D3.2 Initial directions for innovative use of communications and social media
Executive summary
Background
One of ENGAGE’s objectives is to produce validated actionable knowledge on societal resilience by demonstrating the benefits and impact of the project solutions in different types of disasters. In addition, one of the desired results of the project is to find the best practices for communication and social media (R3).
Goal
Deliverable 3.2 aims at proposing new directions for the innovative use of Artificial Intelligence (AI) technology to improve societal resilience and citizens’ engagement. The characterisation of the solution allows its potential implementation in different contexts and strengthens forms of resilience embedded culturally and socially in local contexts. We developed a design concept of a blueprint for a conversational AI-enabled chatbot to be used by authorities and first responders before and during emergencies and disasters to contribute to building societal resilience. The objective of the deliverable is to suggest directions for innovating solutions that offer: (A) The ability to test or revise the assumption of ENGAGE that AI-enabled technologies can contribute to building societal resilience; (B) an AI-enabled chatbot, allowing emergency authorities to provide a contextual, online and zero-delay response to the mass-public before, during and following emergencies; (C) Innovative solutions for neutralising false messages during disasters, based on the rapid detection and tracking of trending misinformation (e.g., false rumours) on social media. The chatbot will not distribute messages actively but will rather answer the public’s questions and provide unbiased information; (D) Innovative solutions for citizens’ engagement and transfer of knowledge from research to the public, leveraging the project’s suggested directions to relevant population groups, according to their specific needs and expectations.
Process
The process of suggesting to design a concept of the blueprint included several steps. Initially, we reviewed already existing solutions. This was done by reviewing the scientific and grey literature and discussing the options with informants from the partners of ENGAGE, Ki-CoP members, and additional professionals. The collection was systematic but included a snowball component of collecting solutions from one informant to the next, based on the recommendations that were shared. In the second step, we analysed the solution according to criteria of algorithms and datasets in use, interface with external sources and other criteria, as described in the process section. In addition, we collected existing blueprints of AI-enabled chatbots generated by leading technology companies that present state-of-the-art solutions. Finally, we suggested a concept design for the blueprint, a roadmap and a step-by-step implementation plan.
Existing Solutions
The review identified 45 AI-enabled chatbot solutions concerning emergency and disaster management. Most of them are health-related (34), and specifically, many AI-enabled chatbots were developed in the last 18 months to provide information about COVID-19 (31). Five were related to natural disasters (water, weather, food and earthquakes), three for general disasterrelated issues and two focused on women in particular. Although some are still active, several stopped working for many reasons (e.g., they were intended initially for a short-term use only). Regarding the communication platform, 25 of the AI-enabled chatbots were accessible only through the web. The rest of the 20 AI-enabled chatbots were accessible through the mobile phone (e.g., SMS, messaging apps) or mobile apps, and four of them are also available by Facebook Messenger, which is accessible by mobile or web platforms. The review of the solutions highlighted that in most cases, the approach of the AI-enabled chatbots was conservative and cautious, using closed scenarios rather than freestyle texts, highlighting several advantages and disadvantages regarding usage statistics, simplicity versus complexity, bias and more. In addition, the use of ML algorithms and datasets to train the chatbot was relatively small and not sufficient for facilitating the cuttingedge abilities of AI-enabled chatbots.
Existing Blueprints
The blueprints which were collected in this deliverable describe the different components of AI-enabled chatbot services that are provided by Microsoft, Google, IBM, Amazon Web Services (AWS) and Facebook. Components that constitute different services that are based on the same technology were clustered together. Each component is described, and whenever relevant, the possible contribution to emergency and disaster AI-enabled chatbots is suggested. The most popular components of the blueprints focus on connecting the chatbot to the various communication channels of authorities and first responders, the chatbot’s logic (“brain and body”), various machine learning (ML) algorithms, accessibility tools and other technological capabilities, deployment and integration, data management, processing, quality assurance and storing, analysis, monitoring and insights, security and authentication and different types of data and datasets.
The Blueprint
The suggested blueprint employs the most popular cutting-edge, state-of-the-art technologies, as they appear in current blueprints, to the possible use of authorities and first responders, in AI-enabled chatbots for emergencies and disasters, to contribute to building societal resilience. The blueprint is based on two critical working assumptions. First, during the professional meetings conducted while working on this deliverable, many of the Ki-CoP members and ENGAGE’s partners expressed concerns and potential barriers of using AI-enabled chatbots, which are most probably relevant to other authorities and first responders who were not represented in those meetings. These concerns ranged from not trusting the technological capabilities of the AI-enabled chatbot, through possible biases and mistakes, to being afraid that the public will not adopt this solution and will prefer human assistance. The second working assumption is the need to validate that the chatbot is doing what it should do satisfactorily. Therefore, the primary initial recommendation is that the chatbot collaborates with a human call centre that will verify that it is functioning correctly, providing accurate answers and maintaining the treatment of false information. The chatbot will be also monitored in later stages, but in a less intense way, than at the beginning.
Recommendations
In the last part of the blueprint, which is a recommendation for implementation, we suggest additional recommendations to overcome barriers and meet needs expressed by authorities, first responders, partners of ENGAGE and Ki-CoP members. We recommend adopting either the complete blueprint or adoption of several sub-methods that make the adaptation process more stepwise. In addition, we draw a roadmap for the full implementation of AI-enabled chatbots by authorities and first responders in emergencies and disasters to contribute to building societal resilience. The roadmap draws six milestones in five categories to fully execute the blueprint – technological capability, trust, user perspectives, information management, and budget and funds. Last, Appendix D presents ten implementation stages to help authorities and first responders strategise their implementation process of AI-enabled chatbots.
Conclusions and Contributions
Deliverable 3.2 has four significant conclusions and contributions. The first conclusion relates to the objective related to the ability to test or revise the assumption of ENGAGE that AI-enabled technologies can contribute to building societal resilience. Based on the design of the blueprint and the review of solutions and technologies, we concluded that the answer to this assumption should be positive, but with a cautious adoption of the blueprint suggested in this deliverable. The second contribution is that the blueprint refers to an AI-enabled chatbot, allowing emergency authorities to provide a contextual, online, and zero-delay response to the public before, during, and after emergencies. The third conclusion, and contribution, relates to innovative solutions for neutralising false messages during disasters, based on the rapid detection and tracking of trending misinformation on social media. The suggested blueprint highlighted the technologies needed to complete this mission, but with the necessary caution of only tracking and highlighting potential false information, leaving the last decision for human fact-checkers. Last, the fourth contribution is developing innovative solutions and citizens’ engagement and transfer of knowledge from research and industry to the public, leveraging the project’s suggested directions to relevant population groups, according to their specific needs and expectations.