In 2018, in the build-up of the 2019 elections, we began thinking of the best ways citizens can get easier access to key information and data. This is because we know that during every electoral cycle, citizens are very particular about accessing information related to campaign finance and spending, agendas and manifestos, voter registration, voting procedures, Permanent Voter’s Card (PVC) distribution, and more. They usually have concerns about electoral violence, ballot box snatching, and where their polling stations would be. Beyond this, election time is a time where mis- and disinformation is know to spread in abnormal amounts. To addres these issues, we decided on the Voteron project.
Our strategy and process
First we considered the already exisiting data in the public domain, both those open and not, and how we could simplify access to them. We collected data relevant to our goal (data from previous elections especially), and began re-structuring them in a way the technology we want to build would deliver them in simplified, and storified narratives, making cumbersome data easier to read and interpret by non-technical citizenry. The tool we decided to build was called 'Vote:ron', which had robotic assitance and automation abilities. Using Natural Language Processing, a technique in Artificial Intelligence, Voteron was trained to undersatand human conversational flow, and how to play around data. As a result, it had the ability to simplify huge datasets into simplified narratives or snippets which can be delivered to any user of the app as push notifications or upon request by the user by asking it simple questions in a chat-like nature. Users could also get simplifed versions of other relevant information to keep them informed before, during and after the elections.
Outcomes and learnings
We built a tool capable of delivering simplified data and key information in an almost human conversational pattern. We gave citizens access to information in a whole new way. Citizens had a tool that they could easily consult to dispell suspected mis- or disinformation. Being that this was the first time we were building something around artificial intelligence, we struggled, especially trying to surmount limitations already exisitng in the technology libraries we employed. We built this tool twice using two different frameworks due to the limitations we had in the first one. The second time we involved a technology partner which gave us a smooth sailing.