People’s Intelligence Update

People’s Intelligence (PI), one of the winners of the Tech Challenge for Atrocity Prevention, is an expert system that automates the crowdsourcing, evaluation and verification of information in hard-to-access areas, provides actionable feedback to the source of the information and sends early warnings to partner organizations. PI addresses many of the shortcomings of current documentation initiatives using crowdsourcing: lack of relevant and quality information, no or limited assessment of the reliability of the sources and the credibility of the collected information, reliance on the Internet, lack of feedback loops and limited empowerment of those reporting information.

Over the last months, thanks to a USAID grant awarded to PI as one of the winners of the Tech Challenge for Atrocity Prevention co-sponsored by Humanity United we designed a preliminary architecture of the future PI platform and developed a web based demo to demonstrate some of its features. We invite you to visit our Demo page to access the demo and test out some of its features.

Screen Shot 2015-12-02 at 9.47.07 AM

Our starting point was the list of features and user stories which we elicited during a previous phase together with our humanitarian (ICRC, IOM, UNHCR), human rights (Amnesty International), media (Free Press Unlimited) and peacebuilding (Liberia Peacebuilding office) stakeholders and partners thanks to a small grant from the Humanitarian Innovation Fund.

Informed by this work we first designed a preliminary architecture of the PI platform along six main modules: (1) protocols and formats to communicate with the PI platform, (2) the questions and questions flows to structure the collection of relevant information, (3) localization of the questions and the interface in different languages, (4) algorithms to process the collected information, (5) rules that define the question logic, and (6) actions or tasks to be performed following a question and answer session). The six modules are held together by three controllers: (a) a question manager, (b) a dialogue manager and (c) an action manager. To provide for maximum flexibility, the architecture of PI allows for incremental improvements of each module and controller independently from the each other.

After having defined a preliminary architecture we decided to build a demo of the future PI platform to demonstrate the benefits of the collection of structured data as well as the automation of a number of typically resource intensive information management processes, which we believe can lead to substantial efficiency gains. We will also use the demo to collect feedback from our partners and stakeholders and discuss with them the development of additional features of the PI platform including the design of thematic questionnaires and their underlying questions’ logic with PI’s stakeholders.

In line with the agile software development methodology, the demo was built iteratively over four weeks by two experience developers. At this stage the demo is based on a simple incident report scenario and is far from a finished and polished product. Important features such as the provision of actionable feedback to those reporting information still need to be developed, while all of the current features can be greatly enhanced, including with more advanced natural language processing algorithms.

This being said, with the demo as it stands you can put yourself in the shoes of an untrained person and

  • Contact the PI platform and report an incident by answering a number of questions via text messaging.

Or from the perspective of a staff member of one of the organizations running PI:

  • See how the PI Demo facilitates the collection of structured data which are automatically geotagged;
  • Experience how the PI Demo automatically triages reports along pre-established thematic categories;
  • You can adjust the level of confidence that a given report is clustered within a certain category;
  • You can triangulate reports which share similar characteristics including a date, a proximate location and at least one thematic category;
  • You refine the list of keywords defining each thematic category to improve automatic categorization by the PI Demo;
  • And of course, you can view and sort reports by geographic area, categories and time.

We also invite all our demo users to complete a survey where we will gather first impressions as well as ideas how to improve our software

Lessons learned: During the design phase of the PI architecture and the development efforts of the PI Demo, we quickly learned the importance of regular face to face quality conversations with the developers’ team. As soon as a first version of the demo was coded, we tested it and communicated what worked seamlessly, bugs to be corrected and changes to features that worked but that needed to be rendered differently. Developers went back to work, delivered a new version, which we tested again. The process ended when we were happy that the features had been rendered as per the agreed specifications and all remaining bugs had been corrected.

The number of processes implemented as well as the algorithmic complexity of the Demo provide us with an indication of the amount of efforts that will be needed to design the PI platform. It will require sustained development efforts to improve current natural language processing algorithms combined with machine learning capabilities and develop the remaining features of the PI platform. Its development will necessitate close interactions with PI stakeholders whose inputs during the design and testing phases will form the foundation of PI’s development efforts.