had changed to make the whole task seem possible. To put it short: I experienced first hand the miraculous potential and efficacy of AI. Though I had barely used—and almost shunned—the tool during college, my AI know-how rapidly developed under the tutelage of my intelligent and wholesomely work-averse colleague. In those few months, I went from barely understanding ChatGPT to knowing which platform was best for performing searches, which was best at synthesizing provided information, and, most importantly, how to successfully craft prompts. Armed with this revelation, I thought once again of my previous dream and this time deemed it attainable.

Starting around December, I began experimenting with different platforms, processes, and prompts to discover the best way to create the most in-depth and structurally sound reports. More than that, I reimagined the concept to fit my AI aspirations. The new plan was to manufacture a system that would constantly scour the internet for news and reports on a country and automatically update a report with that information—creating a real-time wikipedia. Similarly, driven by a perhaps too-strong fascination with Ian Bremmer and a desire to escape from the trappings of my job to a life of political commentary, I shifted the goal from creating an all-you-need-to-know political database to corporate-tailored, on-the-demand geopolitical risk reports. Yet, about a month in, the two changes soon reached the extent of my ability and interest, respectively. On top of not being inspired by corporate interests, I worried about reinforcing the post-colonial trap that has kept so many developing economies poor by focusing on the economic woes of a country and dissuading the investment that is so badly needed. Technologically frustrated and topically bored, I resolved to drop the “risk” and create geopolitical reports for every country using AI as far as I could take it.

From the start, I knew my two principal tools would be Perplexity and Google AI Studio. In my experience, especially at a time when ChatGPT only used information from the internet up to a certain year, Perplexity was the best AI application at gathering information from searches and listing its sources used. This would prove vital in finding the various reports and articles from what I deemed the “most reputable sources” (i.e. multilateral institutions like the UN, prominent think tanks like Brookings, and publishers like Foreign Policy). As for Google AI Studio, I knew it would be the best at analyzing the information in the documents I drew from Perplexity and synthesizing it all into one cohesive report. The reason for this: not only does the application have a much higher token count (i.e. how much information it can remember and keep in mind at one time) than others, but it also only uses information from the documents provided. Curating the reports by hand and using a closed-system application, I hoped to avoid the “garbage in-garbage out” as well as hallucination issues prevalent in AI.

My last challenge, then, was to decide how I would structure each geopolitical report. Initially, I dictated categories to the AI that—as my partner thankfully pointed out to me, although at the time I was exasperated—were too rigid and seemingly arbitrary. I was far too concerned with what I imagined would trigger business-dooming unrest (such as a full section focused on the condition of ethnic minoritized groups) rather than the most pertinent political considerations to each country. I feared, too, that by standardizing a set of categories I may miss out on country-unique information not captured under the existing structure. I found giving the AI free reign, however, produced irregular and similarly limited reports. In the end, as a compromise, I used ChatGPT to help write the prompt used to create the report, keeping some of the most pertinent categories like foreign policy while leaving space for “Unique to Country” topics.

With a system set up at last—though it was far from finalized, I began working on what I called Project ATLAS. The goal: to better understand the political and foreign policy considerations of every country in the world during a time of increasing geopolitical competition and misunderstanding. I thought what better place to start than Africa—home to the greatest number of countries and perhaps America’s most inadvertent and, at times, conscious “grey area” in political awareness. What could I learn about the continent with six out of ten of the fastest growing economies and only increasing geopolitical importance in a time when key heads of America’s Africa policy remain vacant or have frequent turnover?