Three Wolves, A Monkey, and a Dung Beetle Walk into a Hackathon
Hackathons are an incredible medium for creating new ideas, new friends, and finding out about yourself. As a semi non-technical person (i.e. I don’t code, at least not yet) I’ve always looked at hackathons with a sense of awe, as an unapproachable event that I always wanted to participate in. Well, I finally got my chance and it was an absolute blast!
What was the hackathon all about?
The Hackathon for Good, sponsored by the city of The Hague, the Data Science Initiative, AI Lab, The Hague Tech, and many others. This meant that there was an incredible pool of talent present at the hackathon and the setup was outstanding! The Hackathon focused on peace, justice, and security. There were challenges around identifying damage from aerial imagery, disrupting land grabbing, forecasting humanitarian financing, classifying and identifying propaganda, among others. Most of the challenges were framed to use AI and machine learning to tackle humanitarian issues. Everything built was open sourced and really did work towards a better world. The amount of data floating around us that can be used for good is growing and growing, and hackathons like this give a platform to build usable solutions from that data.
What did we build?
Now, the self important part. What did our team, Oracles, actually work on? About a week before kick off we decided to tackle NCIA’s (NATO Communications and Information Agency) challenge of identifying propaganda imagery in the public space. NATO is working to help fight propaganda not with counter-propaganda but with “the truth, it’s facts and figures and documentation,” as NATO Secretary General Jens Stoltenberg stated in a speech in Tbilisi, Georgia.
Our goal was to build a system, app, what have you, that could analyze insane amounts of images and identify what potentially is propaganda. Luckily, one of our team members, Marki, knew of cortical.io. Cortical lets you create a semantic fingerprint which is a map of semantic values plotted in a “fingerprint” that is way more complicated than I have time to get into just now.
The Hack
Day 0: We decided to meet before the hackathon to get a sense of what each one of us brought to the table. For myself, that was still unclear. But, we had a great time getting to know one another and found out each one of our spirit animals. Frans, a wolf. Rudolf, also a wolf. Yubin, also a wolf. Marki, a monkey (famous for his love of bananas). And myself? A dung beetle. Were our three wolves a wolf pack or a collection of lone wolves?
Day 1: Our first priority was a discussion with NCIA, really understanding their motives in bringing this challenge, and getting a better view of what we would end up doing. After this meeting, we went straight to work. We had a stand-up to decide on our direction and tasks. Frans and I have both been project leaders and I took the initiative to begin the discussion with defining and framing “What is Propaganda?”. After push back to my theoretical inclinations, Frans stepped in and led the team through an overview of the technical pipeline. Given Marki’s work and familiarity with Cortical.io, our solution to the challenge was immediately evident. What was even more evident, was my need to step away and let Frans lead the team. As the lone non-technical, I began on a mission to put my business brain to use and create a plan and framework beyond the technical problem. The dung beetle had become the lone wolf.
We hacked, and hacked, and hacked. I found ways to assist Marki with data collection and to do my best not to bother the others hard at work. In the meantime, I was studying up on NATO’s missions and statements in regards to propaganda. When I brought findings to the team, there was definite reluctance as the problem at hand was a technical solution, not a product roadmap and implementation. I felt pretty useless. As I left the office to get a night’s sleep, the rest of the team was still hard at work building our product and trying to put the pieces together.
Day 2: As we got back to work early the next morning, I attended a pitch workshop for the team members that would be pitching their solutions. The workshop just happened to be run by my business partner Ali. He ran the members through his quick tips on how to deliver a captivating presentation. As I’ve seen his workshops before, and I knew everyone would implement his suggestion of creating a personal story that could draw the audience into the issue, I had to be different. A quick shameless plug for Ali’s public speaking workshop happening at The Hague Tech, here.
I went back from the workshop which focused on stripping a lot of tech speak out of the presentations and told my team members what we could expect from the pitch. It was a continual back and forth on how much of the product should be shown and its technical capabilities described. In the end, it took a visit from a mentor, Timo, working with the program that emphasized our need to wow the audience, rather than teach them. It was also Timo who helped me work through the most important part of the pitch but I’ll get to that soon.
I began to outline the pitch, already with an introduction in mind. I would begin with a gameshow “What Is Propaganda!” where I had the audience raise their hands if they thought an image was propaganda. I spent the next few hours while my team was working hard to complete as much of our prototype as possible trying to find an analogy that could explain in layman’s terms exactly what the team had been busting their butts to build. Our product took an image, ran it through Google’s Vision API to extract relevant information and keywords. The relevant information included URLs of where the image could be found online. Then, we took those URLs and scraped them for further context and keywords. We then created a semantic fingerprint which we could run against semantic fingerprints we’d previously made from propaganda and non-propaganda images provided by NCIA. From this, we could conclude if the image was most likely propaganda or not by its correlation to known semantic fingerprints. This is a very top level, simplified view of what we did, but I needed it even less technical as I only had two minutes to present.
As the pitch grew closer, the tension grew amongst our team as there was widespread skepticism about how I would be presenting our solution. After spending the last 36 hours working on the product, I understood the reluctance to allow me to possibly mangle the technically brilliant solution with my presentation. The presentation was only an hour away and our code had just just been uploaded to the judges.
Day 2 — The Presentation: We shuffled into the auditorium with the other teams and took a seat beside each other. I was armed with my newly formulated analogy for how our product worked and was eager to get on stage and show everyone what we’d accomplished.
I finally got my turn and ran up onto the stage with my game show emcee hat on, shouting into the mic like the beginning of Wheel of Fortune, “WHAT IS, PROPAGANDA!”. As I showed the audience individual images, it was easy for them to make a snap judgement on the proposed propaganda. But the next slide, what I called Round 2, was a collage of hundreds of images. I then asked, “Now, quickly tell me every image that is propaganda in 1 second,” an impossible task. But an exercise that exemplified what NATO workers are presented with everyday and why our solution was so valuable. Next, it was time to unveil my analogy I’d been giddy about for an hour. Our solution as a police precinct. We bring in our propaganda suspects, interrogate the hell out of them (Google Vision API) to get all the information we can. Next, our forensics team checks their known associates, alibis, and raids their apartment to find all connected info (URL scraping). Then, we take our suspects fingerprint (Cortical.io) and run it against our analytics engine to see if we have a match for propaganda. The depiction had gone off without a hitch, and I stared to my left at the timer on stage to see I had 20 seconds left. I ended with a call to action, that we’d built something usable right now that could help in an effort against propaganda and that we needed NCIA’s and NATO’s help to finish what we’d started.
After the presentations, our team met and congratulated each other on a job well done. I won’t bore you with the build up to the awards, as we didn’t end up winning the hackathon. But from the feedback we received, we were as close as can be to earning a spot on the podium and our team’s inexperience with hackathons had us overlook crucial elements that we will be keenly aware of next time. And as a silver lining, NCIA was impressed with the work we’d done and we’ll be in touch with them in the future.
What A Weekend.
Getting to work with a group of great guys and being able to contribute as a non-technical team member was an amazing experience that I won’t soon forget. After it was all done and dusted, Marki wrote in our group chat with pride about the work that we’d done. After everyone had chimed in, I simply wrote “I enjoyed it so much that let’s do it again!”.
Thanks to Frans, Marki, Rudolf, and Yubin for being great team members and hard workers. Thanks to The Hague Tech, The City of the Hague, AI Lab, and everyone else who made this experience possible. To everyone involved, I enjoyed it so much that let’s do it again!
#Hackathonforgood
